Merge remote-tracking branch 'remote/develop' into server

35e3be9a · WilliamZhang06 · 0d1f90ad · 5c4df656 · 35e3be9a · 35e3be9a
411 changed file
--- a/.mergify.yml
+++ b/.mergify.yml
@@ -80,6 +80,12 @@ pull_request_rules:
    actions:
      label:
        add: ["CLI"]
+  - name: "auto add label=Server"
+    conditions:
+      - files~=^paddlespeech/server
+    actions:
+      label:
+        add: ["Server"]
  - name: "auto add label=Demo"
    conditions:
      - files~=^demos/
@@ -130,7 +136,7 @@ pull_request_rules:
        add: ["Docker"]
  - name: "auto add label=Deployment"
    conditions:
-      - files~=^speechnn/
+      - files~=^speechx/
    actions:
      label:
        add: ["Deployment"]
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
+repos:
 -   repo: https://github.com/pre-commit/mirrors-yapf.git
-    sha: v0.16.0
+    rev: v0.16.0
    hooks:
    -   id: yapf
        files: \.py$
        exclude: (?=third_party).*(\.py)$
 -   repo: https://github.com/pre-commit/pre-commit-hooks
-    sha: a11d9314b22d8f8c7556443875b731ef05965464
+    rev: a11d9314b22d8f8c7556443875b731ef05965464
    hooks:
    -   id: check-merge-conflict
    -   id: check-symlinks
@@ -31,7 +32,7 @@
        -  --jobs=1
        exclude: (?=third_party).*(\.py)$
 -   repo : https://github.com/Lucas-C/pre-commit-hooks
-    sha: v1.0.1
+    rev: v1.0.1
    hooks:
    -   id: forbid-crlf
        files: \.md$

--- a/CHANGELOG.md
+++ b/CHANGELOG.md
 # Changelog
+Date: 2022-1-29, Author: yt605155624.
+Add features to: T2S:
+  - Update aishell3 vc0 with new Tacotron2.
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1419
+Date: 2022-1-29, Author: yt605155624.
+Add features to: T2S:
+  - Add ljspeech Tacotron2.
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1416
+Date: 2022-1-24, Author: yt605155624.
+Add features to: T2S:
+  - Add csmsc WaveRNN.
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1379
+Date: 2022-1-19, Author: yt605155624.
+Add features to: T2S:
+  - Add csmsc Tacotron2.
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1314
 Date: 2022-1-10, Author: Jackwaterveg.  
-Add features to: CLI:  
+Add features to: CLI:
-  - Support English (librispeech/asr1/transformer).  
+  - Support English (librispeech/asr1/transformer).
  - Support choosing `decode_method` for conformer and transformer models.  
  - Refactor the config, using the unified config.  
  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1297
 ***
+Date: 2022-1-17, Author: Jackwaterveg.  
+Add features to: CLI:
+  - Support deepspeech2 online/offline model(aishell).
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/1356
+***
+Date: 2022-1-24, Author: Jackwaterveg.  
+Add features to: ctc_decoders:  
+  - Support online ctc prefix-beam search decoder. 
+  - Unified ctc online decoder and ctc offline decoder.  
+  - PRLink: https://github.com/PaddlePaddle/PaddleSpeech/pull/821
+***
--- a/README.md
+++ b/README.md
@@ -16,12 +16,15 @@
 <p align="center">
    <a href="./LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-red.svg"></a>
-    <a href="support os"><img src="https://img.shields.io/badge/os-linux-yellow.svg"></a>
+    <a href="https://github.com/PaddlePaddle/PaddleSpeech/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleSpeech?color=ffa"></a>
+    <a href="support os"><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
    <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
    <a href="https://github.com/PaddlePaddle/PaddleSpeech/graphs/contributors"><img src="https://img.shields.io/github/contributors/PaddlePaddle/PaddleSpeech?color=9ea"></a>
    <a href="https://github.com/PaddlePaddle/PaddleSpeech/commits"><img src="https://img.shields.io/github/commit-activity/m/PaddlePaddle/PaddleSpeech?color=3af"></a>
    <a href="https://github.com/PaddlePaddle/PaddleSpeech/issues"><img src="https://img.shields.io/github/issues/PaddlePaddle/PaddleSpeech?color=9cc"></a>
    <a href="https://github.com/PaddlePaddle/PaddleSpeech/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleSpeech?color=ccf"></a>
+    <a href="=https://pypi.org/project/paddlespeech/"><img src="https://img.shields.io/pypi/dm/PaddleSpeech"></a>
+    <a href="=https://pypi.org/project/paddlespeech/"><img src="https://static.pepy.tech/badge/paddlespeech"></a>
    <a href="https://huggingface.co/spaces"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue"></a>
 </p>
@@ -143,6 +146,8 @@ For more synthesized audios, please refer to [PaddleSpeech Text-to-Speech sample
 <div align="center"><a href="https://www.bilibili.com/video/BV1cL411V71o?share_source=copy_web"><img src="https://ai-studio-static-online.cdn.bcebos.com/06fd746ab32042f398fb6f33f873e6869e846fe63c214596ae37860fe8103720" / width="500px"></a></div>
+- [PaddleSpeech Demo Video](https://paddlespeech.readthedocs.io/en/latest/demo_video.html)
 ### 🔥 Hot Activities
 - 2021.12.21~12.24
@@ -236,7 +241,7 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
  </thead>
  <tbody>
    <tr>
-      <td rowspan="3">Speech Recogination</td>
+      <td rowspan="4">Speech Recogination</td>
      <td rowspan="2" >Aishell</td>
      <td >DeepSpeech2 RNN + Conv based Models</td>
      <td>
@@ -249,7 +254,7 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
      <a href = "./examples/aishell/asr1">u2.transformer.conformer-aishell</a>
      </td>
    </tr>
-      <tr>
+    <tr>
      <td> Librispeech</td>
      <td>Transformer based Attention Models </td>
      <td>
@@ -257,6 +262,13 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
      </td>
      </td>
    </tr>
+  <tr>
+      <td>TIMIT</td>
+      <td>Unified Streaming & Non-streaming Two-pass</td>
+      <td>
+    <a href = "./examples/timit/asr1"> u2-timit</a>
+      </td>
+  </tr>
  <tr>
  <td>Alignment</td>
  <td>THCHS30</td>
@@ -266,20 +278,13 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
  </td>
  </tr>
   <tr>
-      <td rowspan="2">Language Model</td>
+      <td rowspan="1">Language Model</td>
      <td colspan = "2">Ngram Language Model</td>
      <td>
      <a href = "./examples/other/ngram_lm">kenlm</a>
      </td>
    </tr>
-    <tr>
+  <tr>
-      <td>TIMIT</td>
-      <td>Unified Streaming & Non-streaming Two-pass</td>
-      <td>
-    <a href = "./examples/timit/asr1"> u2-timit</a>
-      </td>
-    </tr>
-    <tr>
      <td rowspan="2">Speech Translation (English to Chinese)</td> 
      <td rowspan="2">TED En-Zh</td>
      <td>Transformer + ASR MTL</td>
@@ -317,14 +322,15 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
    </tr>
    <tr>
      <td rowspan="4">Acoustic Model</td>
-      <td >Tacotron2</td>
+      <td>Tacotron2</td>
-      <td rowspan="2" >LJSpeech</td>
+      <td>LJSpeech / CSMSC</td>
      <td>
-      <a href = "./examples/ljspeech/tts0">tacotron2-ljspeech</a>
+      <a href = "./examples/ljspeech/tts0">tacotron2-ljspeech</a> / <a href = "./examples/csmsc/tts0">tacotron2-csmsc</a>
      </td>
    </tr>
    <tr>
      <td>Transformer TTS</td>
+      <td>LJSpeech</td>
      <td>
      <a href = "./examples/ljspeech/tts1">transformer-ljspeech</a>
      </td>
@@ -344,7 +350,7 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
      </td>
    </tr>
   <tr>
-      <td rowspan="5">Vocoder</td>
+      <td rowspan="6">Vocoder</td>
      <td >WaveFlow</td>
      <td >LJSpeech</td>
      <td>
@@ -378,7 +384,14 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
      <td>
      <a href = "./examples/csmsc/voc5">HiFiGAN-csmsc</a> 
      </td>
-    <tr>                                                                                                                                       
+    </tr>
+    <tr>
+      <td >WaveRNN</td>
+      <td >CSMSC</td>
+      <td>
+      <a href = "./examples/csmsc/voc6">WaveRNN-csmsc</a>
+      </td>
+    </tr>
    <tr>
      <td rowspan="3">Voice Cloning</td>
      <td>GE2E</td>
@@ -416,7 +429,6 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
    </tr>
  </thead>
  <tbody>
  <tr>
      <td>Audio Classification</td>
      <td>ESC-50</td>
@@ -440,7 +452,6 @@ PaddleSpeech supports a series of most popular models. They are summarized in [r
    </tr>
  </thead>
  <tbody>
  <tr>
      <td>Punctuation Restoration</td>
      <td>IWLST2012_zh</td>
@@ -463,7 +474,6 @@ Normally, [Speech SoTA](https://paperswithcode.com/area/speech), [Audio SoTA](ht
  - [Automatic Speech Recognition](./docs/source/asr/quick_start.md)
    - [Introduction](./docs/source/asr/models_introduction.md)
    - [Data Preparation](./docs/source/asr/data_preparation.md)
-    - [Data Augmentation](./docs/source/asr/augmentation.md)
    - [Ngram LM](./docs/source/asr/ngram_lm.md)
  - [Text-to-Speech](./docs/source/tts/quick_start.md)
    - [Introduction](./docs/source/tts/models_introduction.md)
@@ -489,7 +499,17 @@ author={PaddlePaddle Authors},
 howpublished = {\url{https://github.com/PaddlePaddle/PaddleSpeech}},
 year={2021}
 }
+@inproceedings{zheng2021fused,
+  title={Fused acoustic and text encoding for multimodal bilingual pretraining and speech translation},
+  author={Zheng, Renjie and Chen, Junkun and Ma, Mingbo and Huang, Liang},
+  booktitle={International Conference on Machine Learning},
+  pages={12736--12746},
+  year={2021},
+  organization={PMLR}
+}
 ```
 <a name="contribution"></a>
 ## Contribute to PaddleSpeech
@@ -540,6 +560,7 @@ You are warmly welcome to submit questions in [discussions](https://github.com/P
 - Many thanks to [mymagicpower](https://github.com/mymagicpower) for the Java implementation of ASR upon [short](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_sdk) and [long](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_long_audio_sdk) audio files.
 - Many thanks to [JiehangXie](https://github.com/JiehangXie)/[PaddleBoBo](https://github.com/JiehangXie/PaddleBoBo) for developing Virtual Uploader(VUP)/Virtual YouTuber(VTuber) with PaddleSpeech TTS function.
 - Many thanks to [745165806](https://github.com/745165806)/[PaddleSpeechTask](https://github.com/745165806/PaddleSpeechTask) for contributing Punctuation Restoration model.
+- Many thanks to [kslz](https://github.com/745165806) for supplementary Chinese documents.
 Besides, PaddleSpeech depends on a lot of open source repositories. See [references](./docs/source/reference.md) for more information.

--- a/README_cn.md
+++ b/README_cn.md
@@ -147,6 +147,8 @@ from https://github.com/18F/open-source-guide/blob/18f-pages/pages/making-readme
 <div align="center"><a href="https://www.bilibili.com/video/BV1cL411V71o?share_source=copy_web"><img src="https://ai-studio-static-online.cdn.bcebos.com/06fd746ab32042f398fb6f33f873e6869e846fe63c214596ae37860fe8103720" / width="500px"></a></div>
+- [PaddleSpeech 示例视频](https://paddlespeech.readthedocs.io/en/latest/demo_video.html)
 ### 🔥 热门活动
@@ -233,7 +235,7 @@ PaddleSpeech 的 **语音转文本** 包含语音识别声学模型、语音识
  </thead>
  <tbody>
    <tr>
-      <td rowspan="3">语音识别</td>
+      <td rowspan="4">语音识别</td>
      <td rowspan="2" >Aishell</td>
      <td >DeepSpeech2 RNN + Conv based Models</td>
      <td>
@@ -254,6 +256,13 @@ PaddleSpeech 的 **语音转文本** 包含语音识别声学模型、语音识
      </td>
      </td>
    </tr>
+    <tr>
+      <td>TIMIT</td>
+      <td>Unified Streaming & Non-streaming Two-pass</td>
+      <td>
+    <a href = "./examples/timit/asr1"> u2-timit</a>
+      </td>
+    </tr>
  <tr>
  <td>对齐</td>
  <td>THCHS30</td>
@@ -263,19 +272,12 @@ PaddleSpeech 的 **语音转文本** 包含语音识别声学模型、语音识
  </td>
  </tr>
   <tr>
-      <td rowspan="2">语言模型</td>
+      <td rowspan="1">语言模型</td>
      <td colspan = "2">Ngram 语言模型</td>
      <td>
      <a href = "./examples/other/ngram_lm">kenlm</a>
      </td>
    </tr>
-    <tr>
-      <td>TIMIT</td>
-      <td>Unified Streaming & Non-streaming Two-pass</td>
-      <td>
-    <a href = "./examples/timit/asr1"> u2-timit</a>
-      </td>
-    </tr>
    <tr>
      <td rowspan="2">语音翻译（英译中）</td> 
      <td rowspan="2">TED En-Zh</td>
@@ -315,14 +317,15 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
    </tr>
    <tr>
      <td rowspan="4">声学模型</td>
-      <td >Tacotron2</td>
+      <td>Tacotron2</td>
-      <td rowspan="2" >LJSpeech</td>
+      <td>LJSpeech / CSMSC</td>
      <td>
-      <a href = "./examples/ljspeech/tts0">tacotron2-ljspeech</a>
+      <a href = "./examples/ljspeech/tts0">tacotron2-ljspeech</a> / <a href = "./examples/csmsc/tts0">tacotron2-csmsc</a>
      </td>
    </tr>
    <tr>
      <td>Transformer TTS</td>
+      <td>LJSpeech</td>
      <td>
      <a href = "./examples/ljspeech/tts1">transformer-ljspeech</a>
      </td>
@@ -342,7 +345,7 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
      </td>
    </tr>
   <tr>
-      <td rowspan="5">声码器</td>
+      <td rowspan="6">声码器</td>
      <td >WaveFlow</td>
      <td >LJSpeech</td>
      <td>
@@ -376,7 +379,14 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
      <td>
      <a href = "./examples/csmsc/voc5">HiFiGAN-csmsc</a> 
      </td>
-    <tr>                                                                                                                                       
+    </tr>
+    <tr>
+      <td >WaveRNN</td>
+      <td >CSMSC</td>
+      <td>
+      <a href = "./examples/csmsc/voc6">WaveRNN-csmsc</a>
+      </td>
+    </tr>
    <tr>
      <td rowspan="3">声音克隆</td>
      <td>GE2E</td>
@@ -415,8 +425,6 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
    </tr>
  </thead>
  <tbody>
  <tr>
      <td>声音分类</td>
      <td>ESC-50</td>
@@ -440,7 +448,6 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
    </tr>
  </thead>
  <tbody>
  <tr>
      <td>标点恢复</td>
      <td>IWLST2012_zh</td>
@@ -468,7 +475,6 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声
  - [语音识别自定义训练](./docs/source/asr/quick_start.md)
    - [简介](./docs/source/asr/models_introduction.md)
    - [数据准备](./docs/source/asr/data_preparation.md)
-    - [数据增强](./docs/source/asr/augmentation.md)
    - [Ngram 语言模型](./docs/source/asr/ngram_lm.md)
  - [语音合成自定义训练](./docs/source/tts/quick_start.md)
    - [简介](./docs/source/tts/models_introduction.md)
@@ -549,6 +555,7 @@ year={2021}
 - 非常感谢 [mymagicpower](https://github.com/mymagicpower) 采用PaddleSpeech 对 ASR 的[短语音](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_sdk)及[长语音](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_long_audio_sdk)进行 Java 实现。
 - 非常感谢 [JiehangXie](https://github.com/JiehangXie)/[PaddleBoBo](https://github.com/JiehangXie/PaddleBoBo) 采用 PaddleSpeech 语音合成功能实现 Virtual Uploader(VUP)/Virtual YouTuber(VTuber) 虚拟主播。
 - 非常感谢 [745165806](https://github.com/745165806)/[PaddleSpeechTask](https://github.com/745165806/PaddleSpeechTask) 贡献标点重建相关模型。
+- 非常感谢 [kslz](https://github.com/kslz) 补充中文文档。
 此外，PaddleSpeech 依赖于许多开源存储库。有关更多信息，请参阅 [references](./docs/source/reference.md)。

--- a/dataset/voxceleb/README.md
+++ b/dataset/voxceleb/README.md
+# [VoxCeleb](http://www.robots.ox.ac.uk/~vgg/data/voxceleb/)
+VoxCeleb is an audio-visual dataset consisting of short clips of human speech, extracted from interview videos uploaded to YouTube。
+VoxCeleb contains speech from speakers spanning a wide range of different ethnicities, accents, professions and ages.
+All speaking face-tracks are captured "in the wild", with background chatter, laughter, overlapping speech, pose variation and different lighting conditions.
+VoxCeleb consists of both audio and video. Each segment is at least 3 seconds long.
+The dataset consists of two versions, VoxCeleb1 and VoxCeleb2. Each version has it's own train/test split. For each we provide YouTube URLs, face detections and tracks, audio files, cropped face videos and speaker meta-data. There is no overlap between the two versions.
+more info in details refers to http://www.robots.ox.ac.uk/~vgg/data/voxceleb/
--- a/dataset/voxceleb/voxceleb1.py
+++ b/dataset/voxceleb/voxceleb1.py
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Prepare VoxCeleb1 dataset
+create manifest files.
+Manifest file is a json-format file with each line containing the
+meta data (i.e. audio filepath, transcript and audio duration)
+of each audio file in the data set.
+researchers should download the voxceleb1 dataset yourselves
+through google form to get the username & password and unpack the data
+"""
+import argparse
+import codecs
+import glob
+import json
+import os
+import subprocess
+from pathlib import Path
+import soundfile
+from utils.utility import check_md5sum
+from utils.utility import download
+from utils.utility import unzip
+# all the data will be download in the current data/voxceleb directory default
+DATA_HOME = os.path.expanduser('.')
+# if you use the http://www.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/ as the download base url
+# you need to get the username & password via the google form
+# if you use the https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a as the download base url,
+# you need use --no-check-certificate to connect the target download url 
+BASE_URL = "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a"
+# dev data
+DEV_LIST = {
+    "vox1_dev_wav_partaa": "e395d020928bc15670b570a21695ed96",
+    "vox1_dev_wav_partab": "bbfaaccefab65d82b21903e81a8a8020",
+    "vox1_dev_wav_partac": "017d579a2a96a077f40042ec33e51512",
+    "vox1_dev_wav_partad": "7bb1e9f70fddc7a678fa998ea8b3ba19",
+}
+DEV_TARGET_DATA = "vox1_dev_wav_parta* vox1_dev_wav.zip ae63e55b951748cc486645f532ba230b"
+# test data
+TEST_LIST = {"vox1_test_wav.zip": "185fdc63c3c739954633d50379a3d102"}
+TEST_TARGET_DATA = "vox1_test_wav.zip vox1_test_wav.zip 185fdc63c3c739954633d50379a3d102"
+# kaldi trial
+# this trial file is organized by kaldi according the official file,
+# which is a little different with the official trial veri_test2.txt
+KALDI_BASE_URL = "http://www.openslr.org/resources/49/"
+TRIAL_LIST = {"voxceleb1_test_v2.txt": "29fc7cc1c5d59f0816dc15d6e8be60f7"}
+TRIAL_TARGET_DATA = "voxceleb1_test_v2.txt voxceleb1_test_v2.txt 29fc7cc1c5d59f0816dc15d6e8be60f7"
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument(
+    "--target_dir",
+    default=DATA_HOME + "/voxceleb1/",
+    type=str,
+    help="Directory to save the voxceleb1 dataset. (default: %(default)s)")
+parser.add_argument(
+    "--manifest_prefix",
+    default="manifest",
+    type=str,
+    help="Filepath prefix for output manifests. (default: %(default)s)")
+args = parser.parse_args()
+def create_manifest(data_dir, manifest_path_prefix):
+    print("Creating manifest %s ..." % manifest_path_prefix)
+    json_lines = []
+    data_path = os.path.join(data_dir, "wav", "**", "*.wav")
+    total_sec = 0.0
+    total_text = 0.0
+    total_num = 0
+    speakers = set()
+    for audio_path in glob.glob(data_path, recursive=True):
+        audio_id = "-".join(audio_path.split("/")[-3:])
+        utt2spk = audio_path.split("/")[-3]
+        duration = soundfile.info(audio_path).duration
+        text = ""
+        json_lines.append(
+            json.dumps(
+                {
+                    "utt": audio_id,
+                    "utt2spk": str(utt2spk),
+                    "feat": audio_path,
+                    "feat_shape": (duration, ),
+                    "text": text  # compatible with asr data format
+                },
+                ensure_ascii=False))
+        total_sec += duration
+        total_text += len(text)
+        total_num += 1
+        speakers.add(utt2spk)
+    # data_dir_name refer to dev or test
+    # voxceleb1 is given explicit in the path
+    data_dir_name = Path(data_dir).name
+    manifest_path_prefix = manifest_path_prefix + "." + data_dir_name
+    with codecs.open(manifest_path_prefix, 'w', encoding='utf-8') as f:
+        for line in json_lines:
+            f.write(line + "\n")
+    manifest_dir = os.path.dirname(manifest_path_prefix)
+    meta_path = os.path.join(manifest_dir, "voxceleb1." +
+                             data_dir_name) + ".meta"
+    with codecs.open(meta_path, 'w', encoding='utf-8') as f:
+        print(f"{total_num} utts", file=f)
+        print(f"{len(speakers)} speakers", file=f)
+        print(f"{total_sec / (60 * 60)} h", file=f)
+        print(f"{total_text} text", file=f)
+        print(f"{total_text / total_sec} text/sec", file=f)
+        print(f"{total_sec / total_num} sec/utt", file=f)
+def prepare_dataset(base_url, data_list, target_dir, manifest_path,
+                    target_data):
+    if not os.path.exists(target_dir):
+        os.mkdir(target_dir)
+    # wav directory already exists, it need do nothing
+    if not os.path.exists(os.path.join(target_dir, "wav")):
+        # download all dataset part
+        for zip_part in data_list.keys():
+            download_url = " --no-check-certificate " + base_url + "/" + zip_part
+            download(
+                url=download_url,
+                md5sum=data_list[zip_part],
+                target_dir=target_dir)
+        # pack the all part to target zip file
+        all_target_part, target_name, target_md5sum = target_data.split()
+        target_name = os.path.join(target_dir, target_name)
+        if not os.path.exists(target_name):
+            pack_part_cmd = "cat {}/{} > {}".format(target_dir, all_target_part,
+                                                    target_name)
+            subprocess.call(pack_part_cmd, shell=True)
+        # check the target zip file md5sum
+        if not check_md5sum(target_name, target_md5sum):
+            raise RuntimeError("{} MD5 checkssum failed".format(target_name))
+        else:
+            print("Check {} md5sum successfully".format(target_name))
+        # unzip the all zip file
+        if target_name.endswith(".zip"):
+            unzip(target_name, target_dir)
+    # create the manifest file
+    create_manifest(data_dir=target_dir, manifest_path_prefix=manifest_path)
+def main():
+    if args.target_dir.startswith('~'):
+        args.target_dir = os.path.expanduser(args.target_dir)
+    prepare_dataset(
+        base_url=BASE_URL,
+        data_list=DEV_LIST,
+        target_dir=os.path.join(args.target_dir, "dev"),
+        manifest_path=args.manifest_prefix,
+        target_data=DEV_TARGET_DATA)
+    prepare_dataset(
+        base_url=BASE_URL,
+        data_list=TEST_LIST,
+        target_dir=os.path.join(args.target_dir, "test"),
+        manifest_path=args.manifest_prefix,
+        target_data=TEST_TARGET_DATA)
+    print("Manifest prepare done!")
+if __name__ == '__main__':
+    main()
--- a/docs/images/arch/PaddleSpeech_Server_architecture_diagram.png
+++ b/docs/images/arch/PaddleSpeech_Server_architecture_diagram.png
--- a/docs/images/arch/PaddleSpeech_Server_class_diagram.png
+++ b/docs/images/arch/PaddleSpeech_Server_class_diagram.png
--- a/docs/images/arch/paddlespeech_high_layout.jpg
+++ b/docs/images/arch/paddlespeech_high_layout.jpg
--- a/docs/source/asr/augmentation.md
+++ b/docs/source/asr/augmentation.md
-# Data Augmentation Pipeline
-Data augmentation has often been a highly effective technique to boost deep learning performance. We augment our speech data by synthesizing new audios with small random perturbation (label-invariant transformation) added upon raw audios. You don't have to do the syntheses on your own, as it is already embedded into the data provider and is done on the fly, randomly for each epoch during training.
-Six optional augmentation components are provided to be selected, configured, and inserted into the processing pipeline.
-* Audio
-  - Volume Perturbation
-  - Speed Perturbation
-  - Shifting Perturbation
-  - Online Bayesian normalization
-  - Noise Perturbation (need background noise audio files)
-  - Impulse Response (need impulse audio files)
-* Feature
-  - SpecAugment
-  - Adaptive SpecAugment
-To inform the trainer of what augmentation components are needed and what their processing orders are, it is required to prepare in advance an *augmentation configuration file* in [JSON](http://www.json.org/) format. For example:
-```
-[{
-    "type": "speed",
-    "params": {"min_speed_rate": 0.95,
-               "max_speed_rate": 1.05},
-    "prob": 0.6
-},
-{
-    "type": "shift",
-    "params": {"min_shift_ms": -5,
-               "max_shift_ms": 5},
-    "prob": 0.8
-}]
-```
-When the `augment_conf_file` argument is set to the path of the above example configuration file, every audio clip in every epoch will be processed: with 60% of chance, it will first be speed perturbed with a uniformly random sampled speed-rate between 0.95 and 1.05, and then with 80% of chance it will be shifted in time with a randomly sampled offset between -5 ms and 5 ms. Finally, this newly synthesized audio clip will be fed into the feature extractor for further training.
-For other configuration examples, please refer to `examples/conf/augmentation.example.json`.
-Be careful when utilizing the data augmentation technique, as improper augmentation will harm the training, due to the enlarged train-test gap.
--- a/docs/source/asr/models_introduction.md
+++ b/docs/source/asr/models_introduction.md
@@ -38,7 +38,7 @@ vi examples/librispeech/s0/data/vocab.txt
 ```
 #### CMVN
-For CMVN, a subset of the full of the training set is selected and be used to compute the feature mean and std.
+For CMVN, a subset of or full of the training set is selected and be used to compute the feature mean and std.
 ```
 # The code to compute the feature mean and std
 cd examples/aishell/s0

--- a/docs/source/demo_video.rst
+++ b/docs/source/demo_video.rst
+Demo Video 
+==================
+.. raw:: html
+    <video controls width="1024">
+    <source src="https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/PaddleSpeech_Demo.mp4"
+            type="video/mp4">
+    Sorry, your browser doesn't support embedded videos.
+    </video>
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -27,7 +27,6 @@ Contents
   asr/models_introduction
   asr/data_preparation
-   asr/augmentation
   asr/feature_list
   asr/ngram_lm
@@ -42,6 +41,7 @@ Contents
   tts/gan_vocoder
   tts/demo
   tts/demo_2
 .. toctree::
   :maxdepth: 1
@@ -51,12 +51,14 @@ Contents
 .. toctree::
   :maxdepth: 1
-   :caption: Acknowledgement
+   :caption: Demos
-   asr/reference
+   demo_video
+   tts_demo_video
+.. toctree::
+   :maxdepth: 1
+   :caption: Acknowledgement
+   asr/reference
--- a/docs/source/released_model.md
+++ b/docs/source/released_model.md
 # Released Models
 ## Speech-to-Text Models
@@ -9,9 +10,10 @@ Acoustic Model | Training Data | Token-based | Size | Descriptions | CER | WER |
 [Ds2 Offline Aishell ASR0 Model](https://paddlespeech.bj.bcebos.com/s2t/aishell/asr0/asr0_deepspeech2_aishell_ckpt_0.1.1.model.tar.gz)| Aishell Dataset | Char-based | 306 MB | 2 Conv + 3 bidirectional GRU layers| 0.064 |-| 151 h | [Ds2 Offline Aishell ASR0](../../examples/aishell/asr0) 
 [Conformer Offline Aishell ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/aishell/asr1/asr1_conformer_aishell_ckpt_0.1.1.model.tar.gz) | Aishell Dataset | Char-based | 284 MB  | Encoder:Conformer, Decoder:Transformer, Decoding method: Attention rescoring | 0.056 |-| 151 h | [Conformer Offline Aishell ASR1](../../examples/aishell/asr1) 
 [Transformer Aishell ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/aishell/asr1/asr1_transformer_aishell_ckpt_0.1.1.model.tar.gz) | Aishell Dataset | Char-based | 128 MB | Encoder:Transformer, Decoder:Transformer, Decoding method: Attention rescoring | 0.0523 || 151 h | [Transformer  Aishell ASR1](../../examples/aishell/asr1) 
-[Conformer Librispeech ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr1/asr1_conformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 191 MB | Encoder:Conformer, Decoder:Transformer, Decoding method: Attention rescoring |-| 0.0337 | 960 h | [Conformer Librispeech ASR1](../../example/librispeech/asr1) 
+[Ds2 Offline Librispeech ASR0 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr0/asr0_deepspeech2_librispeech_ckpt_0.1.1.model.tar.gz)| Librispeech Dataset | Char-based | 518 MB | 2 Conv + 3 bidirectional LSTM layers| - |0.0725| 960 h | [Ds2 Offline Librispeech ASR0](../../examples/librispeech/asr0) 
-[Transformer Librispeech ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr1/asr1_transformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 131 MB  | Encoder:Transformer, Decoder:Transformer, Decoding method: Attention rescoring |-| 0.0381 | 960 h | [Transformer Librispeech ASR1](../../example/librispeech/asr1) 
+[Conformer Librispeech ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr1/asr1_conformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 191 MB | Encoder:Conformer, Decoder:Transformer, Decoding method: Attention rescoring |-| 0.0337 | 960 h | [Conformer Librispeech ASR1](../../examples/librispeech/asr1) 
-[Transformer Librispeech ASR2 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr2/asr2_transformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 131 MB  | Encoder:Transformer, Decoder:Transformer, Decoding method: JoinCTC w/ LM |-| 0.0240 | 960 h | [Transformer Librispeech ASR2](../../example/librispeech/asr2) 
+[Transformer Librispeech ASR1 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr1/asr1_transformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 131 MB  | Encoder:Transformer, Decoder:Transformer, Decoding method: Attention rescoring |-| 0.0381 | 960 h | [Transformer Librispeech ASR1](../../examples/librispeech/asr1) 
+[Transformer Librispeech ASR2 Model](https://paddlespeech.bj.bcebos.com/s2t/librispeech/asr2/asr2_transformer_librispeech_ckpt_0.1.1.model.tar.gz) | Librispeech Dataset | subword-based | 131 MB  | Encoder:Transformer, Decoder:Transformer, Decoding method: JoinCTC w/ LM |-| 0.0240 | 960 h | [Transformer Librispeech ASR2](../../examples/librispeech/asr2) 
 ### Language Model based on NGram
 Language Model | Training Data | Token-based | Size | Descriptions
@@ -31,14 +33,15 @@ Language Model | Training Data | Token-based | Size | Descriptions
 ### Acoustic Models
 Model Type | Dataset| Example Link | Pretrained Models|Static Models|Size (static)
 :-------------:| :------------:| :-----: | :-----:| :-----:| :-----:
-Tacotron2|LJSpeech|[tacotron2-vctk](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/tts0)|[tacotron2_ljspeech_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.3.zip)|||
+Tacotron2|LJSpeech|[tacotron2-ljspeech](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/tts0)|[tacotron2_ljspeech_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.2.0.zip)|||
+Tacotron2|CSMSC|[tacotron2-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts0)|[tacotron2_csmsc_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_csmsc_ckpt_0.2.0.zip)|[tacotron2_csmsc_static_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_csmsc_static_0.2.0.zip)|103MB|
 TransformerTTS| LJSpeech| [transformer-ljspeech](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/tts1)|[transformer_tts_ljspeech_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/transformer_tts/transformer_tts_ljspeech_ckpt_0.4.zip)|||
 SpeedySpeech| CSMSC | [speedyspeech-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts2) |[speedyspeech_nosil_baker_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/speedyspeech/speedyspeech_nosil_baker_ckpt_0.5.zip)|[speedyspeech_nosil_baker_static_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/speedyspeech/speedyspeech_nosil_baker_static_0.5.zip)|12MB|
 FastSpeech2| CSMSC |[fastspeech2-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts3)|[fastspeech2_nosil_baker_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_baker_ckpt_0.4.zip)|[fastspeech2_nosil_baker_static_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_baker_static_0.4.zip)|157MB|
 FastSpeech2-Conformer| CSMSC |[fastspeech2-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts3)|[fastspeech2_conformer_baker_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_conformer_baker_ckpt_0.5.zip)|||
 FastSpeech2| AISHELL-3 |[fastspeech2-aishell3](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/tts3)|[fastspeech2_nosil_aishell3_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_aishell3_ckpt_0.4.zip)|||
 FastSpeech2| LJSpeech |[fastspeech2-ljspeech](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/tts3)|[fastspeech2_nosil_ljspeech_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_ljspeech_ckpt_0.5.zip)|||
-FastSpeech2| VCTK |[fastspeech2-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/vctk/tts3)|[fastspeech2_nosil_vctk_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_vctk_ckpt_0.5.zip)|||
+FastSpeech2| VCTK |[fastspeech2-vctk](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/vctk/tts3)|[fastspeech2_nosil_vctk_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_vctk_ckpt_0.5.zip)|||
 ### Vocoders
 Model Type | Dataset| Example Link | Pretrained Models| Static Models|Size (static)
@@ -51,12 +54,14 @@ Parallel WaveGAN| VCTK |[PWGAN-vctk](https://github.com/PaddlePaddle/PaddleSpeec
 |Multi Band MelGAN | CSMSC |[MB MelGAN-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc3) | [mb_melgan_csmsc_ckpt_0.1.1.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/mb_melgan/mb_melgan_csmsc_ckpt_0.1.1.zip) <br>[mb_melgan_baker_finetune_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/mb_melgan/mb_melgan_baker_finetune_ckpt_0.5.zip)|[mb_melgan_csmsc_static_0.1.1.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/mb_melgan/mb_melgan_csmsc_static_0.1.1.zip) |8.2MB|
 Style MelGAN | CSMSC |[Style MelGAN-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc4)|[style_melgan_csmsc_ckpt_0.1.1.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/style_melgan/style_melgan_csmsc_ckpt_0.1.1.zip)| | |
 HiFiGAN | CSMSC |[HiFiGAN-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc5)|[hifigan_csmsc_ckpt_0.1.1.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/hifigan/hifigan_csmsc_ckpt_0.1.1.zip)|[hifigan_csmsc_static_0.1.1.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/hifigan/hifigan_csmsc_static_0.1.1.zip)|50MB|
+WaveRNN | CSMSC |[WaveRNN-csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc6)|[wavernn_csmsc_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/wavernn/wavernn_csmsc_ckpt_0.2.0.zip)|[wavernn_csmsc_static_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/wavernn/wavernn_csmsc_static_0.2.0.zip)|18MB|
 ### Voice Cloning
 Model Type | Dataset| Example Link | Pretrained Models
 :-------------:| :------------:| :-----: | :-----:
 GE2E| AISHELL-3, etc. |[ge2e](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/ge2e)|[ge2e_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/ge2e/ge2e_ckpt_0.3.zip)
-GE2E + Tactron2| AISHELL-3 |[ge2e-tactron2-aishell3](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/vc0)|[tacotron2_aishell3_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_aishell3_ckpt_0.3.zip)
+GE2E + Tactron2| AISHELL-3 |[ge2e-tactron2-aishell3](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/vc0)|[tacotron2_aishell3_ckpt_vc0_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_aishell3_ckpt_vc0_0.2.0.zip)
 GE2E + FastSpeech2 | AISHELL-3  |[ge2e-fastspeech2-aishell3](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/vc1)|[fastspeech2_nosil_aishell3_vc1_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_aishell3_vc1_ckpt_0.5.zip)
@@ -65,7 +70,7 @@ GE2E + FastSpeech2 | AISHELL-3  |[ge2e-fastspeech2-aishell3](https://github.com/
 Model Type | Dataset| Example Link | Pretrained Models
 :-------------:| :------------:| :-----: | :-----:
 PANN | Audioset| [audioset_tagging_cnn](https://github.com/qiuqiangkong/audioset_tagging_cnn) | [panns_cnn6.pdparams](https://bj.bcebos.com/paddleaudio/models/panns_cnn6.pdparams), [panns_cnn10.pdparams](https://bj.bcebos.com/paddleaudio/models/panns_cnn10.pdparams), [panns_cnn14.pdparams](https://bj.bcebos.com/paddleaudio/models/panns_cnn14.pdparams)
-PANN | ESC-50 |[pann-esc50]("./examples/esc50/cls0")|[esc50_cnn6.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn6.tar.gz), [esc50_cnn10.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn10.tar.gz), [esc50_cnn14.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn14.tar.gz)
+PANN | ESC-50 |[pann-esc50](../../examples/esc50/cls0)|[esc50_cnn6.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn6.tar.gz), [esc50_cnn10.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn10.tar.gz), [esc50_cnn14.tar.gz](https://paddlespeech.bj.bcebos.com/cls/esc50/esc50_cnn14.tar.gz)
 ## Punctuation Restoration Models
 Model Type | Dataset| Example Link | Pretrained Models

--- a/docs/source/tts/README.md
+++ b/docs/source/tts/README.md
@@ -71,7 +71,3 @@ Check our [website](https://paddlespeech.readthedocs.io/en/latest/tts/demo.html)
 #### GE2E
 1. [ge2e_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/ge2e_ckpt_0.3.zip)
-## License
-Parakeet is provided under the [Apache-2.0 license](LICENSE).
--- a/docs/source/tts/quick_start.md
+++ b/docs/source/tts/quick_start.md
+([简体中文](./quick_start_cn.md)|English)
 # Quick Start of Text-to-Speech
 The examples in PaddleSpeech are mainly classified by datasets, the TTS datasets we mainly used are:
 * CSMCS (Mandarin single speaker)

--- a/docs/source/tts/quick_start_cn.md
+++ b/docs/source/tts/quick_start_cn.md
+(简体中文|[English](./quick_start.md))
+# 语音合成快速开始
+这些PaddleSpeech中的样例主要按数据集分类，我们主要使用的TTS数据集有：
+* CSMCS (普通话单发音人)
+* AISHELL3 (普通话多发音人)
+* LJSpeech (英文单发音人)
+* VCTK (英文多发音人)
+PaddleSpeech 的 TTS 模型具有以下映射关系：
+* tts0 - Tactron2
+* tts1 - TransformerTTS
+* tts2 - SpeedySpeech
+* tts3 - FastSpeech2
+* voc0 - WaveFlow
+* voc1 - Parallel WaveGAN
+* voc2 - MelGAN
+* voc3 - MultiBand MelGAN
+* voc4 - Style MelGAN
+* voc5 - HiFiGAN
+* vc0 - Tactron2 Voice Clone with GE2E
+* vc1 - FastSpeech2 Voice Clone with GE2E
+## 快速开始
+让我们以 FastSpeech2 + Parallel WaveGAN 和 CSMSC 数据集 为例. [examples/csmsc](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc)
+### 用 CSMSC 数据集训练 Parallel WaveGAN
+- 进入目录
+    ```bash
+    cd examples/csmsc/voc1
+    ```
+- 设置环境变量
+    ```bash
+    source path.sh
+    ```
+    **在你开始做任何事情之前，必须先做这步**
+    将 `MAIN_ROOT` 设置为项目目录. 使用 `parallelwave_gan` 模型作为 `MODEL`.
+- 运行
+    ```bash
+    bash run.sh
+    ```
+    这只是一个演示，请确保源数据已经准备好，并且在下一个 `step` 之前每个 `step` 都运行正常.
+### 用CSMSC数据集训练FastSpeech2
+- 进入目录
+    ```bash
+    cd examples/csmsc/tts3
+    ```
+- 设置环境变量
+    ```bash
+    source path.sh
+    ```
+    **在你开始做任何事情之前，必须先做这步**
+    将 `MAIN_ROOT` 设置为项目目录. 使用 `fastspeech2` 模型作为 `MODEL` 。
+- 运行
+    ```bash
+    bash run.sh
+    ```
+    这只是一个演示，请确保源数据已经准备好，并且在下一个 `step` 之前每个 `step` 都运行正常。
+`run.sh` 中主要包括以下步骤：
+- 设置路径。
+- 预处理数据集，
+- 训练模型。
+- 从 `metadata.jsonl` 中合成波形
+- 从文本文件合成波形。（在声学模型中）
+- 使用静态模型进行推理。（可选）
+有关更多详细信息，请参见 examples 中的 `README.md`
+## TTS 流水线
+本节介绍如何使用 TTS 提供的预训练模型，并对其进行推理。
+TTS中的预训练模型在压缩包中提供。将其解压缩以获得如下文件夹：
+**Acoustic Models:**
+```text
+checkpoint_name
+├── default.yaml
+├── snapshot_iter_*.pdz
+├── speech_stats.npy
+├── phone_id_map.txt
+├── spk_id_map.txt (optimal)
+└── tone_id_map.txt (optimal)
+```
+**Vocoders:**
+```text
+checkpoint_name
+├── default.yaml  
+├── snapshot_iter_*.pdz
+└── stats.npy  
+```
+- `default.yaml` 存储用于训练模型的配置。
+- `snapshot_iter_*.pdz` 是检查点文件，其中`*`是它经过训练的步骤。
+- `*_stats.npy` 是特征的统计文件，如果它在训练前已被标准化。
+- `phone_id_map.txt` 是音素到音素 ID 的映射关系。
+- `tone_id_map.txt` 是在训练声学模型之前分割音调和拼音时，音调到音调 ID 的映射关系。（例如在 csmsc/speedyspeech 的示例中）
+- `spk_id_map.txt` 是多发音人声学模型中 "发音人" 到 "spk_ids" 的映射关系。
+下面的示例代码显示了如何使用模型进行预测。
+### Acoustic Models 声学模型（文本到频谱图）
+下面的代码显示了如何使用 `FastSpeech2` 模型。加载预训练模型后，使用它和 normalizer 对象构建预测对象，然后使用 `fastspeech2_inferencet(phone_ids)` 生成频谱图，频谱图可进一步用于使用声码器合成原始音频。
+```python
+from pathlib import Path
+import numpy as np
+import paddle
+import yaml
+from yacs.config import CfgNode
+from paddlespeech.t2s.models.fastspeech2 import FastSpeech2
+from paddlespeech.t2s.models.fastspeech2 import FastSpeech2Inference
+from paddlespeech.t2s.modules.normalizer import ZScore
+# examples/fastspeech2/baker/frontend.py
+from frontend import Frontend
+# 加载预训练模型
+checkpoint_dir = Path("fastspeech2_nosil_baker_ckpt_0.4")
+with open(checkpoint_dir / "phone_id_map.txt", "r") as f:
+    phn_id = [line.strip().split() for line in f.readlines()]
+vocab_size = len(phn_id)
+with open(checkpoint_dir / "default.yaml") as f:
+    fastspeech2_config = CfgNode(yaml.safe_load(f))
+odim = fastspeech2_config.n_mels
+model = FastSpeech2(
+    idim=vocab_size, odim=odim, **fastspeech2_config["model"])
+model.set_state_dict(
+    paddle.load(args.fastspeech2_checkpoint)["main_params"])
+model.eval()
+# 加载特征文件
+stat = np.load(checkpoint_dir / "speech_stats.npy")
+mu, std = stat
+mu = paddle.to_tensor(mu)
+std = paddle.to_tensor(std)
+fastspeech2_normalizer = ZScore(mu, std)
+# 构建预测对象
+fastspeech2_inference = FastSpeech2Inference(fastspeech2_normalizer, model)
+# load Chinese Frontend
+frontend = Frontend(checkpoint_dir / "phone_id_map.txt")
+# 构建一个中文前端
+sentence = "你好吗？"
+input_ids = frontend.get_input_ids(sentence, merge_sentences=True)
+phone_ids = input_ids["phone_ids"]
+flags = 0
+# 构建预测对象加载中文前端，对中文文本前端的输出进行分段
+for part_phone_ids in phone_ids:
+    with paddle.no_grad():
+        temp_mel = fastspeech2_inference(part_phone_ids)
+        if flags == 0:
+            mel = temp_mel
+            flags = 1
+        else:
+            mel = paddle.concat([mel, temp_mel])
+```
+### Vcoder声码器（谱图到波形）
+下面的代码显示了如何使用 `Parallel WaveGAN` 模型。像上面的例子一样，加载预训练模型后，使用它和 normalizer 对象构建预测对象，然后使用 `pwg_inference(mel)` 生成原始音频（ wav 格式）。
+```python
+from pathlib import Path
+import numpy as np
+import paddle
+import soundfile as sf
+import yaml
+from yacs.config import CfgNode
+from paddlespeech.t2s.models.parallel_wavegan import PWGGenerator
+from paddlespeech.t2s.models.parallel_wavegan import PWGInference
+from paddlespeech.t2s.modules.normalizer import ZScore
+# 加载预训练模型
+checkpoint_dir = Path("parallel_wavegan_baker_ckpt_0.4")
+with open(checkpoint_dir / "pwg_default.yaml") as f:
+    pwg_config = CfgNode(yaml.safe_load(f))
+vocoder = PWGGenerator(**pwg_config["generator_params"])
+vocoder.set_state_dict(paddle.load(args.pwg_params))
+vocoder.remove_weight_norm()
+vocoder.eval()
+# 加载特征文件
+stat = np.load(checkpoint_dir / "pwg_stats.npy")
+mu, std = stat
+mu = paddle.to_tensor(mu)
+std = paddle.to_tensor(std)
+pwg_normalizer = ZScore(mu, std)
+# 加载预训练模型构造预测对象
+pwg_inference = PWGInference(pwg_normalizer, vocoder)
+# 频谱图到波形
+wav = pwg_inference(mel)
+sf.write(
+        audio_path,
+        wav.numpy(),
+        samplerate=fastspeech2_config.fs)
+```
--- a/docs/source/tts/tts_datasets.md
+++ b/docs/source/tts/tts_datasets.md
+# TTS Datasets
+<!--
+see https://openslr.org/
+-->
+## Mandarin
+- [CSMSC](https://www.data-baker.com/open_source.html): Chinese Standard Mandarin Speech Copus
+    - Duration/h: 12
+    - Number of Sentences: 10,000
+    - Size: 2.14GB
+    - Speaker: 1 female, ages 20 ~30
+    - Sample Rate: 48 kHz、16bit
+    - Mean Words per Clip: 16
+- [AISHELL-3](http://www.aishelltech.com/aishell_3)
+    - Duration/h: 85
+    - Number of Sentences: 88,035
+    - Size: 17.75GB
+    - Speaker: 218
+    - Sample Rate: 44.1 kHz、16bit 
+## English
+- [LJSpeech](https://keithito.com/LJ-Speech-Dataset/)
+    - Duration/h: 24
+    - Number of Sentences: 13,100
+    - Size: 2.56GB
+    - Speaker: 1, age 20 ~30
+    - Sample Rate: 22050 Hz、16bit
+    - Mean Words per Clip: 17.23
+- [VCTK](https://datashare.ed.ac.uk/handle/10283/3443)
+    - Number of Sentences: 44,583
+    - Size: 10.94GB
+    - Speaker: 110 
+    - Sample Rate: 48 kHz、16bit
+    - Mean Words per Clip: 17.23
+## Japanese
+<!--
+see https://sites.google.com/site/shinnosuketakamichi/publication/corpus
+-->
+- [tri-jek](https://sites.google.com/site/shinnosuketakamichi/research-topics/tri-jek_corpus): Japanese-English-Korean tri-lingual corpus
+- [JSSS-misc](https://sites.google.com/site/shinnosuketakamichi/research-topics/jsss-misc_corpus): misc tasks of JSSS corpus
+- [JTubeSpeech](https://github.com/sarulab-speech/jtubespeech): Corpus of Japanese speech collected from YouTube
+- [J-MAC](https://sites.google.com/site/shinnosuketakamichi/research-topics/j-mac_corpus): Japanese multi-speaker audiobook corpus
+- [J-KAC](https://sites.google.com/site/shinnosuketakamichi/research-topics/j-kac_corpus): Japanese Kamishibai and audiobook corpus
+- [JMD](https://sites.google.com/site/shinnosuketakamichi/research-topics/jmd_corpus): Japanese multi-dialect corpus
+- [JSSS](https://sites.google.com/site/shinnosuketakamichi/research-topics/jsss_corpus): Japanese multi-style (summarization and simplification) corpus
+- [RWCP-SSD-Onomatopoeia](https://www.ksuke.net/dataset/rwcp-ssd-onomatopoeia): onomatopoeic word dataset for environmental sounds 
+- [Life-m](https://sites.google.com/site/shinnosuketakamichi/research-topics/life-m_corpus): landmark image-themed music corpus
+- [PJS](https://sites.google.com/site/shinnosuketakamichi/research-topics/pjs_corpus): Phoneme-balanced Japanese singing voice corpus
+- [JVS-MuSiC](https://sites.google.com/site/shinnosuketakamichi/research-topics/jvs_music): Japanese multi-speaker singing-voice corpus
+- [JVS](https://sites.google.com/site/shinnosuketakamichi/research-topics/jvs_corpus): Japanese multi-speaker voice corpus
+- [JSUT-book](https://sites.google.com/site/shinnosuketakamichi/publication/jsut-book): audiobook corpus by a single Japanese speaker
+- [JSUT-vi](https://sites.google.com/site/shinnosuketakamichi/publication/jsut-vi): vocal imitation corpus by a single Japanese speaker
+- [JSUT-song](https://sites.google.com/site/shinnosuketakamichi/publication/jsut-song): singing voice corpus by a single Japanese singer
+- [JSUT](https://sites.google.com/site/shinnosuketakamichi/publication/jsut): a large-scaled corpus of reading-style Japanese speech by a single speaker
+## Emotions
+### English
+- [CREMA-D](https://github.com/CheyneyComputerScience/CREMA-D)
+- [Seen and Unseen emotional style transfer for voice conversion with a new emotional speech dataset](https://kunzhou9646.github.io/controllable-evc/)
+    - paper : [Seen and Unseen emotional style transfer for voice conversion with a new emotional speech dataset](https://arxiv.org/abs/2010.14794)
+### Mandarin
+- [EMOVIE Dataset](https://viem-ccy.github.io/EMOVIE/dataset_release )
+    - paper: [EMOVIE: A Mandarin Emotion Speech Dataset with a Simple Emotional Text-to-Speech Model](https://arxiv.org/abs/2106.09317)
+- MASC
+    - paper: [MASC: A Speech Corpus in Mandarin for Emotion Analysis and Affective Speaker Recognition](https://ieeexplore.ieee.org/document/4013501)
+### English && Mandarin
+- [Emotional Voice Conversion: Theory, Databases and ESD](https://github.com/HLTSingapore/Emotional-Speech-Data)    
+    - paper: [Emotional Voice Conversion: Theory, Databases and ESD](https://arxiv.org/abs/2105.14762) 
+## Music
+- [GiantMIDI-Piano](https://github.com/bytedance/GiantMIDI-Piano)
+- [MAESTRO Dataset](https://magenta.tensorflow.org/datasets/maestro)
+     - [tf code](https://www.tensorflow.org/tutorials/audio/music_generation) 
+- [Opencpop](https://wenet.org.cn/opencpop/)
--- a/docs/source/tts_demo_video.rst
+++ b/docs/source/tts_demo_video.rst
+TTS Demo Video
+==================
+.. raw:: html
+    <video controls width="1024">
+    <source src="https://paddlespeech.bj.bcebos.com/Parakeet/docs/demos/paddle2021_with_me.mp4"
+            type="video/mp4">
+    Sorry, your browser doesn't support embedded videos.
+    </video>
--- a/docs/tutorial/asr/tutorial_deepspeech2.ipynb
+++ b/docs/tutorial/asr/tutorial_deepspeech2.ipynb
@@ -265,7 +265,7 @@
   },
   "outputs": [],
   "source": [
-    "!pip install --upgrade pip && pip install paddlespeech"
+    "!pip install --upgrade pip && pip install paddlespeech==0.1.0"
   ]
  },
  {

--- a/docs/tutorial/asr/tutorial_transformer.ipynb
+++ b/docs/tutorial/asr/tutorial_transformer.ipynb
@@ -138,7 +138,7 @@
   },
   "outputs": [],
   "source": [
-    "!pip install --upgrade pip && pip install paddlespeech"
+    "!pip install --upgrade pip && pip install paddlespeech==0.1.0"
   ]
  },
  {

--- a/examples/aishell/asr0/conf/tuning/chunk_decode.yaml
+++ b/examples/aishell/asr0/conf/tuning/chunk_decode.yaml
-chunk_batch_size: 32
+decode_batch_size: 32
 error_rate_type: cer 
 decoding_method: ctc_beam_search
 lang_model_path: data/lm/zh_giga.no_cna_cmn.prune01244.klm
 alpha: 2.2 #1.9
 beta: 4.3
-beam_size: 300
+beam_size: 500
 cutoff_prob: 0.99
 cutoff_top_n: 40
 num_proc_bsearch: 10
--- a/examples/aishell3/tts3/README.md
+++ b/examples/aishell3/tts3/README.md
@@ -257,6 +257,7 @@ python3 ${BIN_DIR}/../synthesize_e2e.py \
  --output_dir=exp/default/test_e2e \
  --phones_dict=fastspeech2_nosil_aishell3_ckpt_0.4/phone_id_map.txt \
  --speaker_dict=fastspeech2_nosil_aishell3_ckpt_0.4/speaker_id_map.txt \
-  --spk_id=0
+  --spk_id=0 \
+  --inference_dir=exp/default/inference
 ```
--- a/examples/aishell3/tts3/conf/default.yaml
+++ b/examples/aishell3/tts3/conf/default.yaml
@@ -16,8 +16,8 @@ fmax: 7600         # Maximum frequency of Mel basis.
 n_mels: 80         # The number of mel basis.
 # Only used for the model using pitch features (e.g. FastSpeech2)
-f0min: 80          # Maximum f0 for pitch extraction.
+f0min: 80          # Minimum f0 for pitch extraction.
-f0max: 400         # Minimum f0 for pitch extraction.
+f0max: 400         # Maximum f0 for pitch extraction.
 ###########################################################
@@ -64,14 +64,14 @@ model:
    pitch_predictor_dropout: 0.5               # dropout rate in pitch predictor
    pitch_embed_kernel_size: 1                 # kernel size of conv embedding layer for pitch
    pitch_embed_dropout: 0.0                   # dropout rate after conv embedding layer for pitch
-    stop_gradient_from_pitch_predictor: true   # whether to stop the gradient from pitch predictor to encoder
+    stop_gradient_from_pitch_predictor: True   # whether to stop the gradient from pitch predictor to encoder
    energy_predictor_layers: 2                 # number of conv layers in energy predictor
    energy_predictor_chans: 256                # number of channels of conv layers in energy predictor
    energy_predictor_kernel_size: 3            # kernel size of conv leyers in energy predictor
    energy_predictor_dropout: 0.5              # dropout rate in energy predictor
    energy_embed_kernel_size: 1                # kernel size of conv embedding layer for energy
    energy_embed_dropout: 0.0                  # dropout rate after conv embedding layer for energy
-    stop_gradient_from_energy_predictor: false # whether to stop the gradient from energy predictor to encoder
+    stop_gradient_from_energy_predictor: False # whether to stop the gradient from energy predictor to encoder
    spk_embed_dim: 256                         # speaker embedding dimension
    spk_embed_integration_type: concat         # speaker embedding integration type
@@ -84,7 +84,6 @@ updater:
    use_masking: True                 # whether to apply masking for padded part in loss calculation
 ###########################################################
 #                     OPTIMIZER SETTING                   #
 ###########################################################

--- a/examples/aishell3/tts3/local/inference.sh
+++ b/examples/aishell3/tts3/local/inference.sh
+#!/bin/bash
+train_output_path=$1
+stage=0
+stop_stage=0
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=fastspeech2_aishell3 \
+        --voc=pwgan_aishell3 \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt \
+        --speaker_dict=dump/speaker_id_map.txt \
+        --spk_id=0
+fi
--- a/examples/aishell3/tts3/local/synthesize_e2e.sh
+++ b/examples/aishell3/tts3/local/synthesize_e2e.sh
@@ -20,4 +20,5 @@ python3 ${BIN_DIR}/../synthesize_e2e.py \
    --output_dir=${train_output_path}/test_e2e \
    --phones_dict=dump/phone_id_map.txt \
    --speaker_dict=dump/speaker_id_map.txt \
-    --spk_id=0
+    --spk_id=0 \
+    --inference_dir=${train_output_path}/inference
--- a/examples/aishell3/vc0/README.md
+++ b/examples/aishell3/vc0/README.md
 # Tacotron2 + AISHELL-3 Voice Cloning
-This example contains code used to train a [Tacotron2 ](https://arxiv.org/abs/1712.05884) model with [AISHELL-3](http://www.aishelltech.com/aishell_3). The trained model can be used in Voice Cloning Task, We refer to the model structure of  [Transfer Learning from Speaker Veriﬁcation to Multispeaker Text-To-Speech Synthesis](https://arxiv.org/pdf/1806.04558.pdf). The general steps are as follows:
+This example contains code used to train a [Tacotron2](https://arxiv.org/abs/1712.05884) model with [AISHELL-3](http://www.aishelltech.com/aishell_3). The trained model can be used in Voice Cloning Task, We refer to the model structure of  [Transfer Learning from Speaker Veriﬁcation to Multispeaker Text-To-Speech Synthesis](https://arxiv.org/pdf/1806.04558.pdf). The general steps are as follows:
-1. Speaker Encoder: We use Speaker Verification to train a speaker encoder. Datasets used in this task are different from those used in Tacotron2 because the transcriptions are not needed, we use more datasets, refer to  [ge2e](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/ge2e).
+1. Speaker Encoder: We use Speaker Verification to train a speaker encoder. Datasets used in this task are different from those used in `Tacotron2` because the transcriptions are not needed, we use more datasets, refer to  [ge2e](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/ge2e).
-2. Synthesizer: We use the trained speaker encoder to generate speaker embedding for each sentence in AISHELL-3. This embedding is an extra input of  Tacotron2 which will be concated with encoder outputs.
+2. Synthesizer: We use the trained speaker encoder to generate speaker embedding for each sentence in AISHELL-3. This embedding is an extra input of  `Tacotron2` which will be concated with encoder outputs.
-3. Vocoder: We use WaveFlow as the neural Vocoder, refer to [waveflow](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/voc0).
+3. Vocoder: We use [Parallel Wave GAN](http://arxiv.org/abs/1910.11480) as the neural Vocoder, refer to [voc1](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/voc1).
+## Dataset
+### Download and Extract
+Download AISHELL-3.
+```bash
+wget https://www.openslr.org/resources/93/data_aishell3.tgz
+```
+Extract AISHELL-3.
+```bash
+mkdir data_aishell3
+tar zxvf data_aishell3.tgz -C data_aishell3
+```
+### Get MFA Result and Extract
+We use [MFA2.x](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) to get phonemes for Tacotron2, the durations of MFA are not needed here.
+You can download from here [aishell3_alignment_tone.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/AISHELL-3/with_tone/aishell3_alignment_tone.tar.gz), or train your MFA model reference to [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) (use MFA1.x now) of our repo.
+## Pretrained GE2E Model
+We use pretrained GE2E model to generate speaker embedding for each sentence.
+Download pretrained GE2E model from here [ge2e_ckpt_0.3.zip](https://bj.bcebos.com/paddlespeech/Parakeet/released_models/ge2e/ge2e_ckpt_0.3.zip), and `unzip` it.
 ## Get Started
 Assume the path to the dataset is `~/datasets/data_aishell3`.
-Assume the path to the MFA result of AISHELL-3 is `./alignment`.
+Assume the path to the MFA result of AISHELL-3 is `./aishell3_alignment_tone`.
-Assume the path to the pretrained ge2e model is `ge2e_ckpt_path=./ge2e_ckpt_0.3/step-3000000`
+Assume the path to the pretrained ge2e model is `./ge2e_ckpt_0.3`.
 Run the command below to
 1. **source path**.
 2. preprocess the dataset.
 3. train the model.
-4. start a voice cloning inference.
+4. synthesize waveform from `metadata.jsonl`.
+5. start a voice cloning inference.
 ```bash
 ./run.sh
 ```
-You can choose a range of stages you want to run, or set `stage` equal to `stop-stage` to use only one stage, for example, run the following command will only preprocess the dataset.
+You can choose a range of stages you want to run, or set `stage` equal to `stop-stage` to use only one stage, for example, running the following command will only preprocess the dataset.
 ```bash
 ./run.sh --stage 0 --stop-stage 0
 ```
 ### Data Preprocessing
 ```bash
-CUDA_VISIBLE_DEVICES=${gpus} ./local/preprocess.sh ${input} ${preprocess_path} ${alignment} ${ge2e_ckpt_path}
+CUDA_VISIBLE_DEVICES=${gpus} ./local/preprocess.sh ${conf_path} ${ge2e_ckpt_path}
 ```
-#### Generate Speaker Embedding
+When it is done. A `dump` folder is created in the current directory. The structure of the dump folder is listed below.
- Use pretrained GE2E (speaker encoder) to generate speaker embedding for each sentence in AISHELL-3, which has the same file structure with wav files and the format is  `.npy`.
+```text
+dump
-```bash
+├── dev
-if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+│   ├── norm
-    python3 ${BIN_DIR}/../ge2e/inference.py \
+│   └── raw
-        --input=${input} \
+├── embed
-        --output=${preprocess_path}/embed \
+│   ├── SSB0005
-        --ngpu=1 \
+│   ├── SSB0009
-        --checkpoint_path=${ge2e_ckpt_path}
+│   ├── ...
-fi
+│   └── ...
+├── phone_id_map.txt
+├── speaker_id_map.txt
+├── test
+│   ├── norm
+│   └──  raw
+└── train
+    ├── norm
+    ├── raw
+    └── speech_stats.npy
 ```
+The `embed` contains the generated speaker embedding for each sentence in AISHELL-3, which has the same file structure with wav files and the format is  `.npy`.
 The computing time of utterance embedding can be x hours.
-#### Process Wav
-There is silence in the edge of AISHELL-3's wavs, and the audio amplitude is very small, so, we need to remove the silence and normalize the audio. You can the silence remove method based on volume or energy, but the effect is not very good, We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) to get the alignment of text and speech, then utilize the alignment results to remove the silence.
-We use Montreal Force Aligner 1.0. The label in  aishell3 includes pinyin，so the lexicon we provided to MFA is pinyin rather than Chinese characters. And the prosody marks(`$`  and `%`) need to be removed. You should preprocess the dataset into the format which MFA needs, the texts have the same name with wavs and have the suffix `.lab`.
+The dataset is split into 3 parts, namely `train`, `dev`, and` test`, each of which contains a `norm` and `raw` subfolder. The raw folder contains speech features of each utterance, while the norm folder contains normalized ones. The statistics used to normalize features are computed from the training set, which is located in `dump/train/*_stats.npy`.
-We use [lexicon.txt](https://github.com/PaddlePaddle/PaddleSpeech/blob/develop/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/lexicon.txt) as the lexicon.
+Also, there is a `metadata.jsonl` in each subfolder. It is a table-like file that contains phones, text_lengths, speech_lengths, durations, the path of speech features, speaker, and id of each utterance.
-You can download the alignment results from here [alignment_aishell3.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/AISHELL-3/alignment_aishell3.tar.gz), or train your MFA model reference to [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) (use MFA1.x now) of our repo.
+The preprocessing step is very similar to that one of [tts0](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts0), but there is one more `ge2e/inference` step here.
+### Model Training
+`./local/train.sh` calls `${BIN_DIR}/train.py`.
 ```bash
-if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
-    echo "Process wav ..."
-    python3 ${BIN_DIR}/process_wav.py \
-        --input=${input}/wav \
-        --output=${preprocess_path}/normalized_wav \
-        --alignment=${alignment}
-fi
 ```
+The training step is very similar to that one of [tts0](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts0), but we should set `--voice-cloning=True` when calling `${BIN_DIR}/train.py`.
-#### Preprocess Transcription
+### Synthesizing
-We revert the transcription into `phones` and  `tones`. It is worth noting that our processing here is different from that used for MFA, we separated the tones. This is a processing method, of course, you can only segment initials and vowels.
+We use [parallel wavegan](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/voc1) as the neural vocoder.
+Download pretrained parallel wavegan model from [pwg_aishell3_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_aishell3_ckpt_0.5.zip) and unzip it.
 ```bash
-if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+unzip pwg_aishell3_ckpt_0.5.zip
-    python3 ${BIN_DIR}/preprocess_transcription.py \
-        --input=${input} \
-        --output=${preprocess_path}
-fi
 ```
-The default input is  `~/datasets/data_aishell3/train`，which contains `label_train-set.txt`, the processed results are `metadata.yaml` and  `metadata.pickle`. the former is a text format for easy viewing, and the latter is a binary format for direct reading.
+Parallel WaveGAN checkpoint contains files listed below.
-#### Extract Mel
+```text
-```python
+pwg_aishell3_ckpt_0.5
-if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+├── default.yaml                   # default config used to train parallel wavegan
-    python3 ${BIN_DIR}/extract_mel.py \
+├── feats_stats.npy                # statistics used to normalize spectrogram when training parallel wavegan
-        --input=${preprocess_path}/normalized_wav \
+└── snapshot_iter_1000000.pdz      # generator parameters of parallel wavegan
-        --output=${preprocess_path}/mel
-fi
 ```
+`./local/synthesize.sh` calls `${BIN_DIR}/../synthesize.py`, which can synthesize waveform from `metadata.jsonl`.
-### Model Training
 ```bash
-CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${preprocess_path} ${train_output_path}
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
 ```
+The synthesizing step is very similar to that one of [tts0](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/tts0), but we should set `--voice-cloning=True` when calling `${BIN_DIR}/../synthesize.py`.
-Our model removes stop token prediction in Tacotron2, because of the problem of the extremely unbalanced proportion of positive and negative samples of stop token prediction, and it's very sensitive to the clip of audio silence. We use the last symbol from the highest point of attention to the encoder side as the termination condition.
-In addition, to accelerate the convergence of the model, we add `guided attention loss` to induce the alignment between encoder and decoder to show diagonal lines faster.
 ### Voice Cloning
+Assume there are some  reference audios in `./ref_audio`
+```text
+ref_audio
+├── 001238.wav
+├── LJ015-0254.wav
+└── audio_self_test.mp3
+```
+`./local/voice_cloning.sh` calls `${BIN_DIR}/../voice_cloning.py`
 ```bash
-CUDA_VISIBLE_DEVICES=${gpus} ./local/voice_cloning.sh ${ge2e_params_path} ${tacotron2_params_path} ${waveflow_params_path} ${vc_input} ${vc_output}
+CUDA_VISIBLE_DEVICES=${gpus} ./local/voice_cloning.sh ${conf_path} ${train_output_path} ${ckpt_name} ${ge2e_params_path} ${ref_audio_dir}
 ```
 ## Pretrained Model
-[tacotron2_aishell3_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_aishell3_ckpt_0.3.zip).
+[tacotron2_aishell3_ckpt_vc0_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_aishell3_ckpt_vc0_0.2.0.zip)
+Model | Step | eval/loss | eval/l1_loss | eval/mse_loss | eval/bce_loss| eval/attn_loss
+:-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:
+default| 2(gpu) x 37596|0.58704|0.39623|0.15073|0.039|1.9981e-04|
+Tacotron2 checkpoint contains files listed below.
+(There is no need for `speaker_id_map.txt` here )
+```text
+tacotron2_aishell3_ckpt_vc0_0.2.0
+├── default.yaml            # default config used to train tacotron2
+├── phone_id_map.txt        # phone vocabulary file when training tacotron2
+├── snapshot_iter_37596.pdz # model parameters and optimizer states
+└── speech_stats.npy        # statistics used to normalize spectrogram when training tacotron2
+```
+## More
+We strongly recommend that you use [FastSpeech2 + AISHELL-3 Voice Cloning](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/aishell3/vc1) which works better.
--- a/examples/aishell3/vc0/conf/default.yaml
+++ b/examples/aishell3/vc0/conf/default.yaml
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+fs: 24000          # sr
+n_fft: 2048        # FFT size (samples).
+n_shift: 300       # Hop size (samples). 12.5ms
+win_length: 1200   # Window length (samples). 50ms
+                   # If set to null, it will be the same as fft_size.
+window: "hann"     # Window function.
+# Only used for feats_type != raw
+fmin: 80           # Minimum frequency of Mel basis.
+fmax: 7600         # Maximum frequency of Mel basis.
+n_mels: 80         # The number of mel basis.
+###########################################################
+#                       DATA SETTING                      #
+###########################################################
+batch_size: 64
+num_workers: 2
+###########################################################
+#                       MODEL SETTING                     #
+###########################################################
+model:                          # keyword arguments for the selected model
+    embed_dim: 512               # char or phn embedding dimension
+    elayers: 1                   # number of blstm layers in encoder
+    eunits: 512                  # number of blstm units
+    econv_layers: 3              # number of convolutional layers in encoder
+    econv_chans: 512             # number of channels in convolutional layer
+    econv_filts: 5               # filter size of convolutional layer
+    atype: location              # attention function type
+    adim: 512                    # attention dimension
+    aconv_chans: 32              # number of channels in convolutional layer of attention
+    aconv_filts: 15              # filter size of convolutional layer of attention
+    cumulate_att_w: True         # whether to cumulate attention weight
+    dlayers: 2                   # number of lstm layers in decoder
+    dunits: 1024                 # number of lstm units in decoder
+    prenet_layers: 2             # number of layers in prenet
+    prenet_units: 256            # number of units in prenet
+    postnet_layers: 5            # number of layers in postnet
+    postnet_chans: 512           # number of channels in postnet
+    postnet_filts: 5             # filter size of postnet layer
+    output_activation: null      # activation function for the final output
+    use_batch_norm: True         # whether to use batch normalization in encoder
+    use_concate: True            # whether to concatenate encoder embedding with decoder outputs
+    use_residual: False          # whether to use residual connection in encoder
+    dropout_rate: 0.5            # dropout rate
+    zoneout_rate: 0.1            # zoneout rate
+    reduction_factor: 1          # reduction factor
+    spk_embed_dim: 256           # speaker embedding dimension
+    spk_embed_integration_type: concat # how to integrate speaker embedding
+###########################################################
+#                       UPDATER SETTING                   #
+###########################################################
+updater:
+    use_masking: True            # whether to apply masking for padded part in loss calculation
+    bce_pos_weight: 5.0          # weight of positive sample in binary cross entropy calculation
+    use_guided_attn_loss: True   # whether to use guided attention loss
+    guided_attn_loss_sigma: 0.4  # sigma of guided attention loss
+    guided_attn_loss_lambda: 1.0 # strength of guided attention loss
+##########################################################
+#                  OPTIMIZER SETTING                     #
+##########################################################
+optimizer:
+    optim: adam              # optimizer type
+    learning_rate: 1.0e-03   # learning rate
+    epsilon: 1.0e-06         # epsilon
+    weight_decay: 0.0        # weight decay coefficient
+###########################################################
+#                     TRAINING SETTING                    #
+###########################################################
+max_epoch: 100
+num_snapshots: 5
+###########################################################
+#                       OTHER SETTING                     #
+###########################################################
+seed: 42
\ No newline at end of file
--- a/examples/aishell3/vc0/local/preprocess.sh
+++ b/examples/aishell3/vc0/local/preprocess.sh
 #!/bin/bash
-stage=0
+stage=3
 stop_stage=100
-input=$1
+config_path=$1
-preprocess_path=$2
+ge2e_ckpt_path=$2
-alignment=$3
-ge2e_ckpt_path=$4
+# gen speaker embedding
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    python3 ${MAIN_ROOT}/paddlespeech/vector/exps/ge2e/inference.py \
-        --input=${input}/wav \
+        --input=~/datasets/data_aishell3/train/wav/ \
-        --output=${preprocess_path}/embed \
+        --output=dump/embed \
        --checkpoint_path=${ge2e_ckpt_path}
 fi
+# copy from tts3/preprocess
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    echo "Process wav ..."
+    # get durations from MFA's result
-    python3 ${BIN_DIR}/process_wav.py \
+    echo "Generate durations.txt from MFA results ..."
-        --input=${input}/wav \
+    python3 ${MAIN_ROOT}/utils/gen_duration_from_textgrid.py \
-        --output=${preprocess_path}/normalized_wav \
+        --inputdir=./aishell3_alignment_tone \
-        --alignment=${alignment}
+        --output durations.txt \
+        --config=${config_path}
 fi
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    python3 ${BIN_DIR}/preprocess_transcription.py \
+    # extract features
-        --input=${input} \
+    echo "Extract features ..."
-        --output=${preprocess_path}
+    python3 ${BIN_DIR}/preprocess.py \
+        --dataset=aishell3 \
+        --rootdir=~/datasets/data_aishell3/ \
+        --dumpdir=dump \
+        --dur-file=durations.txt \
+        --config=${config_path} \
+        --num-cpu=20 \
+        --cut-sil=True \
+        --spk_emb_dir=dump/embed
 fi
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
-    python3 ${BIN_DIR}/extract_mel.py \
+    # get features' stats(mean and std)
-        --input=${preprocess_path}/normalized_wav \
+    echo "Get features' stats ..."
-        --output=${preprocess_path}/mel
+    python3 ${MAIN_ROOT}/utils/compute_statistics.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --field-name="speech"
+fi
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    # normalize and covert phone to id, dev and test should use train's stats
+    echo "Normalize ..."
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --dumpdir=dump/train/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/dev/raw/metadata.jsonl \
+        --dumpdir=dump/dev/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/test/raw/metadata.jsonl \
+        --dumpdir=dump/test/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
 fi
--- a/examples/aishell3/vc0/local/synthesize.sh
+++ b/examples/aishell3/vc0/local/synthesize.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+ckpt_name=$3
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize.py \
+    --am=tacotron2_aishell3 \
+    --am_config=${config_path} \
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_stat=dump/train/speech_stats.npy \
+    --voc=pwgan_aishell3 \
+    --voc_config=pwg_aishell3_ckpt_0.5/default.yaml \
+    --voc_ckpt=pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz \
+    --voc_stat=pwg_aishell3_ckpt_0.5/feats_stats.npy \
+    --test_metadata=dump/test/norm/metadata.jsonl \
+    --output_dir=${train_output_path}/test \
+    --phones_dict=dump/phone_id_map.txt \
+    --speaker_dict=dump/speaker_id_map.txt \
+    --voice-cloning=True
--- a/examples/aishell3/vc0/local/train.sh
+++ b/examples/aishell3/vc0/local/train.sh
 #!/bin/bash
-preprocess_path=$1
+config_path=$1
 train_output_path=$2
 python3 ${BIN_DIR}/train.py \
-    --data=${preprocess_path} \
+    --train-metadata=dump/train/norm/metadata.jsonl \
-    --output=${train_output_path} \
+    --dev-metadata=dump/dev/norm/metadata.jsonl \
-    --ngpu=1
+    --config=${config_path} \
\ No newline at end of file
+    --output-dir=${train_output_path} \
+    --ngpu=2 \
+    --phones-dict=dump/phone_id_map.txt \
+    --voice-cloning=True
\ No newline at end of file
--- a/examples/aishell3/vc0/local/voice_cloning.sh
+++ b/examples/aishell3/vc0/local/voice_cloning.sh
 #!/bin/bash
-ge2e_params_path=$1
+config_path=$1
-tacotron2_params_path=$2
+train_output_path=$2
-waveflow_params_path=$3
+ckpt_name=$3
-vc_input=$4
+ge2e_params_path=$4
-vc_output=$5
+ref_audio_dir=$5
-python3 ${BIN_DIR}/voice_cloning.py \
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../voice_cloning.py \
+    --am=tacotron2_aishell3 \
+    --am_config=${config_path} \
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_stat=dump/train/speech_stats.npy \
+    --voc=pwgan_aishell3 \
+    --voc_config=pwg_aishell3_ckpt_0.5/default.yaml \
+    --voc_ckpt=pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz \
+    --voc_stat=pwg_aishell3_ckpt_0.5/feats_stats.npy \
    --ge2e_params_path=${ge2e_params_path} \
-    --tacotron2_params_path=${tacotron2_params_path} \
+    --text="凯莫瑞安联合体的经济崩溃迫在眉睫。" \
-    --waveflow_params_path=${waveflow_params_path} \
+    --input-dir=${ref_audio_dir} \
-    --input-dir=${vc_input} \
+    --output-dir=${train_output_path}/vc_syn \
-    --output-dir=${vc_output}
+    --phones-dict=dump/phone_id_map.txt
\ No newline at end of file
--- a/examples/aishell3/vc0/path.sh
+++ b/examples/aishell3/vc0/path.sh
@@ -9,5 +9,5 @@ export PYTHONDONTWRITEBYTECODE=1
 export PYTHONIOENCODING=UTF-8
 export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
-MODEL=voice_cloning/tacotron2_ge2e
+MODEL=tacotron2
 export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
--- a/examples/aishell3/vc0/run.sh
+++ b/examples/aishell3/vc0/run.sh
@@ -3,25 +3,20 @@
 set -e
 source path.sh
-gpus=0
+gpus=0,1
 stage=0
 stop_stage=100
-input=~/datasets/data_aishell3/train
+conf_path=conf/default.yaml
-preprocess_path=dump
+train_output_path=exp/default
-alignment=./alignment
+ckpt_name=snapshot_iter_482.pdz
+ref_audio_dir=ref_audio
 # not include ".pdparams" here
 ge2e_ckpt_path=./ge2e_ckpt_0.3/step-3000000
-train_output_path=output
 # include ".pdparams" here
 ge2e_params_path=${ge2e_ckpt_path}.pdparams
-tacotron2_params_path=${train_output_path}/checkpoints/step-1000.pdparams
-# pretrained model
-# tacotron2_params_path=./tacotron2_aishell3_ckpt_0.3/step-450000.pdparams
-waveflow_params_path=./waveflow_ljspeech_ckpt_0.3/step-2000000.pdparams
-vc_input=ref_audio
-vc_output=syn_audio
 # with the following command, you can choose the stage range you want to run
 # such as `./run.sh --stage 0 --stop-stage 0`
@@ -30,15 +25,20 @@ source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    # prepare data
-    CUDA_VISIBLE_DEVICES=${gpus} ./local/preprocess.sh ${input} ${preprocess_path} ${alignment} ${ge2e_ckpt_path} || exit -1
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/preprocess.sh ${conf_path} ${ge2e_ckpt_path} || exit -1
 fi
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${preprocess_path} ${train_output_path} || exit -1
+    # train model, all `ckpt` under `train_output_path/checkpoints/` dir
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1
 fi
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    CUDA_VISIBLE_DEVICES=${gpus} ./local/voice_cloning.sh ${ge2e_params_path} ${tacotron2_params_path} ${waveflow_params_path} ${vc_input} ${vc_output} || exit -1
+    # synthesize, vocoder is pwgan
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
 fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # synthesize, vocoder is pwgan
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/voice_cloning.sh ${conf_path} ${train_output_path} ${ckpt_name} ${ge2e_params_path} ${ref_audio_dir} || exit -1
+fi
--- a/examples/aishell3/vc1/README.md
+++ b/examples/aishell3/vc1/README.md
 # FastSpeech2 + AISHELL-3 Voice Cloning
 This example contains code used to train a [FastSpeech2](https://arxiv.org/abs/2006.04558) model with [AISHELL-3](http://www.aishelltech.com/aishell_3). The trained model can be used in Voice Cloning Task, We refer to the model structure of  [Transfer Learning from Speaker Veriﬁcation to Multispeaker Text-To-Speech Synthesis](https://arxiv.org/pdf/1806.04558.pdf). The general steps are as follows:
 1. Speaker Encoder: We use Speaker Verification to train a speaker encoder. Datasets used in this task are different from those used in `FastSpeech2` because the transcriptions are not needed, we use more datasets, refer to  [ge2e](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/ge2e).
@@ -114,7 +113,7 @@ ref_audio
 ├── LJ015-0254.wav
 └── audio_self_test.mp3
 ```
-`./local/voice_cloning.sh` calls `${BIN_DIR}/voice_cloning.py`
+`./local/voice_cloning.sh` calls `${BIN_DIR}/../voice_cloning.py`
 ```bash
 CUDA_VISIBLE_DEVICES=${gpus} ./local/voice_cloning.sh ${conf_path} ${train_output_path} ${ckpt_name} ${ge2e_params_path} ${ref_audio_dir}

--- a/examples/aishell3/vc1/conf/default.yaml
+++ b/examples/aishell3/vc1/conf/default.yaml
@@ -16,8 +16,8 @@ fmax: 7600         # Maximum frequency of Mel basis.
 n_mels: 80         # The number of mel basis.
 # Only used for the model using pitch features (e.g. FastSpeech2)
-f0min: 80          # Maximum f0 for pitch extraction.
+f0min: 80          # Minimum f0 for pitch extraction.
-f0max: 400         # Minimum f0 for pitch extraction.
+f0max: 400         # Maximum f0 for pitch extraction.
 ###########################################################
@@ -64,14 +64,14 @@ model:
    pitch_predictor_dropout: 0.5               # dropout rate in pitch predictor
    pitch_embed_kernel_size: 1                 # kernel size of conv embedding layer for pitch
    pitch_embed_dropout: 0.0                   # dropout rate after conv embedding layer for pitch
-    stop_gradient_from_pitch_predictor: true   # whether to stop the gradient from pitch predictor to encoder
+    stop_gradient_from_pitch_predictor: True   # whether to stop the gradient from pitch predictor to encoder
    energy_predictor_layers: 2                 # number of conv layers in energy predictor
    energy_predictor_chans: 256                # number of channels of conv layers in energy predictor
    energy_predictor_kernel_size: 3            # kernel size of conv leyers in energy predictor
    energy_predictor_dropout: 0.5              # dropout rate in energy predictor
    energy_embed_kernel_size: 1                # kernel size of conv embedding layer for energy
    energy_embed_dropout: 0.0                  # dropout rate after conv embedding layer for energy
-    stop_gradient_from_energy_predictor: false # whether to stop the gradient from energy predictor to encoder
+    stop_gradient_from_energy_predictor: False # whether to stop the gradient from energy predictor to encoder
    spk_embed_dim: 256                         # speaker embedding dimension
    spk_embed_integration_type: concat         # speaker embedding integration type

--- a/examples/aishell3/vc1/local/voice_cloning.sh
+++ b/examples/aishell3/vc1/local/voice_cloning.sh
@@ -8,13 +8,15 @@ ref_audio_dir=$5
 FLAGS_allocator_strategy=naive_best_fit \
 FLAGS_fraction_of_gpu_memory_to_use=0.01 \
-python3 ${BIN_DIR}/voice_cloning.py \
+python3 ${BIN_DIR}/../voice_cloning.py \
-    --fastspeech2-config=${config_path} \
+    --am=fastspeech2_aishell3 \
-    --fastspeech2-checkpoint=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_config=${config_path} \
-    --fastspeech2-stat=dump/train/speech_stats.npy \
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
-    --pwg-config=pwg_aishell3_ckpt_0.5/default.yaml \
+    --am_stat=dump/train/speech_stats.npy \
-    --pwg-checkpoint=pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz \
+    --voc=pwgan_aishell3 \
-    --pwg-stat=pwg_aishell3_ckpt_0.5/feats_stats.npy \
+    --voc_config=pwg_aishell3_ckpt_0.5/default.yaml \
+    --voc_ckpt=pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz \
+    --voc_stat=pwg_aishell3_ckpt_0.5/feats_stats.npy \
    --ge2e_params_path=${ge2e_params_path} \
    --text="凯莫瑞安联合体的经济崩溃迫在眉睫。" \
    --input-dir=${ref_audio_dir} \

--- a/examples/aishell3/voc1/conf/default.yaml
+++ b/examples/aishell3/voc1/conf/default.yaml
@@ -33,7 +33,7 @@ generator_params:
    aux_context_window: 2 # Context window size for auxiliary feature.
                          # If set to 2, previous 2 and future 2 frames will be considered.
    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    use_weight_norm: true # Whether to use weight norm.
+    use_weight_norm: True # Whether to use weight norm.
                          # If set to true, it will be applied to all of the conv layers.
    upsample_scales: [4, 5, 3, 5]     # Upsampling scales. prod(upsample_scales) == n_shift
@@ -46,8 +46,8 @@ discriminator_params:
    kernel_size: 3        # Number of output channels.
    layers: 10            # Number of conv layers.
    conv_channels: 64     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
+    bias: True            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
+    use_weight_norm: True # Whether to use weight norm.
                          # If set to true, it will be applied to all of the conv layers.
    nonlinear_activation: "leakyrelu" # Nonlinear function after each conv.
    nonlinear_activation_params:      # Nonlinear function parameters

--- a/examples/ami/README.md
+++ b/examples/ami/README.md
+# Speaker Diarization on AMI corpus
+* sd0 - speaker diarization by AHC,SC base on x-vectors
--- a/examples/ami/sd0/.gitignore
+++ b/examples/ami/sd0/.gitignore
+results
\ No newline at end of file
--- a/examples/ami/sd0/README.md
+++ b/examples/ami/sd0/README.md
+# Speaker Diarization on AMI corpus
+## About the AMI corpus:
+"The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals synchronized to a common timeline. These include close-talking and far-field microphones, individual and room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings, the participants also have unsynchronized pens available to them that record what is written. The meetings were recorded in English using three different rooms with different acoustic properties, and include mostly non-native speakers." See [ami overview](http://groups.inf.ed.ac.uk/ami/corpus/overview.shtml) for more details.
+## About the example
+The script performs diarization using x-vectors(TDNN,ECAPA-TDNN) on the AMI mix-headset data. We demonstrate the use of different clustering methods: AHC, spectral.
+## How to Run
+Use the following command to run diarization on AMI corpus.
+`bash ./run.sh` 
+## Results (DER) coming soon! :)
--- a/examples/ami/sd0/local/ami_prepare.py
+++ b/examples/ami/sd0/local/ami_prepare.py
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Data preparation.
+Download: http://groups.inf.ed.ac.uk/ami/download/
+Prepares metadata files (JSON) from manual annotations "segments/" using RTTM format (Oracle VAD).
+Authors
+ * qingenz123@126.com (Qingen ZHAO) 2022
+"""
+import os
+import logging
+import argparse
+import xml.etree.ElementTree as et
+import glob
+import json
+from ami_splits import get_AMI_split
+from distutils.util import strtobool
+from dataio import (
+    load_pkl,
+    save_pkl, )
+logger = logging.getLogger(__name__)
+SAMPLERATE = 16000
+def prepare_ami(
+        data_folder,
+        manual_annot_folder,
+        save_folder,
+        ref_rttm_dir,
+        meta_data_dir,
+        split_type="full_corpus_asr",
+        skip_TNO=True,
+        mic_type="Mix-Headset",
+        vad_type="oracle",
+        max_subseg_dur=3.0,
+        overlap=1.5, ):
+    """
+    Prepares reference RTTM and JSON files for the AMI dataset.
+    Arguments
+    ---------
+    data_folder : str
+        Path to the folder where the original amicorpus is stored.
+    manual_annot_folder : str
+        Directory where the manual annotations are stored.
+    save_folder : str
+        The save directory in results.
+    ref_rttm_dir : str
+        Directory to store reference RTTM files.
+    meta_data_dir : str
+        Directory to store the meta data (json) files.
+    split_type : str
+        Standard dataset split. See ami_splits.py for more information.
+        Allowed split_type: "scenario_only", "full_corpus" or "full_corpus_asr"
+    skip_TNO: bool
+        Skips TNO meeting recordings if True.
+    mic_type : str
+        Type of microphone to be used.
+    vad_type : str
+        Type of VAD. Kept for future when VAD will be added.
+    max_subseg_dur : float
+        Duration in seconds of a subsegments to be prepared from larger segments.
+    overlap : float
+        Overlap duration in seconds between adjacent subsegments
+    Example
+    -------
+    >>> from dataset.ami.ami_prepare import prepare_ami
+    >>> data_folder = '/home/data/ami/amicorpus/'
+    >>> manual_annot_folder = '/home/data/ami/ami_public_manual/'
+    >>> save_folder = './results/
+    >>> split_type = 'full_corpus_asr'
+    >>> mic_type = 'Mix-Headset'
+    >>> prepare_ami(data_folder, manual_annot_folder, save_folder, split_type, mic_type)
+    """
+    # Meta files
+    meta_files = [
+        os.path.join(meta_data_dir, "ami_train." + mic_type + ".subsegs.json"),
+        os.path.join(meta_data_dir, "ami_dev." + mic_type + ".subsegs.json"),
+        os.path.join(meta_data_dir, "ami_eval." + mic_type + ".subsegs.json"),
+    ]
+    # Create configuration for easily skipping data_preparation stage
+    conf = {
+        "data_folder": data_folder,
+        "save_folder": save_folder,
+        "ref_rttm_dir": ref_rttm_dir,
+        "meta_data_dir": meta_data_dir,
+        "split_type": split_type,
+        "skip_TNO": skip_TNO,
+        "mic_type": mic_type,
+        "vad": vad_type,
+        "max_subseg_dur": max_subseg_dur,
+        "overlap": overlap,
+        "meta_files": meta_files,
+    }
+    if not os.path.exists(save_folder):
+        os.makedirs(save_folder)
+    # Setting output option files.
+    opt_file = "opt_ami_prepare." + mic_type + ".pkl"
+    # Check if this phase is already done (if so, skip it)
+    if skip(save_folder, conf, meta_files, opt_file):
+        logger.info(
+            "Skipping data preparation, as it was completed in previous run.")
+        return
+    msg = "\tCreating meta-data file for the AMI Dataset.."
+    logger.debug(msg)
+    # Get the split
+    train_set, dev_set, eval_set = get_AMI_split(split_type)
+    # Prepare RTTM from XML(manual annot) and store are groundtruth
+    # Create ref_RTTM directory
+    if not os.path.exists(ref_rttm_dir):
+        os.makedirs(ref_rttm_dir)
+    # Create reference RTTM files
+    splits = ["train", "dev", "eval"]
+    for i in splits:
+        rttm_file = ref_rttm_dir + "/fullref_ami_" + i + ".rttm"
+        if i == "train":
+            prepare_segs_for_RTTM(
+                train_set,
+                rttm_file,
+                data_folder,
+                manual_annot_folder,
+                i,
+                skip_TNO, )
+        if i == "dev":
+            prepare_segs_for_RTTM(
+                dev_set,
+                rttm_file,
+                data_folder,
+                manual_annot_folder,
+                i,
+                skip_TNO, )
+        if i == "eval":
+            prepare_segs_for_RTTM(
+                eval_set,
+                rttm_file,
+                data_folder,
+                manual_annot_folder,
+                i,
+                skip_TNO, )
+    # Create meta_files for splits
+    meta_data_dir = meta_data_dir
+    if not os.path.exists(meta_data_dir):
+        os.makedirs(meta_data_dir)
+    for i in splits:
+        rttm_file = ref_rttm_dir + "/fullref_ami_" + i + ".rttm"
+        meta_filename_prefix = "ami_" + i
+        prepare_metadata(
+            rttm_file,
+            meta_data_dir,
+            data_folder,
+            meta_filename_prefix,
+            max_subseg_dur,
+            overlap,
+            mic_type, )
+    save_opt_file = os.path.join(save_folder, opt_file)
+    save_pkl(conf, save_opt_file)
+def get_RTTM_per_rec(segs, spkrs_list, rec_id):
+    """Prepares rttm for each recording
+    """
+    rttm = []
+    # Prepare header
+    for spkr_id in spkrs_list:
+        # e.g. SPKR-INFO ES2008c 0 <NA> <NA> <NA> unknown ES2008c.A_PM <NA> <NA>
+        line = ("SPKR-INFO " + rec_id + " 0 <NA> <NA> <NA> unknown " + spkr_id +
+                " <NA> <NA>")
+        rttm.append(line)
+    # Append remaining lines
+    for row in segs:
+        # e.g. SPEAKER ES2008c 0 37.880 0.590 <NA> <NA> ES2008c.A_PM <NA> <NA>
+        if float(row[1]) < float(row[0]):
+            msg1 = (
+                "Possibly Incorrect Annotation Found!! transcriber_start (%s) > transcriber_end (%s)"
+                % (row[0], row[1]))
+            msg2 = (
+                "Excluding this incorrect row from the RTTM : %s, %s, %s, %s" %
+                (rec_id, row[0], str(round(float(row[1]) - float(row[0]), 4)),
+                 str(row[2]), ))
+            logger.info(msg1)
+            logger.info(msg2)
+            continue
+        line = ("SPEAKER " + rec_id + " 0 " + str(round(float(row[0]), 4)) + " "
+                + str(round(float(row[1]) - float(row[0]), 4)) + " <NA> <NA> " +
+                str(row[2]) + " <NA> <NA>")
+        rttm.append(line)
+    return rttm
+def prepare_segs_for_RTTM(list_ids, out_rttm_file, audio_dir, annot_dir,
+                          split_type, skip_TNO):
+    RTTM = []  # Stores all RTTMs clubbed together for a given dataset split
+    for main_meet_id in list_ids:
+        # Skip TNO meetings from dev and eval sets
+        if (main_meet_id.startswith("TS") and split_type != "train" and
+                skip_TNO is True):
+            msg = ("Skipping TNO meeting in AMI " + str(split_type) + " set : "
+                   + str(main_meet_id))
+            logger.info(msg)
+            continue
+        list_sessions = glob.glob(audio_dir + "/" + main_meet_id + "*")
+        list_sessions.sort()
+        for sess in list_sessions:
+            rec_id = os.path.basename(sess)
+            path = annot_dir + "/segments/" + rec_id
+            f = path + ".*.segments.xml"
+            list_spkr_xmls = glob.glob(f)
+            list_spkr_xmls.sort()  # A, B, C, D, E etc (Speakers)
+            segs = []
+            spkrs_list = (
+                [])  # Since non-scenario recordings contains 3-5 speakers
+            for spkr_xml_file in list_spkr_xmls:
+                # Speaker ID
+                spkr = os.path.basename(spkr_xml_file).split(".")[1]
+                spkr_ID = rec_id + "." + spkr
+                spkrs_list.append(spkr_ID)
+                # Parse xml tree
+                tree = et.parse(spkr_xml_file)
+                root = tree.getroot()
+                # Start, end and speaker_ID from xml file
+                segs = segs + [[
+                    elem.attrib["transcriber_start"],
+                    elem.attrib["transcriber_end"],
+                    spkr_ID,
+                ] for elem in root.iter("segment")]
+            # Sort rows as per the start time (per recording)
+            segs.sort(key=lambda x: float(x[0]))
+            rttm_per_rec = get_RTTM_per_rec(segs, spkrs_list, rec_id)
+            RTTM = RTTM + rttm_per_rec
+    # Write one RTTM as groundtruth. For example, "fullref_eval.rttm"
+    with open(out_rttm_file, "w") as f:
+        for item in RTTM:
+            f.write("%s\n" % item)
+def is_overlapped(end1, start2):
+    """Returns True if the two segments overlap
+    Arguments
+    ---------
+    end1 : float
+        End time of the first segment.
+    start2 : float
+        Start time of the second segment.
+    """
+    if start2 > end1:
+        return False
+    else:
+        return True
+def merge_rttm_intervals(rttm_segs):
+    """Merges adjacent segments in rttm if they overlap.
+    """
+    # For one recording
+    # rec_id = rttm_segs[0][1]
+    rttm_segs.sort(key=lambda x: float(x[3]))
+    # first_seg = rttm_segs[0] # first interval.. as it is
+    merged_segs = [rttm_segs[0]]
+    strt = float(rttm_segs[0][3])
+    end = float(rttm_segs[0][3]) + float(rttm_segs[0][4])
+    for row in rttm_segs[1:]:
+        s = float(row[3])
+        e = float(row[3]) + float(row[4])
+        if is_overlapped(end, s):
+            # Update only end. The strt will be same as in last segment
+            # Just update last row in the merged_segs
+            end = max(end, e)
+            merged_segs[-1][3] = str(round(strt, 4))
+            merged_segs[-1][4] = str(round((end - strt), 4))
+            merged_segs[-1][7] = "overlap"  # previous_row[7] + '-'+ row[7]
+        else:
+            # Add a new disjoint segment
+            strt = s
+            end = e
+            merged_segs.append(row)  # this will have 1 spkr ID
+    return merged_segs
+def get_subsegments(merged_segs, max_subseg_dur=3.0, overlap=1.5):
+    """Divides bigger segments into smaller sub-segments
+    """
+    shift = max_subseg_dur - overlap
+    subsegments = []
+    # These rows are in RTTM format
+    for row in merged_segs:
+        seg_dur = float(row[4])
+        rec_id = row[1]
+        if seg_dur > max_subseg_dur:
+            num_subsegs = int(seg_dur / shift)
+            # Taking 0.01 sec as small step
+            seg_start = float(row[3])
+            seg_end = seg_start + seg_dur
+            # Now divide this segment (new_row) in smaller subsegments
+            for i in range(num_subsegs):
+                subseg_start = seg_start + i * shift
+                subseg_end = min(subseg_start + max_subseg_dur - 0.01, seg_end)
+                subseg_dur = subseg_end - subseg_start
+                new_row = [
+                    "SPEAKER",
+                    rec_id,
+                    "0",
+                    str(round(float(subseg_start), 4)),
+                    str(round(float(subseg_dur), 4)),
+                    "<NA>",
+                    "<NA>",
+                    row[7],
+                    "<NA>",
+                    "<NA>",
+                ]
+                subsegments.append(new_row)
+                # Break if exceeding the boundary
+                if subseg_end >= seg_end:
+                    break
+        else:
+            subsegments.append(row)
+    return subsegments
+def prepare_metadata(rttm_file, save_dir, data_dir, filename, max_subseg_dur,
+                     overlap, mic_type):
+    # Read RTTM, get unique meeting_IDs (from RTTM headers)
+    # For each MeetingID. select that meetID -> merge -> subsegment -> json -> append
+    # Read RTTM
+    RTTM = []
+    with open(rttm_file, "r") as f:
+        for line in f:
+            entry = line[:-1]
+            RTTM.append(entry)
+    spkr_info = filter(lambda x: x.startswith("SPKR-INFO"), RTTM)
+    rec_ids = list(set([row.split(" ")[1] for row in spkr_info]))
+    rec_ids.sort()  # sorting just to make JSON look in proper sequence
+    # For each recording merge segments and then perform subsegmentation
+    MERGED_SEGMENTS = []
+    SUBSEGMENTS = []
+    for rec_id in rec_ids:
+        segs_iter = filter(lambda x: x.startswith("SPEAKER " + str(rec_id)),
+                           RTTM)
+        gt_rttm_segs = [row.split(" ") for row in segs_iter]
+        # Merge, subsegment and then convert to json format.
+        merged_segs = merge_rttm_intervals(
+            gt_rttm_segs)  # We lose speaker_ID after merging
+        MERGED_SEGMENTS = MERGED_SEGMENTS + merged_segs
+        # Divide segments into smaller sub-segments
+        subsegs = get_subsegments(merged_segs, max_subseg_dur, overlap)
+        SUBSEGMENTS = SUBSEGMENTS + subsegs
+    # Write segment AND sub-segments (in RTTM format)
+    segs_file = save_dir + "/" + filename + ".segments.rttm"
+    subsegment_file = save_dir + "/" + filename + ".subsegments.rttm"
+    with open(segs_file, "w") as f:
+        for row in MERGED_SEGMENTS:
+            line_str = " ".join(row)
+            f.write("%s\n" % line_str)
+    with open(subsegment_file, "w") as f:
+        for row in SUBSEGMENTS:
+            line_str = " ".join(row)
+            f.write("%s\n" % line_str)
+    # Create JSON from subsegments
+    json_dict = {}
+    for row in SUBSEGMENTS:
+        rec_id = row[1]
+        strt = str(round(float(row[3]), 4))
+        end = str(round((float(row[3]) + float(row[4])), 4))
+        subsegment_ID = rec_id + "_" + strt + "_" + end
+        dur = row[4]
+        start_sample = int(float(strt) * SAMPLERATE)
+        end_sample = int(float(end) * SAMPLERATE)
+        # If multi-mic audio is selected
+        if mic_type == "Array1":
+            wav_file_base_path = (data_dir + "/" + rec_id + "/audio/" + rec_id +
+                                  "." + mic_type + "-")
+            f = []  # adding all 8 mics
+            for i in range(8):
+                f.append(wav_file_base_path + str(i + 1).zfill(2) + ".wav")
+            audio_files_path_list = f
+            # Note: key "files" with 's' is used for multi-mic
+            json_dict[subsegment_ID] = {
+                "wav": {
+                    "files": audio_files_path_list,
+                    "duration": float(dur),
+                    "start": int(start_sample),
+                    "stop": int(end_sample),
+                },
+            }
+        else:
+            # Single mic audio
+            wav_file_path = (data_dir + "/" + rec_id + "/audio/" + rec_id + "."
+                             + mic_type + ".wav")
+            # Note: key "file" without 's' is used for single-mic
+            json_dict[subsegment_ID] = {
+                "wav": {
+                    "file": wav_file_path,
+                    "duration": float(dur),
+                    "start": int(start_sample),
+                    "stop": int(end_sample),
+                },
+            }
+    out_json_file = save_dir + "/" + filename + "." + mic_type + ".subsegs.json"
+    with open(out_json_file, mode="w") as json_f:
+        json.dump(json_dict, json_f, indent=2)
+    msg = "%s JSON prepared" % (out_json_file)
+    logger.debug(msg)
+def skip(save_folder, conf, meta_files, opt_file):
+    """
+    Detects if the AMI data_preparation has been already done.
+    If the preparation has been done, we can skip it.
+    Returns
+    -------
+    bool
+        if True, the preparation phase can be skipped.
+        if False, it must be done.
+    """
+    # Checking if meta (json) files are available
+    skip = True
+    for file_path in meta_files:
+        if not os.path.isfile(file_path):
+            skip = False
+    # Checking saved options
+    save_opt_file = os.path.join(save_folder, opt_file)
+    if skip is True:
+        if os.path.isfile(save_opt_file):
+            opts_old = load_pkl(save_opt_file)
+            if opts_old == conf:
+                skip = True
+            else:
+                skip = False
+        else:
+            skip = False
+    return skip
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(
+        prog='python ami_prepare.py  --data_folder /home/data/ami/amicorpus \
+            --manual_annot_folder /home/data/ami/ami_public_manual_1.6.2 \
+            --save_folder ./results/ --ref_rttm_dir ./results/ref_rttms \
+            --meta_data_dir ./results/metadata',
+        description='AMI Data preparation')
+    parser.add_argument(
+        '--data_folder',
+        required=True,
+        help='Path to the folder where the original amicorpus is stored')
+    parser.add_argument(
+        '--manual_annot_folder',
+        required=True,
+        help='Directory where the manual annotations are stored')
+    parser.add_argument(
+        '--save_folder', required=True, help='The save directory in results')
+    parser.add_argument(
+        '--ref_rttm_dir',
+        required=True,
+        help='Directory to store reference RTTM files')
+    parser.add_argument(
+        '--meta_data_dir',
+        required=True,
+        help='Directory to store the meta data (json) files')
+    parser.add_argument(
+        '--split_type',
+        default="full_corpus_asr",
+        help='Standard dataset split. See ami_splits.py for more information')
+    parser.add_argument(
+        '--skip_TNO',
+        default=True,
+        type=strtobool,
+        help='Skips TNO meeting recordings if True')
+    parser.add_argument(
+        '--mic_type',
+        default="Mix-Headset",
+        help='Type of microphone to be used')
+    parser.add_argument(
+        '--vad_type',
+        default="oracle",
+        help='Type of VAD. Kept for future when VAD will be added')
+    parser.add_argument(
+        '--max_subseg_dur',
+        default=3.0,
+        type=float,
+        help='Duration in seconds of a subsegments to be prepared from larger segments'
+    )
+    parser.add_argument(
+        '--overlap',
+        default=1.5,
+        type=float,
+        help='Overlap duration in seconds between adjacent subsegments')
+    args = parser.parse_args()
+    prepare_ami(args.data_folder, args.manual_annot_folder, args.save_folder,
+                args.ref_rttm_dir, args.meta_data_dir)
--- a/examples/ami/sd0/local/ami_splits.py
+++ b/examples/ami/sd0/local/ami_splits.py
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+AMI corpus contained 100 hours of meeting recording.
+This script returns the standard train, dev and eval split for AMI corpus.
+For more information on dataset please refer to http://groups.inf.ed.ac.uk/ami/corpus/datasets.shtml
+Authors
+ * qingenz123@126.com (Qingen ZHAO) 2022
+"""
+ALLOWED_OPTIONS = ["scenario_only", "full_corpus", "full_corpus_asr"]
+def get_AMI_split(split_option):
+    """
+    Prepares train, dev, and test sets for given split_option
+    Arguments
+    ---------
+    split_option: str
+        The standard split option.
+        Allowed options: "scenario_only", "full_corpus", "full_corpus_asr"
+    Returns
+    -------
+        Meeting IDs for train, dev, and test sets for given split_option
+    """
+    if split_option not in ALLOWED_OPTIONS:
+        print(
+            f'Invalid split "{split_option}" requested!\nValid split_options are: ',
+            ALLOWED_OPTIONS, )
+        return
+    if split_option == "scenario_only":
+        train_set = [
+            "ES2002",
+            "ES2005",
+            "ES2006",
+            "ES2007",
+            "ES2008",
+            "ES2009",
+            "ES2010",
+            "ES2012",
+            "ES2013",
+            "ES2015",
+            "ES2016",
+            "IS1000",
+            "IS1001",
+            "IS1002",
+            "IS1003",
+            "IS1004",
+            "IS1005",
+            "IS1006",
+            "IS1007",
+            "TS3005",
+            "TS3008",
+            "TS3009",
+            "TS3010",
+            "TS3011",
+            "TS3012",
+        ]
+        dev_set = [
+            "ES2003",
+            "ES2011",
+            "IS1008",
+            "TS3004",
+            "TS3006",
+        ]
+        test_set = [
+            "ES2004",
+            "ES2014",
+            "IS1009",
+            "TS3003",
+            "TS3007",
+        ]
+    if split_option == "full_corpus":
+        # List of train: SA (TRAINING PART OF SEEN DATA)
+        train_set = [
+            "ES2002",
+            "ES2005",
+            "ES2006",
+            "ES2007",
+            "ES2008",
+            "ES2009",
+            "ES2010",
+            "ES2012",
+            "ES2013",
+            "ES2015",
+            "ES2016",
+            "IS1000",
+            "IS1001",
+            "IS1002",
+            "IS1003",
+            "IS1004",
+            "IS1005",
+            "IS1006",
+            "IS1007",
+            "TS3005",
+            "TS3008",
+            "TS3009",
+            "TS3010",
+            "TS3011",
+            "TS3012",
+            "EN2001",
+            "EN2003",
+            "EN2004",
+            "EN2005",
+            "EN2006",
+            "EN2009",
+            "IN1001",
+            "IN1002",
+            "IN1005",
+            "IN1007",
+            "IN1008",
+            "IN1009",
+            "IN1012",
+            "IN1013",
+            "IN1014",
+            "IN1016",
+        ]
+        # List of dev: SB (DEV PART OF SEEN DATA)
+        dev_set = [
+            "ES2003",
+            "ES2011",
+            "IS1008",
+            "TS3004",
+            "TS3006",
+            "IB4001",
+            "IB4002",
+            "IB4003",
+            "IB4004",
+            "IB4010",
+            "IB4011",
+        ]
+        # List of test: SC (UNSEEN DATA FOR EVALUATION)
+        # Note that IB4005 does not appear because it has speakers in common with two sets of data.
+        test_set = [
+            "ES2004",
+            "ES2014",
+            "IS1009",
+            "TS3003",
+            "TS3007",
+            "EN2002",
+        ]
+    if split_option == "full_corpus_asr":
+        train_set = [
+            "ES2002",
+            "ES2003",
+            "ES2005",
+            "ES2006",
+            "ES2007",
+            "ES2008",
+            "ES2009",
+            "ES2010",
+            "ES2012",
+            "ES2013",
+            "ES2014",
+            "ES2015",
+            "ES2016",
+            "IS1000",
+            "IS1001",
+            "IS1002",
+            "IS1003",
+            "IS1004",
+            "IS1005",
+            "IS1006",
+            "IS1007",
+            "TS3005",
+            "TS3006",
+            "TS3007",
+            "TS3008",
+            "TS3009",
+            "TS3010",
+            "TS3011",
+            "TS3012",
+            "EN2001",
+            "EN2003",
+            "EN2004",
+            "EN2005",
+            "EN2006",
+            "EN2009",
+            "IN1001",
+            "IN1002",
+            "IN1005",
+            "IN1007",
+            "IN1008",
+            "IN1009",
+            "IN1012",
+            "IN1013",
+            "IN1014",
+            "IN1016",
+        ]
+        dev_set = [
+            "ES2011",
+            "IS1008",
+            "TS3004",
+            "IB4001",
+            "IB4002",
+            "IB4003",
+            "IB4004",
+            "IB4010",
+            "IB4011",
+        ]
+        test_set = [
+            "ES2004",
+            "IS1009",
+            "TS3003",
+            "EN2002",
+        ]
+    return train_set, dev_set, test_set
--- a/examples/ami/sd0/local/data.sh
+++ b/examples/ami/sd0/local/data.sh
+#!/bin/bash
+stage=1
+TARGET_DIR=${MAIN_ROOT}/dataset/ami
+data_folder=${TARGET_DIR}/amicorpus #e.g., /path/to/amicorpus/
+manual_annot_folder=${TARGET_DIR}/ami_public_manual_1.6.2 #e.g., /path/to/ami_public_manual_1.6.2/
+save_folder=${MAIN_ROOT}/examples/ami/sd0/data
+ref_rttm_dir=${save_folder}/ref_rttms
+meta_data_dir=${save_folder}/metadata
+set=L
+. ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
+set -u
+set -o pipefail
+mkdir -p ${save_folder}
+if [ ${stage} -le 0 ]; then
+    # Download AMI corpus, You need around 10GB of free space to get whole data
+    # The signals are too large to package in this way,
+    # so you need to use the chooser to indicate which ones you wish to download
+    echo "Please follow https://groups.inf.ed.ac.uk/ami/download/ to download the data."
+    echo "Annotations: AMI manual annotations v1.6.2 "
+    echo "Signals: "
+    echo "1) Select one or more AMI meetings: the IDs please follow ./ami_split.py"
+    echo "2) Select media streams: Just select Headset mix"
+    exit 0;
+fi
+if [ ${stage} -le 1 ]; then
+    echo "AMI Data preparation"
+    python local/ami_prepare.py  --data_folder ${data_folder} \
+            --manual_annot_folder ${manual_annot_folder} \
+            --save_folder ${save_folder} --ref_rttm_dir ${ref_rttm_dir} \
+            --meta_data_dir ${meta_data_dir} 
+    if [ $? -ne 0 ]; then
+        echo "Prepare AMI failed. Please check log message."
+        exit 1
+    fi
+fi
+echo "AMI data preparation done."
+exit 0
--- a/examples/ami/sd0/local/dataio.py
+++ b/examples/ami/sd0/local/dataio.py
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Data reading and writing.
+Authors
+ * qingenz123@126.com (Qingen ZHAO) 2022
+"""
+import os
+import pickle
+def save_pkl(obj, file):
+    """Save an object in pkl format.
+    Arguments
+    ---------
+    obj : object
+        Object to save in pkl format
+    file : str
+        Path to the output file
+    sampling_rate : int
+        Sampling rate of the audio file, TODO: this is not used?
+    Example
+    -------
+    >>> tmpfile = os.path.join(getfixture('tmpdir'), "example.pkl")
+    >>> save_pkl([1, 2, 3, 4, 5], tmpfile)
+    >>> load_pkl(tmpfile)
+    [1, 2, 3, 4, 5]
+    """
+    with open(file, "wb") as f:
+        pickle.dump(obj, f)
+def load_pickle(pickle_path):
+    """Utility function for loading .pkl pickle files.
+    Arguments
+    ---------
+    pickle_path : str
+        Path to pickle file.
+    Returns
+    -------
+    out : object
+        Python object loaded from pickle.
+    """
+    with open(pickle_path, "rb") as f:
+        out = pickle.load(f)
+    return out
+def load_pkl(file):
+    """Loads a pkl file.
+    For an example, see `save_pkl`.
+    Arguments
+    ---------
+    file : str
+        Path to the input pkl file.
+    Returns
+    -------
+    The loaded object.
+    """
+    # Deals with the situation where two processes are trying
+    # to access the same label dictionary by creating a lock
+    count = 100
+    while count > 0:
+        if os.path.isfile(file + ".lock"):
+            time.sleep(1)
+            count -= 1
+        else:
+            break
+    try:
+        open(file + ".lock", "w").close()
+        with open(file, "rb") as f:
+            return pickle.load(f)
+    finally:
+        if os.path.isfile(file + ".lock"):
+            os.remove(file + ".lock")
--- a/examples/ami/sd0/path.sh
+++ b/examples/ami/sd0/path.sh
+export MAIN_ROOT=`realpath ${PWD}/../../../`
+export PATH=${MAIN_ROOT}:${MAIN_ROOT}/utils:${PATH}
+export LC_ALL=C
+export PYTHONDONTWRITEBYTECODE=1
+# Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
+export PYTHONIOENCODING=UTF-8
+export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+# model exp
+#MODEL=ECAPA_TDNN
+#export BIN_DIR=${MAIN_ROOT}/paddlespeech/vector/exps/${MODEL}/bin
--- a/examples/ami/sd0/run.sh
+++ b/examples/ami/sd0/run.sh
+#!/bin/bash
+. path.sh || exit 1;
+set -e
+stage=1
+. ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
+if [ ${stage} -le 1 ]; then
+    # prepare data
+    bash ./local/data.sh || exit -1
+fi
\ No newline at end of file
--- a/examples/ami/sd0/utils
+++ b/examples/ami/sd0/utils
+../../../utils
\ No newline at end of file
--- a/examples/callcenter/README.md
+++ b/examples/callcenter/README.md
+# Callcenter 8k sample rate
+Data distribution:
+```
+676048 utts
+491.4004722221223 h
+4357792.0 text
+2.4633630739178654 text/sec
+2.6167397877068495 sec/utt
+```
+train/dev/test partition:
+```
+    33802 manifest.dev
+    67606 manifest.test
+   574640 manifest.train
+   676048 total
+```
--- a/examples/csmsc/README.md
+++ b/examples/csmsc/README.md
@@ -10,3 +10,5 @@
 * voc2 - MelGAN
 * voc3 - MultiBand MelGAN
 * voc4 - Style MelGAN
+* voc5 - HiFiGAN
+* voc6 - WaveRNN
--- a/examples/csmsc/tts0/README.md
+++ b/examples/csmsc/tts0/README.md
+# Tacotron2 with CSMSC
+This example contains code used to train a [Tacotron2](https://arxiv.org/abs/1712.05884) model with [Chinese Standard Mandarin Speech Copus](https://www.data-baker.com/open_source.html).
+## Dataset
+### Download and Extract
+Download CSMSC from it's [Official Website](https://test.data-baker.com/data/index/source).
+### Get MFA Result and Extract
+We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) to get phonemes for Tacotron2, the durations of MFA are not needed here.
+You can download from here [baker_alignment_tone.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/BZNSYP/with_tone/baker_alignment_tone.tar.gz), or train your MFA model reference to [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) of our repo.
+## Get Started
+Assume the path to the dataset is `~/datasets/BZNSYP`.
+Assume the path to the MFA result of CSMSC is `./baker_alignment_tone`.
+Run the command below to
+1. **source path**.
+2. preprocess the dataset.
+3. train the model.
+4. synthesize wavs.
+    - synthesize waveform from `metadata.jsonl`.
+    - synthesize waveform from a text file.
+```bash
+./run.sh
+```
+You can choose a range of stages you want to run, or set `stage` equal to `stop-stage` to use only one stage, for example, running the following command will only preprocess the dataset.
+```bash
+./run.sh --stage 0 --stop-stage 0
+```
+### Data Preprocessing
+```bash
+./local/preprocess.sh ${conf_path}
+```
+When it is done. A `dump` folder is created in the current directory. The structure of the dump folder is listed below.
+```text
+dump
+├── dev
+│   ├── norm
+│   └── raw
+├── phone_id_map.txt
+├── speaker_id_map.txt
+├── test
+│   ├── norm
+│   └── raw
+└── train
+    ├── norm
+    ├── raw
+    └── speech_stats.npy
+```
+The dataset is split into 3 parts, namely `train`, `dev`, and` test`, each of which contains a `norm` and `raw` subfolder. The raw folder contains speech features of each utterance, while the norm folder contains normalized ones. The statistics used to normalize features are computed from the training set, which is located in `dump/train/*_stats.npy`.
+Also, there is a `metadata.jsonl` in each subfolder. It is a table-like file that contains phones, text_lengths, speech_lengths, durations, the path of speech features, speaker, and the id of each utterance.
+### Model Training
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
+```
+`./local/train.sh` calls `${BIN_DIR}/train.py`.
+Here's the complete help message.
+```text
+usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
+                [--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
+                [--ngpu NGPU] [--phones-dict PHONES_DICT]
+Train a Tacotron2 model.
+optional arguments:
+  -h, --help            show this help message and exit
+  --config CONFIG       tacotron2 config file.
+  --train-metadata TRAIN_METADATA
+                        training data.
+  --dev-metadata DEV_METADATA
+                        dev data.
+  --output-dir OUTPUT_DIR
+                        output dir.
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --phones-dict PHONES_DICT
+                        phone vocabulary file.
+```
+1. `--config` is a config file in yaml format to overwrite the default config, which can be found at `conf/default.yaml`.
+2. `--train-metadata` and `--dev-metadata` should be the metadata file in the normalized subfolder of `train` and `dev` in the `dump` folder.
+3. `--output-dir` is the directory to save the results of the experiment. Checkpoints are saved in `checkpoints/` inside this directory.
+4. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
+5. `--phones-dict` is the path of the phone vocabulary file.
+### Synthesizing
+We use [parallel wavegan](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc1) as the neural vocoder.
+Download pretrained parallel wavegan model from [pwg_baker_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_ckpt_0.4.zip) and unzip it.
+```bash
+unzip pwg_baker_ckpt_0.4.zip
+```
+Parallel WaveGAN checkpoint contains files listed below.
+```text
+pwg_baker_ckpt_0.4
+├── pwg_default.yaml               # default config used to train parallel wavegan
+├── pwg_snapshot_iter_400000.pdz   # model parameters of parallel wavegan
+└── pwg_stats.npy                  # statistics used to normalize spectrogram when training parallel wavegan
+```
+`./local/synthesize.sh` calls `${BIN_DIR}/../synthesize.py`, which can synthesize waveform from `metadata.jsonl`.
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize.py [-h]
+                     [--am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}]
+                     [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                     [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                     [--tones_dict TONES_DICT] [--speaker_dict SPEAKER_DICT]
+                     [--voice-cloning VOICE_CLONING]
+                     [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}]
+                     [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                     [--voc_stat VOC_STAT] [--ngpu NGPU]
+                     [--test_metadata TEST_METADATA] [--output_dir OUTPUT_DIR]
+Synthesize with acoustic model & vocoder
+optional arguments:
+  -h, --help            show this help message and exit
+  --am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}
+                        Choose acoustic model type of tts task.
+  --am_config AM_CONFIG
+                        Config of acoustic model. Use deault config when it is
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --voice-cloning VOICE_CLONING
+                        whether training voice cloning model.
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --test_metadata TEST_METADATA
+                        test metadata.
+  --output_dir OUTPUT_DIR
+                        output dir.
+```
+`./local/synthesize_e2e.sh` calls `${BIN_DIR}/../synthesize_e2e.py`, which can synthesize waveform from text file.
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize_e2e.py [-h]
+                         [--am {speedyspeech_csmsc,speedyspeech_aishell3,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}]
+                         [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                         [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                         [--tones_dict TONES_DICT]
+                         [--speaker_dict SPEAKER_DICT] [--spk_id SPK_ID]
+                         [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc,style_melgan_csmsc,hifigan_csmsc}]
+                         [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                         [--voc_stat VOC_STAT] [--lang LANG]
+                         [--inference_dir INFERENCE_DIR] [--ngpu NGPU]
+                         [--text TEXT] [--output_dir OUTPUT_DIR]
+Synthesize with acoustic model & vocoder
+optional arguments:
+  -h, --help            show this help message and exit
+  --am {speedyspeech_csmsc,speedyspeech_aishell3,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}
+                        Choose acoustic model type of tts task.
+  --am_config AM_CONFIG
+                        Config of acoustic model. Use deault config when it is
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --spk_id SPK_ID       spk id for multi speaker acoustic model
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc,style_melgan_csmsc,hifigan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
+  --lang LANG           Choose model language. zh or en
+  --inference_dir INFERENCE_DIR
+                        dir to save inference models
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --text TEXT           text to synthesize, a 'utt_id sentence' pair per line.
+  --output_dir OUTPUT_DIR
+                        output dir.
+```
+1. `--am` is acoustic model type with the format {model_name}_{dataset}
+2. `--am_config`, `--am_checkpoint`, `--am_stat` and `--phones_dict` are arguments for acoustic model, which correspond to the 4 files in the Tacotron2 pretrained model.
+3. `--voc` is vocoder type with the format {model_name}_{dataset}
+4. `--voc_config`, `--voc_checkpoint`, `--voc_stat` are arguments for vocoder, which correspond to the 3 files in the parallel wavegan pretrained model.
+5. `--lang` is the model language, which can be `zh` or `en`.
+6. `--test_metadata` should be the metadata file in the normalized subfolder of `test`  in the `dump` folder.
+7. `--text` is the text file, which contains sentences to synthesize.
+8. `--output_dir` is the directory to save synthesized audio files.
+9. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
+## Pretrained Model
+Pretrained Tacotron2 model with no silence in the edge of audios:
+- [tacotron2_csmsc_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_csmsc_ckpt_0.2.0.zip)
+The static model can be downloaded here [tacotron2_csmsc_static_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_csmsc_static_0.2.0.zip).
+Model | Step | eval/loss | eval/l1_loss | eval/mse_loss | eval/bce_loss| eval/attn_loss 
+:-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:
+default| 1(gpu) x 30600|0.57185|0.39614|0.14642|0.029|5.8e-05|
+Tacotron2 checkpoint contains files listed below.
+```text
+tacotron2_csmsc_ckpt_0.2.0
+├── default.yaml            # default config used to train Tacotron2
+├── phone_id_map.txt        # phone vocabulary file when training Tacotron2
+├── snapshot_iter_30600.pdz # model parameters and optimizer states
+└── speech_stats.npy        # statistics used to normalize spectrogram when training Tacotron2
+```
+You can use the following scripts to synthesize for `${BIN_DIR}/../sentences.txt` using pretrained Tacotron2 and parallel wavegan models.
+```bash
+source path.sh
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize_e2e.py \
+  --am=tacotron2_csmsc \
+  --am_config=tacotron2_csmsc_ckpt_0.2.0/default.yaml \
+  --am_ckpt=tacotron2_csmsc_ckpt_0.2.0/snapshot_iter_30600.pdz \
+  --am_stat=tacotron2_csmsc_ckpt_0.2.0/speech_stats.npy  \
+  --voc=pwgan_csmsc \
+  --voc_config=pwg_baker_ckpt_0.4/pwg_default.yaml \
+  --voc_ckpt=pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz \
+  --voc_stat=pwg_baker_ckpt_0.4/pwg_stats.npy \
+  --lang=zh \
+  --text=${BIN_DIR}/../sentences.txt \
+  --output_dir=exp/default/test_e2e \
+  --inference_dir=exp/default/inference \
+  --phones_dict=tacotron2_csmsc_ckpt_0.2.0/phone_id_map.txt
+```
--- a/examples/csmsc/tts0/conf/default.yaml
+++ b/examples/csmsc/tts0/conf/default.yaml
+# This configuration is for Paddle to train Tacotron 2. Compared to the
+# original paper, this configuration additionally use the guided attention
+# loss to accelerate the learning of the diagonal attention. It requires
+# only a single GPU with 12 GB memory and it takes ~1 days to finish the
+# training on Titan V.
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+fs: 24000          # sr
+n_fft: 2048        # FFT size (samples).
+n_shift: 300       # Hop size (samples). 12.5ms
+win_length: 1200   # Window length (samples). 50ms
+                   # If set to null, it will be the same as fft_size.
+window: "hann"     # Window function.
+# Only used for feats_type != raw
+fmin: 80           # Minimum frequency of Mel basis.
+fmax: 7600         # Maximum frequency of Mel basis.
+n_mels: 80         # The number of mel basis.
+###########################################################
+#                       DATA SETTING                      #
+###########################################################
+batch_size: 64
+num_workers: 2
+###########################################################
+#                       MODEL SETTING                     #
+###########################################################
+model:                          # keyword arguments for the selected model
+    embed_dim: 512               # char or phn embedding dimension
+    elayers: 1                   # number of blstm layers in encoder
+    eunits: 512                  # number of blstm units
+    econv_layers: 3              # number of convolutional layers in encoder
+    econv_chans: 512             # number of channels in convolutional layer
+    econv_filts: 5               # filter size of convolutional layer
+    atype: location              # attention function type
+    adim: 512                    # attention dimension
+    aconv_chans: 32              # number of channels in convolutional layer of attention
+    aconv_filts: 15              # filter size of convolutional layer of attention
+    cumulate_att_w: True         # whether to cumulate attention weight
+    dlayers: 2                   # number of lstm layers in decoder
+    dunits: 1024                 # number of lstm units in decoder
+    prenet_layers: 2             # number of layers in prenet
+    prenet_units: 256            # number of units in prenet
+    postnet_layers: 5            # number of layers in postnet
+    postnet_chans: 512           # number of channels in postnet
+    postnet_filts: 5             # filter size of postnet layer
+    output_activation: null      # activation function for the final output
+    use_batch_norm: True         # whether to use batch normalization in encoder
+    use_concate: True            # whether to concatenate encoder embedding with decoder outputs
+    use_residual: False          # whether to use residual connection in encoder
+    dropout_rate: 0.5            # dropout rate
+    zoneout_rate: 0.1            # zoneout rate
+    reduction_factor: 1          # reduction factor
+    spk_embed_dim: null          # speaker embedding dimension
+###########################################################
+#                       UPDATER SETTING                   #
+###########################################################
+updater:
+    use_masking: True            # whether to apply masking for padded part in loss calculation
+    bce_pos_weight: 5.0          # weight of positive sample in binary cross entropy calculation
+    use_guided_attn_loss: True   # whether to use guided attention loss
+    guided_attn_loss_sigma: 0.4  # sigma of guided attention loss
+    guided_attn_loss_lambda: 1.0 # strength of guided attention loss
+##########################################################
+#                  OPTIMIZER SETTING                     #
+##########################################################
+optimizer:
+    optim: adam              # optimizer type
+    learning_rate: 1.0e-03   # learning rate
+    epsilon: 1.0e-06         # epsilon
+    weight_decay: 0.0        # weight decay coefficient
+###########################################################
+#                     TRAINING SETTING                    #
+###########################################################
+max_epoch: 200
+num_snapshots: 5
+###########################################################
+#                       OTHER SETTING                     #
+###########################################################
+seed: 42
\ No newline at end of file
--- a/examples/csmsc/tts0/local/inference.sh
+++ b/examples/csmsc/tts0/local/inference.sh
+#!/bin/bash
+train_output_path=$1
+stage=0
+stop_stage=0
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=tacotron2_csmsc \
+        --voc=pwgan_csmsc \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt
+fi
+# for more GAN Vocoders
+# multi band melgan
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=tacotron2_csmsc \
+        --voc=mb_melgan_csmsc \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt
+fi
+# style melgan
+# style melgan's Dygraph to Static Graph is not ready now
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=tacotron2_csmsc \
+        --voc=style_melgan_csmsc \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt
+fi
+# hifigan
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=tacotron2_csmsc \
+        --voc=hifigan_csmsc \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt
+fi
\ No newline at end of file
--- a/examples/csmsc/tts0/local/preprocess.sh
+++ b/examples/csmsc/tts0/local/preprocess.sh
+#!/bin/bash
+stage=0
+stop_stage=100
+config_path=$1
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # get durations from MFA's result
+    echo "Generate durations.txt from MFA results ..."
+    python3 ${MAIN_ROOT}/utils/gen_duration_from_textgrid.py \
+        --inputdir=./baker_alignment_tone \
+        --output=durations.txt \
+        --config=${config_path}
+fi
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # extract features
+    echo "Extract features ..."
+    python3 ${BIN_DIR}/preprocess.py \
+        --dataset=baker \
+        --rootdir=~/datasets/BZNSYP/ \
+        --dumpdir=dump \
+        --dur-file=durations.txt \
+        --config=${config_path} \
+        --num-cpu=20 \
+        --cut-sil=True
+fi
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # get features' stats(mean and std)
+    echo "Get features' stats ..."
+    python3 ${MAIN_ROOT}/utils/compute_statistics.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --field-name="speech"
+fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # normalize and covert phone to id, dev and test should use train's stats
+    echo "Normalize ..."
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --dumpdir=dump/train/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/dev/raw/metadata.jsonl \
+        --dumpdir=dump/dev/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/test/raw/metadata.jsonl \
+        --dumpdir=dump/test/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+fi
--- a/examples/csmsc/tts0/local/synthesize.sh
+++ b/examples/csmsc/tts0/local/synthesize.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+ckpt_name=$3
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize.py \
+    --am=tacotron2_csmsc \
+    --am_config=${config_path} \
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_stat=dump/train/speech_stats.npy \
+    --voc=pwgan_csmsc \
+    --voc_config=pwg_baker_ckpt_0.4/pwg_default.yaml \
+    --voc_ckpt=pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz \
+    --voc_stat=pwg_baker_ckpt_0.4/pwg_stats.npy \
+    --test_metadata=dump/test/norm/metadata.jsonl \
+    --output_dir=${train_output_path}/test \
+    --phones_dict=dump/phone_id_map.txt
--- a/examples/csmsc/tts0/local/synthesize_e2e.sh
+++ b/examples/csmsc/tts0/local/synthesize_e2e.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+ckpt_name=$3
+stage=0
+stop_stage=0
+# TODO: tacotron2 动转静的结果没有静态图的响亮, 可能还是 decode 的时候某个函数动静不对齐
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=pwgan_csmsc \
+        --voc_config=pwg_baker_ckpt_0.4/pwg_default.yaml \
+        --voc_ckpt=pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz \
+        --voc_stat=pwg_baker_ckpt_0.4/pwg_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --phones_dict=dump/phone_id_map.txt \
+        --inference_dir=${train_output_path}/inference
+fi
+# for more GAN Vocoders
+# multi band melgan
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=mb_melgan_csmsc \
+        --voc_config=mb_melgan_baker_finetune_ckpt_0.5/finetune.yaml \
+        --voc_ckpt=mb_melgan_baker_finetune_ckpt_0.5/snapshot_iter_2000000.pdz\
+        --voc_stat=mb_melgan_baker_finetune_ckpt_0.5/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --inference_dir=${train_output_path}/inference \
+        --phones_dict=dump/phone_id_map.txt
+fi
+# the pretrained models haven't release now
+# style melgan
+# style melgan's Dygraph to Static Graph is not ready now
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=style_melgan_csmsc \
+        --voc_config=style_melgan_csmsc_ckpt_0.1.1/default.yaml \
+        --voc_ckpt=style_melgan_csmsc_ckpt_0.1.1/snapshot_iter_1500000.pdz \
+        --voc_stat=style_melgan_csmsc_ckpt_0.1.1/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --phones_dict=dump/phone_id_map.txt
+        # --inference_dir=${train_output_path}/inference
+fi
+# hifigan
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    echo "in hifigan syn_e2e"
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=hifigan_csmsc \
+        --voc_config=hifigan_csmsc_ckpt_0.1.1/default.yaml \
+        --voc_ckpt=hifigan_csmsc_ckpt_0.1.1/snapshot_iter_2500000.pdz \
+        --voc_stat=hifigan_csmsc_ckpt_0.1.1/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --inference_dir=${train_output_path}/inference \
+        --phones_dict=dump/phone_id_map.txt
+fi
+# wavernn
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "in wavernn syn_e2e"
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=wavernn_csmsc \
+        --voc_config=wavernn_csmsc_ckpt_0.2.0/default.yaml \
+        --voc_ckpt=wavernn_csmsc_ckpt_0.2.0/snapshot_iter_400000.pdz \
+        --voc_stat=wavernn_csmsc_ckpt_0.2.0/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --phones_dict=dump/phone_id_map.txt \
+        --inference_dir=${train_output_path}/inference
+fi
\ No newline at end of file
--- a/examples/csmsc/tts0/local/train.sh
+++ b/examples/csmsc/tts0/local/train.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+python3 ${BIN_DIR}/train.py \
+    --train-metadata=dump/train/norm/metadata.jsonl \
+    --dev-metadata=dump/dev/norm/metadata.jsonl \
+    --config=${config_path} \
+    --output-dir=${train_output_path} \
+    --ngpu=1 \
+    --phones-dict=dump/phone_id_map.txt
\ No newline at end of file
--- a/examples/csmsc/tts0/path.sh
+++ b/examples/csmsc/tts0/path.sh
+#!/bin/bash
+export MAIN_ROOT=`realpath ${PWD}/../../../`
+export PATH=${MAIN_ROOT}:${MAIN_ROOT}/utils:${PATH}
+export LC_ALL=C
+export PYTHONDONTWRITEBYTECODE=1
+# Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
+export PYTHONIOENCODING=UTF-8
+export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
+MODEL=tacotron2
+export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
--- a/examples/csmsc/tts0/run.sh
+++ b/examples/csmsc/tts0/run.sh
+#!/bin/bash
+set -e
+source path.sh
+gpus=0,1
+stage=0
+stop_stage=100
+conf_path=conf/default.yaml
+train_output_path=exp/default
+ckpt_name=snapshot_iter_153.pdz
+# with the following command, you can choose the stage range you want to run
+# such as `./run.sh --stage 0 --stop-stage 0`
+# this can not be mixed use with `$1`, `$2` ...
+source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # prepare data
+    ./local/preprocess.sh ${conf_path} || exit -1
+fi
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # train model, all `ckpt` under `train_output_path/checkpoints/` dir
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1
+fi
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # synthesize, vocoder is pwgan
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
+fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # synthesize_e2e, vocoder is pwgan
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
+fi
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    # inference with static model
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${train_output_path} || exit -1
+fi
--- a/examples/csmsc/tts2/local/synthesize_e2e.sh
+++ b/examples/csmsc/tts2/local/synthesize_e2e.sh
@@ -92,3 +92,26 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
        --phones_dict=dump/phone_id_map.txt \
        --tones_dict=dump/tone_id_map.txt
 fi
+# wavernn
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "in wavernn syn_e2e"
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=speedyspeech_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/feats_stats.npy \
+        --voc=wavernn_csmsc \
+        --voc_config=wavernn_csmsc_ckpt_0.2.0/default.yaml \
+        --voc_ckpt=wavernn_csmsc_ckpt_0.2.0/snapshot_iter_400000.pdz \
+        --voc_stat=wavernn_csmsc_ckpt_0.2.0/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --phones_dict=dump/phone_id_map.txt \
+        --tones_dict=dump/tone_id_map.txt \
+        --inference_dir=${train_output_path}/inference
+fi
--- a/examples/csmsc/tts3/README.md
+++ b/examples/csmsc/tts3/README.md
+([简体中文](./README_cn.md)|English)
 # FastSpeech2 with CSMSC
 This example contains code used to train a [Fastspeech2](https://arxiv.org/abs/2006.04558) model with [Chinese Standard Mandarin Speech Copus](https://www.data-baker.com/open_source.html).
@@ -242,6 +243,8 @@ fastspeech2_nosil_baker_ckpt_0.4
 └── speech_stats.npy        # statistics used to normalize spectrogram when training fastspeech2
 ```
 You can use the following scripts to synthesize for `${BIN_DIR}/../sentences.txt` using pretrained fastspeech2 and parallel wavegan models.
+If you want to use fastspeech2_conformer, you must delete this line `--inference_dir=exp/default/inference \` to skip the step of dygraph to static graph, cause we haven't tested dygraph to static graph for fastspeech2_conformer till now.
 ```bash
 source path.sh

--- a/examples/csmsc/tts3/README_cn.md
+++ b/examples/csmsc/tts3/README_cn.md
+(简体中文|[English](./README.md))
+# 用 CSMSC 数据集训练 FastSpeech2 模型
+本用例包含用于训练 [Fastspeech2](https://arxiv.org/abs/2006.04558) 模型的代码，使用 [Chinese Standard Mandarin Speech Copus](https://www.data-baker.com/open_source.html) 数据集。
+## 数据集
+### 下载并解压
+从 [官方网站](https://test.data-baker.com/data/index/source) 下载数据集
+### 获取MFA结果并解压
+我们使用 [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) 去获得 fastspeech2 的音素持续时间。
+你们可以从这里下载 [baker_alignment_tone.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/BZNSYP/with_tone/baker_alignment_tone.tar.gz), 或参考 [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) 训练你自己的模型。
+## 开始
+假设数据集的路径是 `~/datasets/BZNSYP`.
+假设CSMSC的MFA结果路径为 `./baker_alignment_tone`.
+运行下面的命令会进行如下操作：
+1. **设置原路径**。
+2. 对数据集进行预处理。
+3. 训练模型
+4. 合成波形
+    - 从 `metadata.jsonl` 合成波形。
+    - 从文本文件合成波形。
+5. 使用静态模型进行推理。
+```bash
+./run.sh
+```
+您可以选择要运行的一系列阶段，或者将 `stage` 设置为 `stop-stage` 以仅使用一个阶段，例如，运行以下命令只会预处理数据集。
+```bash
+./run.sh --stage 0 --stop-stage 0
+```
+### 数据预处理
+```bash
+./local/preprocess.sh ${conf_path}
+```
+当它完成时。将在当前目录中创建 `dump` 文件夹。转储文件夹的结构如下所示。
+```text
+dump
+├── dev
+│   ├── norm
+│   └── raw
+├── phone_id_map.txt
+├── speaker_id_map.txt
+├── test
+│   ├── norm
+│   └── raw
+└── train
+    ├── energy_stats.npy
+    ├── norm
+    ├── pitch_stats.npy
+    ├── raw
+    └── speech_stats.npy
+```
+数据集分为三个部分，即 `train` 、 `dev` 和 `test` ，每个部分都包含一个 `norm` 和 `raw` 子文件夹。原始文件夹包含每个话语的语音、音调和能量特征，而 `norm` 文件夹包含规范化的特征。用于规范化特征的统计数据是从 `dump/train/*_stats.npy` 中的训练集计算出来的。
+此外，还有一个 `metadata.jsonl` 在每个子文件夹中。它是一个类似表格的文件，包含音素、文本长度、语音长度、持续时间、语音特征路径、音调特征路径、能量特征路径、说话人和每个话语的 id。
+### 模型训练
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
+```
+`./local/train.sh` 调用 `${BIN_DIR}/train.py` 。
+以下是完整的帮助信息。
+```text
+usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
+                [--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
+                [--ngpu NGPU] [--phones-dict PHONES_DICT]
+                [--speaker-dict SPEAKER_DICT] [--voice-cloning VOICE_CLONING]
+Train a FastSpeech2 model.
+optional arguments:
+  -h, --help            show this help message and exit
+  --config CONFIG       fastspeech2 config file.
+  --train-metadata TRAIN_METADATA
+                        training data.
+  --dev-metadata DEV_METADATA
+                        dev data.
+  --output-dir OUTPUT_DIR
+                        output dir.
+  --ngpu NGPU           if ngpu=0, use cpu.
+  --phones-dict PHONES_DICT
+                        phone vocabulary file.
+  --speaker-dict SPEAKER_DICT
+                        speaker id map file for multiple speaker model.
+  --voice-cloning VOICE_CLONING
+                        whether training voice cloning model.
+```
+1. `--config` 是一个 yaml 格式的配置文件，用于覆盖默认配置，位于 `conf/default.yaml`.
+2. `--train-metadata` 和 `--dev-metadata` 应为 `dump` 文件夹中 `train` 和 `dev` 下的规范化元数据文件
+3. `--output-dir` 是保存结果的目录。 检查点保存在此目录中的 `checkpoints/` 目录下。
+4. `--ngpu` 要使用的 GPU 数，如果 ngpu==0，则使用 cpu 。
+5. `--phones-dict` 是音素词汇表文件的路径。
+### 合成
+我们使用 [parallel wavegan](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/csmsc/voc1) 作为神经声码器（vocoder）。
+从 [pwg_baker_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_ckpt_0.4.zip) 下载预训练的 parallel wavegan 模型并将其解压。
+```bash
+unzip pwg_baker_ckpt_0.4.zip
+```
+Parallel WaveGAN 检查点包含如下文件。
+```text
+pwg_baker_ckpt_0.4
+├── pwg_default.yaml               # 用于训练 parallel wavegan 的默认配置
+├── pwg_snapshot_iter_400000.pdz   # parallel wavegan 的模型参数
+└── pwg_stats.npy                  # 训练平行波形时用于规范化谱图的统计数据
+```
+`./local/synthesize.sh` 调用 `${BIN_DIR}/../synthesize.py` 即可从 `metadata.jsonl`中合成波形。
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize.py [-h]
+                     [--am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk}]
+                     [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                     [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                     [--tones_dict TONES_DICT] [--speaker_dict SPEAKER_DICT]
+                     [--voice-cloning VOICE_CLONING]
+                     [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}]
+                     [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                     [--voc_stat VOC_STAT] [--ngpu NGPU]
+                     [--test_metadata TEST_METADATA] [--output_dir OUTPUT_DIR]
+Synthesize with acoustic model & vocoder
+optional arguments:
+  -h, --help            show this help message and exit
+  --am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk}
+                        Choose acoustic model type of tts task.
+  --am_config AM_CONFIG
+                        Config of acoustic model. Use deault config when it is
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --voice-cloning VOICE_CLONING
+                        whether training voice cloning model.
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --test_metadata TEST_METADATA
+                        test metadata.
+  --output_dir OUTPUT_DIR
+                        output dir.
+```
+`./local/synthesize_e2e.sh` 调用 `${BIN_DIR}/../synthesize_e2e.py`，即可从文本文件中合成波形。
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize_e2e.py [-h]
+                         [--am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk}]
+                         [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                         [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                         [--tones_dict TONES_DICT]
+                         [--speaker_dict SPEAKER_DICT] [--spk_id SPK_ID]
+                         [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}]
+                         [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                         [--voc_stat VOC_STAT] [--lang LANG]
+                         [--inference_dir INFERENCE_DIR] [--ngpu NGPU]
+                         [--text TEXT] [--output_dir OUTPUT_DIR]
+Synthesize with acoustic model & vocoder
+optional arguments:
+  -h, --help            show this help message and exit
+  --am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk}
+                        Choose acoustic model type of tts task.
+  --am_config AM_CONFIG
+                        Config of acoustic model. Use deault config when it is
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --spk_id SPK_ID       spk id for multi speaker acoustic model
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
+  --lang LANG           Choose model language. zh or en
+  --inference_dir INFERENCE_DIR
+                        dir to save inference models
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --text TEXT           text to synthesize, a 'utt_id sentence' pair per line.
+  --output_dir OUTPUT_DIR
+                        output dir.
+```
+1. `--am` 声学模型格式是否符合 {model_name}_{dataset}
+2. `--am_config`, `--am_checkpoint`, `--am_stat` 和 `--phones_dict` 是声学模型的参数，对应于 fastspeech2 预训练模型中的 4 个文件。
+3. `--voc` 声码器(vocoder)格式是否符合 {model_name}_{dataset}
+4. `--voc_config`, `--voc_checkpoint`, `--voc_stat` 是声码器的参数，对应于 parallel wavegan 预训练模型中的 3 个文件。
+5. `--lang` 对应模型的语言可以是 `zh` 或 `en` 。
+6. `--test_metadata` 应为 `dump` 文件夹中 `test` 下的规范化元数据文件、
+7. `--text` 是文本文件，其中包含要合成的句子。
+8. `--output_dir` 是保存合成音频文件的目录。
+9. `--ngpu` 要使用的GPU数，如果 ngpu==0，则使用 cpu 。
+### 推理
+在合成之后，我们将在 `${train_output_path}/inference` 中得到 fastspeech2 和 pwgan 的静态模型
+`./local/inference.sh` 调用 `${BIN_DIR}/inference.py` 为 fastspeech2 + pwgan 综合提供了一个 paddle 静态模型推理示例。
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${train_output_path}
+```
+## 预训练模型
+预先训练的 FastSpeech2 模型，在音频边缘没有空白音频：
+- [fastspeech2_nosil_baker_ckpt_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_baker_ckpt_0.4.zip)
+- [fastspeech2_conformer_baker_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_conformer_baker_ckpt_0.5.zip)
+静态模型可以在这里下载 [fastspeech2_nosil_baker_static_0.4.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_nosil_baker_static_0.4.zip).
+Model | Step | eval/loss | eval/l1_loss | eval/duration_loss | eval/pitch_loss| eval/energy_loss 
+:-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:
+default| 2(gpu) x 76000|1.0991|0.59132|0.035815|0.31915|0.15287|
+conformer| 2(gpu) x 76000|1.0675|0.56103|0.035869|0.31553|0.15509|
+FastSpeech2检查点包含下列文件。
+```text
+fastspeech2_nosil_baker_ckpt_0.4
+├── default.yaml            # 用于训练 fastspeech2 的默认配置
+├── phone_id_map.txt        # 训练 fastspeech2 时的音素词汇文件
+├── snapshot_iter_76000.pdz # 模型参数和优化器状态
+└── speech_stats.npy        # 训练 fastspeech2 时用于规范化频谱图的统计数据
+```
+您可以使用以下脚本通过使用预训练的 fastspeech2 和 parallel wavegan 模型为 `${BIN_DIR}/../sentences.txt` 合成句子
+```bash
+source path.sh
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize_e2e.py \
+  --am=fastspeech2_csmsc \
+  --am_config=fastspeech2_nosil_baker_ckpt_0.4/default.yaml \
+  --am_ckpt=fastspeech2_nosil_baker_ckpt_0.4/snapshot_iter_76000.pdz \
+  --am_stat=fastspeech2_nosil_baker_ckpt_0.4/speech_stats.npy  \
+  --voc=pwgan_csmsc \
+  --voc_config=pwg_baker_ckpt_0.4/pwg_default.yaml \
+  --voc_ckpt=pwg_baker_ckpt_0.4/pwg_snapshot_iter_400000.pdz \
+  --voc_stat=pwg_baker_ckpt_0.4/pwg_stats.npy \
+  --lang=zh \
+  --text=${BIN_DIR}/../sentences.txt \
+  --output_dir=exp/default/test_e2e \
+  --inference_dir=exp/default/inference \
+  --phones_dict=fastspeech2_nosil_baker_ckpt_0.4/phone_id_map.txt
+```
--- a/examples/csmsc/tts3/conf/conformer.yaml
+++ b/examples/csmsc/tts3/conf/conformer.yaml
@@ -16,8 +16,8 @@ fmax: 7600         # Maximum frequency of Mel basis.
 n_mels: 80         # The number of mel basis.
 # Only used for the model using pitch features (e.g. FastSpeech2)
-f0min: 80          # Maximum f0 for pitch extraction.
+f0min: 80          # Minimum f0 for pitch extraction.
-f0max: 400         # Minimum f0 for pitch extraction.
+f0max: 400         # Maximum f0 for pitch extraction.
 ###########################################################
@@ -53,8 +53,8 @@ model:
    conformer_pos_enc_layer_type: rel_pos        # conformer positional encoding type
    conformer_self_attn_layer_type: rel_selfattn # conformer self-attention type
    conformer_activation_type: swish             # conformer activation type
-    use_macaron_style_in_conformer: true         # whether to use macaron style in conformer
+    use_macaron_style_in_conformer: True         # whether to use macaron style in conformer
-    use_cnn_in_conformer: true                   # whether to use CNN in conformer
+    use_cnn_in_conformer: True                   # whether to use CNN in conformer
    conformer_enc_kernel_size: 7                 # kernel size in CNN module of conformer-based encoder
    conformer_dec_kernel_size: 31                # kernel size in CNN module of conformer-based decoder
    init_type: xavier_uniform         # initialization type
@@ -70,14 +70,14 @@ model:
    pitch_predictor_dropout: 0.5               # dropout rate in pitch predictor
    pitch_embed_kernel_size: 1                 # kernel size of conv embedding layer for pitch
    pitch_embed_dropout: 0.0                   # dropout rate after conv embedding layer for pitch
-    stop_gradient_from_pitch_predictor: true   # whether to stop the gradient from pitch predictor to encoder
+    stop_gradient_from_pitch_predictor: True   # whether to stop the gradient from pitch predictor to encoder
    energy_predictor_layers: 2                 # number of conv layers in energy predictor
    energy_predictor_chans: 256                # number of channels of conv layers in energy predictor
    energy_predictor_kernel_size: 3            # kernel size of conv leyers in energy predictor
    energy_predictor_dropout: 0.5              # dropout rate in energy predictor
    energy_embed_kernel_size: 1                # kernel size of conv embedding layer for energy
    energy_embed_dropout: 0.0                  # dropout rate after conv embedding layer for energy
-    stop_gradient_from_energy_predictor: false # whether to stop the gradient from energy predictor to encoder
+    stop_gradient_from_energy_predictor: False # whether to stop the gradient from energy predictor to encoder

--- a/examples/csmsc/tts3/conf/default.yaml
+++ b/examples/csmsc/tts3/conf/default.yaml
@@ -16,8 +16,8 @@ fmax: 7600         # Maximum frequency of Mel basis.
 n_mels: 80         # The number of mel basis.
 # Only used for the model using pitch features (e.g. FastSpeech2)
-f0min: 80          # Maximum f0 for pitch extraction.
+f0min: 80          # Minimum f0 for pitch extraction.
-f0max: 400         # Minimum f0 for pitch extraction.
+f0max: 400         # Maximum f0 for pitch extraction.
 ###########################################################
@@ -64,14 +64,14 @@ model:
    pitch_predictor_dropout: 0.5               # dropout rate in pitch predictor
    pitch_embed_kernel_size: 1                 # kernel size of conv embedding layer for pitch
    pitch_embed_dropout: 0.0                   # dropout rate after conv embedding layer for pitch
-    stop_gradient_from_pitch_predictor: true   # whether to stop the gradient from pitch predictor to encoder
+    stop_gradient_from_pitch_predictor: True   # whether to stop the gradient from pitch predictor to encoder
    energy_predictor_layers: 2                 # number of conv layers in energy predictor
    energy_predictor_chans: 256                # number of channels of conv layers in energy predictor
    energy_predictor_kernel_size: 3            # kernel size of conv leyers in energy predictor
    energy_predictor_dropout: 0.5              # dropout rate in energy predictor
    energy_embed_kernel_size: 1                # kernel size of conv embedding layer for energy
    energy_embed_dropout: 0.0                  # dropout rate after conv embedding layer for energy
-    stop_gradient_from_energy_predictor: false # whether to stop the gradient from energy predictor to encoder
+    stop_gradient_from_energy_predictor: False # whether to stop the gradient from energy predictor to encoder
@@ -82,7 +82,6 @@ updater:
    use_masking: True                 # whether to apply masking for padded part in loss calculation
 ###########################################################
 #                     OPTIMIZER SETTING                   #
 ###########################################################

--- a/examples/csmsc/tts3/local/inference.sh
+++ b/examples/csmsc/tts3/local/inference.sh
@@ -48,4 +48,15 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
        --text=${BIN_DIR}/../sentences.txt \
        --output_dir=${train_output_path}/pd_infer_out \
        --phones_dict=dump/phone_id_map.txt
+fi
+# wavernn
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    python3 ${BIN_DIR}/../inference.py \
+        --inference_dir=${train_output_path}/inference \
+        --am=fastspeech2_csmsc \
+        --voc=wavernn_csmsc \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/pd_infer_out \
+        --phones_dict=dump/phone_id_map.txt
 fi
\ No newline at end of file
--- a/examples/csmsc/tts3/local/synthesize_e2e.sh
+++ b/examples/csmsc/tts3/local/synthesize_e2e.sh
@@ -89,3 +89,25 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
        --inference_dir=${train_output_path}/inference \
        --phones_dict=dump/phone_id_map.txt
 fi
+# wavernn
+if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
+    echo "in wavernn syn_e2e"
+    FLAGS_allocator_strategy=naive_best_fit \
+    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+    python3 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=fastspeech2_csmsc \
+        --am_config=${config_path} \
+        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+        --am_stat=dump/train/speech_stats.npy \
+        --voc=wavernn_csmsc \
+        --voc_config=wavernn_csmsc_ckpt_0.2.0/default.yaml \
+        --voc_ckpt=wavernn_csmsc_ckpt_0.2.0/snapshot_iter_400000.pdz \
+        --voc_stat=wavernn_csmsc_ckpt_0.2.0/feats_stats.npy \
+        --lang=zh \
+        --text=${BIN_DIR}/../sentences.txt \
+        --output_dir=${train_output_path}/test_e2e \
+        --phones_dict=dump/phone_id_map.txt \
+        --inference_dir=${train_output_path}/inference
+fi
--- a/examples/csmsc/tts3/run.sh
+++ b/examples/csmsc/tts3/run.sh
@@ -18,7 +18,7 @@ source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    # prepare data
-    bash ./local/preprocess.sh ${conf_path} || exit -1
+    ./local/preprocess.sh ${conf_path} || exit -1
 fi
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
@@ -40,3 +40,4 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
    # inference with static model
    CUDA_VISIBLE_DEVICES=${gpus} ./local/inference.sh ${train_output_path} || exit -1
 fi
--- a/examples/csmsc/voc1/conf/default.yaml
+++ b/examples/csmsc/voc1/conf/default.yaml
@@ -34,10 +34,10 @@ generator_params:
    aux_context_window: 2 # Context window size for auxiliary feature.
                          # If set to 2, previous 2 and future 2 frames will be considered.
    dropout: 0.0          # Dropout rate. 0.0 means no dropout applied.
-    bias: true            # use bias in residual blocks
+    bias: True            # use bias in residual blocks
-    use_weight_norm: true # Whether to use weight norm.
+    use_weight_norm: True # Whether to use weight norm.
                          # If set to true, it will be applied to all of the conv layers.
-    use_causal_conv: false               # use causal conv in residual blocks and upsample layers
+    use_causal_conv: False               # use causal conv in residual blocks and upsample layers
    upsample_scales: [4, 5, 3, 5]     # Upsampling scales. Prodcut of these must be the same as hop size.
    interpolate_mode: "nearest" # upsample net interpolate mode
    freq_axis_kernel_size: 1 # upsamling net: convolution kernel size in frequencey axis
@@ -53,8 +53,8 @@ discriminator_params:
    kernel_size: 3        # Number of output channels.
    layers: 10            # Number of conv layers.
    conv_channels: 64     # Number of chnn layers.
-    bias: true            # Whether to use bias parameter in conv.
+    bias: True            # Whether to use bias parameter in conv.
-    use_weight_norm: true # Whether to use weight norm.
+    use_weight_norm: True # Whether to use weight norm.
                          # If set to true, it will be applied to all of the conv layers.
    nonlinear_activation: "leakyrelu" # Nonlinear function after each conv.
    nonlinear_activation_params:      # Nonlinear function parameters

--- a/examples/csmsc/voc3/conf/default.yaml
+++ b/examples/csmsc/voc3/conf/default.yaml
@@ -63,13 +63,13 @@ discriminator_params:
 ###########################################################
 #                   STFT LOSS SETTING                     #
 ###########################################################
-use_stft_loss: true
+use_stft_loss: True
 stft_loss_params:
    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss.
    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
    window: "hann"                # Window function for STFT-based loss
-use_subband_stft_loss: true
+use_subband_stft_loss: True
 subband_stft_loss_params:
    fft_sizes: [384, 683, 171]  # List of FFT size for STFT-based loss.
    hop_sizes: [30, 60, 10]     # List of hop size for STFT-based loss
@@ -79,7 +79,7 @@ subband_stft_loss_params:
 ###########################################################
 #               ADVERSARIAL LOSS SETTING                  #
 ###########################################################
-use_feat_match_loss: false # Whether to use feature matching loss.
+use_feat_match_loss: False # Whether to use feature matching loss.
 lambda_adv: 2.5            # Loss balancing coefficient for adversarial loss.
 ###########################################################

--- a/examples/csmsc/voc3/conf/finetune.yaml
+++ b/examples/csmsc/voc3/conf/finetune.yaml
@@ -63,13 +63,13 @@ discriminator_params:
 ###########################################################
 #                   STFT LOSS SETTING                     #
 ###########################################################
-use_stft_loss: true
+use_stft_loss: True
 stft_loss_params:
    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
    win_lengths: [600, 1200, 240] # List of window length for STFT-based loss.
    window: "hann"                # Window function for STFT-based loss
-use_subband_stft_loss: true
+use_subband_stft_loss: True
 subband_stft_loss_params:
    fft_sizes: [384, 683, 171]  # List of FFT size for STFT-based loss.
    hop_sizes: [30, 60, 10]     # List of hop size for STFT-based loss.
@@ -79,7 +79,7 @@ subband_stft_loss_params:
 ###########################################################
 #               ADVERSARIAL LOSS SETTING                  #
 ###########################################################
-use_feat_match_loss: false # Whether to use feature matching loss.
+use_feat_match_loss: False # Whether to use feature matching loss.
 lambda_adv: 2.5            # Loss balancing coefficient for adversarial loss.
 ###########################################################

--- a/examples/csmsc/voc4/conf/default.yaml
+++ b/examples/csmsc/voc4/conf/default.yaml
@@ -65,7 +65,7 @@ discriminator_params:
 ###########################################################
 #                   STFT LOSS SETTING                     #
 ###########################################################
-use_stft_loss: true
+use_stft_loss: True
 stft_loss_params:
    fft_sizes: [1024, 2048, 512]  # List of FFT size for STFT-based loss.
    hop_sizes: [120, 240, 50]     # List of hop size for STFT-based loss
@@ -78,9 +78,9 @@ lambda_aux: 1.0                   # Loss balancing coefficient for aux loss.
 ###########################################################
 lambda_adv: 1.0 # Loss balancing coefficient for adv loss.
 generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
 discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
 ###########################################################
 #                  DATA LOADER SETTING                    #

--- a/examples/csmsc/voc5/conf/default.yaml
+++ b/examples/csmsc/voc5/conf/default.yaml
@@ -35,12 +35,12 @@ generator_params:
        - [1, 3, 5]
        - [1, 3, 5]
        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
+    use_additional_convs: True            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
+    bias: True                            # Whether to use bias parameter in conv.
    nonlinear_activation: "leakyrelu"     # Nonlinear activation type.
    nonlinear_activation_params:          # Nonlinear activation paramters.
        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
+    use_weight_norm: True                 # Whether to apply weight normalization.
 ###########################################################
@@ -60,12 +60,12 @@ discriminator_params:
        channels: 128                      # Initial number of channels.
        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
        max_groups: 16                     # Maximum number of groups in downsampling conv layers.
-        bias: true
+        bias: True
        downsample_scales: [4, 4, 4, 4, 1] # Downsampling scales.
        nonlinear_activation: "leakyrelu"  # Nonlinear activation.
        nonlinear_activation_params:
            negative_slope: 0.1
-    follow_official_norm: true             # Whether to follow the official norm setting.
+    follow_official_norm: True             # Whether to follow the official norm setting.
    periods: [2, 3, 5, 7, 11]              # List of period for multi-period discriminator.
    period_discriminator_params:
        in_channels: 1                     # Number of input channels.
@@ -74,19 +74,19 @@ discriminator_params:
        channels: 32                       # Initial number of channels.
        downsample_scales: [3, 3, 3, 3, 1] # Downsampling scales.
        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        bias: true                         # Whether to use bias parameter in conv layer."
+        bias: True                         # Whether to use bias parameter in conv layer."
        nonlinear_activation: "leakyrelu"  # Nonlinear activation.
        nonlinear_activation_params:       # Nonlinear activation paramters.
            negative_slope: 0.1
-        use_weight_norm: true              # Whether to apply weight normalization.
+        use_weight_norm: True              # Whether to apply weight normalization.
-        use_spectral_norm: false           # Whether to apply spectral normalization.
+        use_spectral_norm: False           # Whether to apply spectral normalization.
 ###########################################################
 #                   STFT LOSS SETTING                     #
 ###########################################################
-use_stft_loss: false                 # Whether to use multi-resolution STFT loss.
+use_stft_loss: False                 # Whether to use multi-resolution STFT loss.
-use_mel_loss: true                   # Whether to use Mel-spectrogram loss.
+use_mel_loss: True                   # Whether to use Mel-spectrogram loss.
 mel_loss_params:
    fs: 24000
    fft_size: 2048
@@ -98,14 +98,14 @@ mel_loss_params:
    fmax: 12000
    log_base: null
 generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
 discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
-use_feat_match_loss: true
+use_feat_match_loss: True
 feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
-    average_by_layers: false         # Whether to average loss by #layers in each discriminator.
+    average_by_layers: False         # Whether to average loss by #layers in each discriminator.
-    include_final_outputs: false     # Whether to include final outputs in feat match loss calculation.
+    include_final_outputs: False     # Whether to include final outputs in feat match loss calculation.
 ###########################################################
 #               ADVERSARIAL LOSS SETTING                  #

--- a/examples/csmsc/voc5/conf/finetune.yaml
+++ b/examples/csmsc/voc5/conf/finetune.yaml
@@ -35,12 +35,12 @@ generator_params:
        - [1, 3, 5]
        - [1, 3, 5]
        - [1, 3, 5]
-    use_additional_convs: true            # Whether to use additional conv layer in residual blocks.
+    use_additional_convs: True            # Whether to use additional conv layer in residual blocks.
-    bias: true                            # Whether to use bias parameter in conv.
+    bias: True                            # Whether to use bias parameter in conv.
    nonlinear_activation: "leakyrelu"     # Nonlinear activation type.
    nonlinear_activation_params:          # Nonlinear activation paramters.
        negative_slope: 0.1
-    use_weight_norm: true                 # Whether to apply weight normalization.
+    use_weight_norm: True                 # Whether to apply weight normalization.
 ###########################################################
@@ -60,12 +60,12 @@ discriminator_params:
        channels: 128                      # Initial number of channels.
        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
        max_groups: 16                     # Maximum number of groups in downsampling conv layers.
-        bias: true
+        bias: True
        downsample_scales: [4, 4, 4, 4, 1] # Downsampling scales.
        nonlinear_activation: "leakyrelu"  # Nonlinear activation.
        nonlinear_activation_params:
            negative_slope: 0.1
-    follow_official_norm: true             # Whether to follow the official norm setting.
+    follow_official_norm: True             # Whether to follow the official norm setting.
    periods: [2, 3, 5, 7, 11]              # List of period for multi-period discriminator.
    period_discriminator_params:
        in_channels: 1                     # Number of input channels.
@@ -74,19 +74,19 @@ discriminator_params:
        channels: 32                       # Initial number of channels.
        downsample_scales: [3, 3, 3, 3, 1] # Downsampling scales.
        max_downsample_channels: 1024      # Maximum number of channels in downsampling conv layers.
-        bias: true                         # Whether to use bias parameter in conv layer."
+        bias: True                         # Whether to use bias parameter in conv layer."
        nonlinear_activation: "leakyrelu"  # Nonlinear activation.
        nonlinear_activation_params:       # Nonlinear activation paramters.
            negative_slope: 0.1
-        use_weight_norm: true              # Whether to apply weight normalization.
+        use_weight_norm: True              # Whether to apply weight normalization.
-        use_spectral_norm: false           # Whether to apply spectral normalization.
+        use_spectral_norm: False           # Whether to apply spectral normalization.
 ###########################################################
 #                   STFT LOSS SETTING                     #
 ###########################################################
-use_stft_loss: false                 # Whether to use multi-resolution STFT loss.
+use_stft_loss: False                 # Whether to use multi-resolution STFT loss.
-use_mel_loss: true                   # Whether to use Mel-spectrogram loss.
+use_mel_loss: True                   # Whether to use Mel-spectrogram loss.
 mel_loss_params:
    fs: 24000
    fft_size: 2048
@@ -98,14 +98,14 @@ mel_loss_params:
    fmax: 12000
    log_base: null
 generator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
 discriminator_adv_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
-use_feat_match_loss: true
+use_feat_match_loss: True
 feat_match_loss_params:
-    average_by_discriminators: false # Whether to average loss by #discriminators.
+    average_by_discriminators: False # Whether to average loss by #discriminators.
-    average_by_layers: false         # Whether to average loss by #layers in each discriminator.
+    average_by_layers: False         # Whether to average loss by #layers in each discriminator.
-    include_final_outputs: false     # Whether to include final outputs in feat match loss calculation.
+    include_final_outputs: False     # Whether to include final outputs in feat match loss calculation.
 ###########################################################
 #               ADVERSARIAL LOSS SETTING                  #

--- a/examples/csmsc/voc6/README.md
+++ b/examples/csmsc/voc6/README.md
+# WaveRNN with CSMSC
+This example contains code used to train a [WaveRNN](https://arxiv.org/abs/1802.08435) model with [Chinese Standard Mandarin Speech Copus](https://www.data-baker.com/open_source.html).
+## Dataset
+### Download and Extract
+Download CSMSC from the [official website](https://www.data-baker.com/data/index/source) and extract it to `~/datasets`. Then the dataset is in the directory `~/datasets/BZNSYP`.
+### Get MFA Result and Extract
+We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) results to cut silence at the edge of audio.
+You can download from here [baker_alignment_tone.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/BZNSYP/with_tone/baker_alignment_tone.tar.gz), or train your MFA model reference to  [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) of our repo.
+## Get Started
+Assume the path to the dataset is `~/datasets/BZNSYP`.
+Assume the path to the MFA result of CSMSC is `./baker_alignment_tone`.
+Run the command below to
+1. **source path**.
+2. preprocess the dataset.
+3. train the model.
+4. synthesize wavs.
+    - synthesize waveform from `metadata.jsonl`.
+```bash
+./run.sh
+```
+You can choose a range of stages you want to run, or set `stage` equal to `stop-stage` to use only one stage, for example, running the following command will only preprocess the dataset.
+```bash
+./run.sh --stage 0 --stop-stage 0
+```
+### Data Preprocessing
+```bash
+./local/preprocess.sh ${conf_path}
+```
+When it is done. A `dump` folder is created in the current directory. The structure of the dump folder is listed below.
+```text
+dump
+├── dev
+│   ├── norm
+│   └── raw
+├── test
+│   ├── norm
+│   └── raw
+└── train
+    ├── norm
+    ├── raw
+    └── feats_stats.npy
+```
+The dataset is split into 3 parts, namely `train`, `dev`, and `test`, each of which contains a `norm` and `raw` subfolder. The `raw` folder contains the log magnitude of the mel spectrogram of each utterance, while the norm folder contains the normalized spectrogram. The statistics used to normalize the spectrogram are computed from the training set, which is located in `dump/train/feats_stats.npy`.
+Also, there is a `metadata.jsonl` in each subfolder. It is a table-like file that contains id and paths to the spectrogram of each utterance.
+### Model Training
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
+```
+`./local/train.sh` calls `${BIN_DIR}/train.py`.
+Here's the complete help message.
+```text
+usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
+                [--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
+                [--ngpu NGPU]
+Train a WaveRNN model.
+optional arguments:
+  -h, --help            show this help message and exit
+  --config CONFIG       config file to overwrite default config.
+  --train-metadata TRAIN_METADATA
+                        training data.
+  --dev-metadata DEV_METADATA
+                        dev data.
+  --output-dir OUTPUT_DIR
+                        output dir.
+  --ngpu NGPU           if ngpu == 0, use cpu.
+```
+1. `--config` is a config file in yaml format to overwrite the default config, which can be found at `conf/default.yaml`.
+2. `--train-metadata` and `--dev-metadata` should be the metadata file in the normalized subfolder of `train` and `dev` in the `dump` folder.
+3. `--output-dir` is the directory to save the results of the experiment. Checkpoints are saved in `checkpoints/` inside this directory.
+4. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
+### Synthesizing
+`./local/synthesize.sh` calls `${BIN_DIR}/synthesize.py`, which can synthesize waveform from `metadata.jsonl`.
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize.py [-h] [--config CONFIG] [--checkpoint CHECKPOINT]
+                     [--test-metadata TEST_METADATA] [--output-dir OUTPUT_DIR]
+                     [--ngpu NGPU]
+Synthesize with WaveRNN.
+optional arguments:
+  -h, --help            show this help message and exit
+  --config CONFIG       Vocoder config file.
+  --checkpoint CHECKPOINT
+                        snapshot to load.
+  --test-metadata TEST_METADATA
+                        dev data.
+  --output-dir OUTPUT_DIR
+                        output dir.
+  --ngpu NGPU           if ngpu == 0, use cpu.
+```
+1. `--config` wavernn config file. You should use the same config with which the model is trained.
+2. `--checkpoint` is the checkpoint to load. Pick one of the checkpoints from `checkpoints` inside the training output directory.
+3. `--test-metadata` is the metadata of the test dataset. Use the `metadata.jsonl` in the `dev/norm` subfolder from the processed directory.
+4. `--output-dir` is the directory to save the synthesized audio files.
+5. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
+## Pretrained Models
+The pretrained model can be downloaded here [wavernn_csmsc_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/wavernn/wavernn_csmsc_ckpt_0.2.0.zip).
+The static model can be downloaded here [wavernn_csmsc_static_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/wavernn/wavernn_csmsc_static_0.2.0.zip).
+Model | Step | eval/loss
+:-------------:|:------------:| :------------:
+default| 1(gpu) x 400000|2.602768
+WaveRNN checkpoint contains files listed below.
+```text
+wavernn_csmsc_ckpt_0.2.0
+├── default.yaml                   # default config used to train wavernn
+├── feats_stats.npy                # statistics used to normalize spectrogram when training wavernn
+└── snapshot_iter_400000.pdz       # parameters of wavernn
+```
--- a/examples/csmsc/voc6/conf/default.yaml
+++ b/examples/csmsc/voc6/conf/default.yaml
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+fs: 24000                # Sampling rate.
+n_fft: 2048              # FFT size (samples).
+n_shift: 300             # Hop size (samples). 12.5ms
+win_length: 1200         # Window length (samples). 50ms
+                         # If set to null, it will be the same as fft_size.
+window: "hann"           # Window function.
+n_mels: 80               # Number of mel basis.
+fmin: 80                 # Minimum freq in mel basis calculation. (Hz)
+fmax: 7600               # Maximum frequency in mel basis calculation. (Hz)
+mu_law: True             # Recommended to suppress noise if using raw bitsexit()
+###########################################################
+#                       MODEL SETTING                     #
+###########################################################
+model:
+    rnn_dims: 512                     # Hidden dims of RNN Layers.
+    fc_dims: 512
+    bits: 9                           # Bit depth of signal
+    aux_context_window: 2             # Context window size for auxiliary feature.
+                                      # If set to 2, previous 2 and future 2 frames will be considered.
+    aux_channels: 80                  # Number of channels for auxiliary feature conv.
+                                      # Must be the same as num_mels.
+    upsample_scales: [4, 5, 3, 5]     # Upsampling scales. Prodcut of these must be the same as hop size, same with pwgan here
+    compute_dims: 128                 # Dims of Conv1D in MelResNet.
+    res_out_dims: 128                 # Dims of output in MelResNet.
+    res_blocks: 10                    # Number of residual blocks.
+    mode: RAW                         # either 'raw'(softmax on raw bits) or 'mold' (sample from mixture of logistics)
+inference:
+    gen_batched: True                 # whether to genenate sample in batch mode
+    target: 12000                     # target number of samples to be generated in each batch entry
+    overlap: 600                      # number of samples for crossfading between batches
+###########################################################
+#                  DATA LOADER SETTING                    #
+###########################################################
+batch_size: 64              # Batch size.
+batch_max_steps: 4500       # Length of each audio in batch. Make sure dividable by hop_size.
+num_workers: 2              # Number of workers in DataLoader.
+###########################################################
+#                     OPTIMIZER SETTING                   #
+###########################################################
+grad_clip: 4.0
+learning_rate: 1.0e-4                
+###########################################################
+#                    INTERVAL SETTING                     #
+###########################################################
+train_max_steps: 400000               # Number of training steps.
+save_interval_steps: 5000             # Interval steps to save checkpoint.
+eval_interval_steps: 1000             # Interval steps to evaluate the network.
+gen_eval_samples_interval_steps: 5000 # the iteration interval of generating valid samples
+generate_num: 5                       # number of samples to generate at each checkpoint
+###########################################################
+#                     OTHER SETTING                       #
+###########################################################
+num_snapshots: 10                 # max number of snapshots to keep while training
+seed: 42                          # random seed for paddle, random, and np.random
--- a/examples/csmsc/voc6/local/preprocess.sh
+++ b/examples/csmsc/voc6/local/preprocess.sh
+#!/bin/bash
+stage=0
+stop_stage=100
+config_path=$1
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # get durations from MFA's result
+    echo "Generate durations.txt from MFA results ..."
+    python3 ${MAIN_ROOT}/utils/gen_duration_from_textgrid.py \
+        --inputdir=./baker_alignment_tone \
+        --output=durations.txt \
+        --config=${config_path}
+fi
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # extract features
+    echo "Extract features ..."
+    python3 ${BIN_DIR}/../gan_vocoder/preprocess.py \
+        --rootdir=~/datasets/BZNSYP/ \
+        --dataset=baker \
+        --dumpdir=dump \
+        --dur-file=durations.txt \
+        --config=${config_path} \
+        --cut-sil=True \
+        --num-cpu=20
+fi
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # get features' stats(mean and std)
+    echo "Get features' stats ..."
+    python3 ${MAIN_ROOT}/utils/compute_statistics.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --field-name="feats"
+fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # normalize, dev and test should use train's stats
+    echo "Normalize ..."
+    python3 ${BIN_DIR}/../gan_vocoder/normalize.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --dumpdir=dump/train/norm \
+        --stats=dump/train/feats_stats.npy
+    python3 ${BIN_DIR}/../gan_vocoder/normalize.py \
+        --metadata=dump/dev/raw/metadata.jsonl \
+        --dumpdir=dump/dev/norm \
+        --stats=dump/train/feats_stats.npy
+    python3 ${BIN_DIR}/../gan_vocoder/normalize.py \
+        --metadata=dump/test/raw/metadata.jsonl \
+        --dumpdir=dump/test/norm \
+        --stats=dump/train/feats_stats.npy
+fi
--- a/examples/csmsc/voc6/local/synthesize.sh
+++ b/examples/csmsc/voc6/local/synthesize.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+ckpt_name=$3
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/synthesize.py \
+    --config=${config_path} \
+    --checkpoint=${train_output_path}/checkpoints/${ckpt_name} \
+    --test-metadata=dump/test/norm/metadata.jsonl \
+    --output-dir=${train_output_path}/test
--- a/examples/csmsc/voc6/local/train.sh
+++ b/examples/csmsc/voc6/local/train.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+FLAGS_cudnn_exhaustive_search=true \
+FLAGS_conv_workspace_size_limit=4000 \
+python ${BIN_DIR}/train.py \
+    --train-metadata=dump/train/norm/metadata.jsonl \
+    --dev-metadata=dump/dev/norm/metadata.jsonl \
+    --config=${config_path} \
+    --output-dir=${train_output_path} \
+    --ngpu=1
--- a/examples/csmsc/voc6/path.sh
+++ b/examples/csmsc/voc6/path.sh
+#!/bin/bash
+export MAIN_ROOT=`realpath ${PWD}/../../../`
+export PATH=${MAIN_ROOT}:${MAIN_ROOT}/utils:${PATH}
+export LC_ALL=C
+export PYTHONDONTWRITEBYTECODE=1
+# Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
+export PYTHONIOENCODING=UTF-8
+export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
+MODEL=wavernn
+export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
\ No newline at end of file
--- a/examples/csmsc/voc6/run.sh
+++ b/examples/csmsc/voc6/run.sh
+#!/bin/bash
+set -e
+source path.sh
+gpus=0,1
+stage=0
+stop_stage=100
+conf_path=conf/default.yaml
+train_output_path=exp/default
+test_input=dump/dump_gta_test
+ckpt_name=snapshot_iter_100000.pdz
+source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # prepare data
+    ./local/preprocess.sh ${conf_path} || exit -1
+fi
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # prepare data
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1
+fi
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # synthesize
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
+fi
--- a/examples/esc50/README.md
+++ b/examples/esc50/README.md
@@ -122,3 +122,6 @@ $ CUDA_VISIBLE_DEVICES=0 ./run.sh 4 cpu ./export /audio/dog.wav
 - `device`: 指定模型预测时使用的设备。
 - `model_dir`: 导出静态图模型和参数文件的保存目录。
 - `wav`: 指定预测的音频文件。
+## Reference
+* [PANNs(PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition)](https://arxiv.org/abs/1912.10211)
--- a/examples/ljspeech/tts0/README.md
+++ b/examples/ljspeech/tts0/README.md
-# Tacotron2 with LJSpeech
+# Tacotron2 with LJSpeech-1.1
-PaddlePaddle dynamic graph implementation of Tacotron2, a neural network architecture for speech synthesis directly from the text. The implementation is based on [Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions](https://arxiv.org/abs/1712.05884).
+This example contains code used to train a [Tacotron2](https://arxiv.org/abs/1712.05884) model with [LJSpeech-1.1](https://keithito.com/LJ-Speech-Dataset/)
 ## Dataset
-We experiment with the LJSpeech dataset. Download and unzip [LJSpeech](https://keithito.com/LJ-Speech-Dataset/).
+### Download and Extract
+Download LJSpeech-1.1 from the [official website](https://keithito.com/LJ-Speech-Dataset/).
+### Get MFA Result and Extract
+We use [MFA](https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner) to get phonemes for Tacotron2, the durations of MFA are not needed here.
+You can download from here [ljspeech_alignment.tar.gz](https://paddlespeech.bj.bcebos.com/MFA/LJSpeech-1.1/ljspeech_alignment.tar.gz), or train your MFA model reference to [mfa example](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/other/mfa) of our repo.
-```bash
-wget https://data.keithito.com/data/speech/LJSpeech-1.1.tar.bz2
-tar xjvf LJSpeech-1.1.tar.bz2
-```
 ## Get Started
 Assume the path to the dataset is `~/datasets/LJSpeech-1.1`.
+Assume the path to the MFA result of LJSpeech-1.1 is `./ljspeech_alignment`.
 Run the command below to
 1. **source path**.
 2. preprocess the dataset.
 3. train the model.
-4. synthesize mels.
+4. synthesize wavs.
+    - synthesize waveform from `metadata.jsonl`.
+    - synthesize waveform from a text file.
 ```bash
 ./run.sh
 ```
@@ -26,64 +31,217 @@ You can choose a range of stages you want to run, or set `stage` equal to `stop-
 ```bash
 ./local/preprocess.sh ${conf_path}
 ```
+When it is done. A `dump` folder is created in the current directory. The structure of the dump folder is listed below.
+```text
+dump
+├── dev
+│   ├── norm
+│   └── raw
+├── phone_id_map.txt
+├── speaker_id_map.txt
+├── test
+│   ├── norm
+│   └── raw
+└── train
+    ├── norm
+    ├── raw
+    └── speech_stats.npy
+```
+The dataset is split into 3 parts, namely `train`, `dev`, and` test`, each of which contains a `norm` and `raw` subfolder. The raw folder contains speech features of each utterance, while the norm folder contains normalized ones. The statistics used to normalize features are computed from the training set, which is located in `dump/train/*_stats.npy`.
+Also, there is a `metadata.jsonl` in each subfolder. It is a table-like file that contains phones, text_lengths, speech_lengths, durations, the path of speech features, speaker, and the id of each utterance.
 ### Model Training
-`./local/train.sh` calls `${BIN_DIR}/train.py`.
 ```bash
 CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path}
 ```
+`./local/train.sh` calls `${BIN_DIR}/train.py`.
 Here's the complete help message.
 ```text
-usage: train.py [-h] [--config FILE] [--data DATA_DIR] [--output OUTPUT_DIR]
+usage: train.py [-h] [--config CONFIG] [--train-metadata TRAIN_METADATA]
-                [--checkpoint_path CHECKPOINT_PATH] [--ngpu NGPU] [--opts ...]
+                [--dev-metadata DEV_METADATA] [--output-dir OUTPUT_DIR]
+                [--ngpu NGPU] [--phones-dict PHONES_DICT]
+Train a Tacotron2 model.
 optional arguments:
  -h, --help            show this help message and exit
-  --config FILE         path of the config file to overwrite to default config
+  --config CONFIG       tacotron2 config file.
-                        with.
+  --train-metadata TRAIN_METADATA
-  --data DATA_DIR       path to the dataset.
+                        training data.
-  --output OUTPUT_DIR   path to save checkpoint and logs.
+  --dev-metadata DEV_METADATA
-  --checkpoint_path CHECKPOINT_PATH
+                        dev data.
-                        path of the checkpoint to load
+  --output-dir OUTPUT_DIR
+                        output dir.
  --ngpu NGPU           if ngpu == 0, use cpu.
-  --opts ...            options to overwrite --config file and the default
+  --phones-dict PHONES_DICT
-                        config, passing in KEY VALUE pairs
+                        phone vocabulary file.
 ```
+1. `--config` is a config file in yaml format to overwrite the default config, which can be found at `conf/default.yaml`.
-If you want to train on CPU, just set `--ngpu=0`.
+2. `--train-metadata` and `--dev-metadata` should be the metadata file in the normalized subfolder of `train` and `dev` in the `dump` folder.
-If you want to train on multiple GPUs, just set `--ngpu` as the num of GPU.
+3. `--output-dir` is the directory to save the results of the experiment. Checkpoints are saved in `checkpoints/` inside this directory.
-By default, training will be resumed from the latest checkpoint in `--output`, if you want to start a new training, please use a new `${OUTPUTPATH}` with no checkpoint.
+4. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
-And if you want to resume from another existing model, you should set `checkpoint_path` to be the checkpoint path you want to load.
+5. `--phones-dict` is the path of the phone vocabulary file.
-**Note: The checkpoint path cannot contain the file extension.**
 ### Synthesizing
-`./local/synthesize.sh` calls `${BIN_DIR}/synthesize.py`,  which synthesize **mels**  from text_list here.
+We use [parallel wavegan](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/voc1) as the neural vocoder.
+Download pretrained parallel wavegan model from [pwg_ljspeech_ckpt_0.5.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_ljspeech_ckpt_0.5.zip) and unzip it.
+```bash
+unzip pwg_ljspeech_ckpt_0.5.zip
+```
+Parallel WaveGAN checkpoint contains files listed below.
+```text
+pwg_ljspeech_ckpt_0.5
+├── pwg_default.yaml              # default config used to train parallel wavegan
+├── pwg_snapshot_iter_400000.pdz  # generator parameters of parallel wavegan
+└── pwg_stats.npy                 # statistics used to normalize spectrogram when training parallel wavegan
+```
+`./local/synthesize.sh` calls `${BIN_DIR}/../synthesize.py`, which can synthesize waveform from `metadata.jsonl`.
 ```bash
-CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${train_output_path} ${ckpt_name}
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name}
 ```
 ```text
-usage: synthesize.py [-h] [--config FILE] [--checkpoint_path CHECKPOINT_PATH]
+usage: synthesize.py [-h]
-                     [--input INPUT] [--output OUTPUT] [--ngpu NGPU]
+                     [--am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}]
-                     [--opts ...] [-v]
+                     [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                     [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                     [--tones_dict TONES_DICT] [--speaker_dict SPEAKER_DICT]
+                     [--voice-cloning VOICE_CLONING]
+                     [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}]
+                     [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                     [--voc_stat VOC_STAT] [--ngpu NGPU]
+                     [--test_metadata TEST_METADATA] [--output_dir OUTPUT_DIR]
-generate mel spectrogram with TransformerTTS.
+Synthesize with acoustic model & vocoder
 optional arguments:
  -h, --help            show this help message and exit
-  --config FILE         extra config to overwrite the default config
+  --am {speedyspeech_csmsc,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}
-  --checkpoint_path CHECKPOINT_PATH
+                        Choose acoustic model type of tts task.
-                        path of the checkpoint to load.
+  --am_config AM_CONFIG
-  --input INPUT         path of the text sentences
+                        Config of acoustic model. Use deault config when it is
-  --output OUTPUT       path to save outputs
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --voice-cloning VOICE_CLONING
+                        whether training voice cloning model.
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
  --ngpu NGPU           if ngpu == 0, use cpu.
-  --opts ...            options to overwrite --config file and the default
+  --test_metadata TEST_METADATA
-                        config, passing in KEY VALUE pairs
+                        test metadata.
-  -v, --verbose         print msg
+  --output_dir OUTPUT_DIR
+                        output dir.
 ```
-**Ps.** You can use [waveflow](https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/examples/ljspeech/voc0) as the neural vocoder to synthesize mels to wavs. (Please  refer to `synthesize.sh` in our  LJSpeech waveflow example)
+`./local/synthesize_e2e.sh` calls `${BIN_DIR}/../synthesize_e2e.py`, which can synthesize waveform from text file.
+```bash
+CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name}
+```
+```text
+usage: synthesize_e2e.py [-h]
+                         [--am {speedyspeech_csmsc,speedyspeech_aishell3,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}]
+                         [--am_config AM_CONFIG] [--am_ckpt AM_CKPT]
+                         [--am_stat AM_STAT] [--phones_dict PHONES_DICT]
+                         [--tones_dict TONES_DICT]
+                         [--speaker_dict SPEAKER_DICT] [--spk_id SPK_ID]
+                         [--voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc,style_melgan_csmsc,hifigan_csmsc}]
+                         [--voc_config VOC_CONFIG] [--voc_ckpt VOC_CKPT]
+                         [--voc_stat VOC_STAT] [--lang LANG]
+                         [--inference_dir INFERENCE_DIR] [--ngpu NGPU]
+                         [--text TEXT] [--output_dir OUTPUT_DIR]
+Synthesize with acoustic model & vocoder
+optional arguments:
+  -h, --help            show this help message and exit
+  --am {speedyspeech_csmsc,speedyspeech_aishell3,fastspeech2_csmsc,fastspeech2_ljspeech,fastspeech2_aishell3,fastspeech2_vctk,tacotron2_csmsc}
+                        Choose acoustic model type of tts task.
+  --am_config AM_CONFIG
+                        Config of acoustic model. Use deault config when it is
+                        None.
+  --am_ckpt AM_CKPT     Checkpoint file of acoustic model.
+  --am_stat AM_STAT     mean and standard deviation used to normalize
+                        spectrogram when training acoustic model.
+  --phones_dict PHONES_DICT
+                        phone vocabulary file.
+  --tones_dict TONES_DICT
+                        tone vocabulary file.
+  --speaker_dict SPEAKER_DICT
+                        speaker id map file.
+  --spk_id SPK_ID       spk id for multi speaker acoustic model
+  --voc {pwgan_csmsc,pwgan_ljspeech,pwgan_aishell3,pwgan_vctk,mb_melgan_csmsc,style_melgan_csmsc,hifigan_csmsc}
+                        Choose vocoder type of tts task.
+  --voc_config VOC_CONFIG
+                        Config of voc. Use deault config when it is None.
+  --voc_ckpt VOC_CKPT   Checkpoint file of voc.
+  --voc_stat VOC_STAT   mean and standard deviation used to normalize
+                        spectrogram when training voc.
+  --lang LANG           Choose model language. zh or en
+  --inference_dir INFERENCE_DIR
+                        dir to save inference models
+  --ngpu NGPU           if ngpu == 0, use cpu.
+  --text TEXT           text to synthesize, a 'utt_id sentence' pair per line.
+  --output_dir OUTPUT_DIR
+                        output dir.
+```
+1. `--am` is acoustic model type with the format {model_name}_{dataset}
+2. `--am_config`, `--am_checkpoint`, `--am_stat` and `--phones_dict` are arguments for acoustic model, which correspond to the 4 files in the Tacotron2 pretrained model.
+3. `--voc` is vocoder type with the format {model_name}_{dataset}
+4. `--voc_config`, `--voc_checkpoint`, `--voc_stat` are arguments for vocoder, which correspond to the 3 files in the parallel wavegan pretrained model.
+5. `--lang` is the model language, which can be `zh` or `en`.
+6. `--test_metadata` should be the metadata file in the normalized subfolder of `test`  in the `dump` folder.
+7. `--text` is the text file, which contains sentences to synthesize.
+8. `--output_dir` is the directory to save synthesized audio files.
+9. `--ngpu` is the number of gpus to use, if ngpu == 0, use cpu.
+## Pretrained Model
+Pretrained Tacotron2 model with no silence in the edge of audios:
+- [tacotron2_ljspeech_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.2.0.zip)
-## Pretrained Models
+Model | Step | eval/loss | eval/l1_loss | eval/mse_loss | eval/bce_loss| eval/attn_loss 
-Pretrained Models can be downloaded from the links below. We provide 2 models with different configurations.
+:-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:
+default| 1(gpu) x 60300|0.554092|0.394260|0.141046|0.018747|3.8e-05|
-1. This model uses a binary classifier to predict the stop token. [tacotron2_ljspeech_ckpt_0.3.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.3.zip)
+Tacotron2 checkpoint contains files listed below.
+```text
+tacotron2_ljspeech_ckpt_0.2.0
+├── default.yaml            # default config used to train Tacotron2
+├── phone_id_map.txt        # phone vocabulary file when training Tacotron2
+├── snapshot_iter_60300.pdz # model parameters and optimizer states
+└── speech_stats.npy        # statistics used to normalize spectrogram when training Tacotron2
+```
+You can use the following scripts to synthesize for `${BIN_DIR}/../sentences_en.txt` using pretrained Tacotron2 and parallel wavegan models.
+```bash
+source path.sh
-2. This model does not have a stop token predictor. It uses the attention peak position to decide whether all the contents have been uttered. Also, guided attention loss is used to speed up training. This model is trained with `configs/alternative.yaml`.[tacotron2_ljspeech_ckpt_0.3_alternative.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.3_alternative.zip)
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize_e2e.py \
+  --am=tacotron2_ljspeech \
+  --am_config=tacotron2_ljspeech_ckpt_0.2.0/default.yaml \
+  --am_ckpt=tacotron2_ljspeech_ckpt_0.2.0/snapshot_iter_60300.pdz \
+  --am_stat=tacotron2_ljspeech_ckpt_0.2.0/speech_stats.npy  \
+  --voc=pwgan_ljspeech\
+  --voc_config=pwg_ljspeech_ckpt_0.5/pwg_default.yaml \
+  --voc_ckpt=pwg_ljspeech_ckpt_0.5/pwg_snapshot_iter_400000.pdz  \
+  --voc_stat=pwg_ljspeech_ckpt_0.5/pwg_stats.npy \
+  --lang=en \
+  --text=${BIN_DIR}/../sentences_en.txt \
+  --output_dir=exp/default/test_e2e \
+  --phones_dict=tacotron2_ljspeech_ckpt_0.2.0/phone_id_map.txt
+```
--- a/examples/ljspeech/tts0/conf/default.yaml
+++ b/examples/ljspeech/tts0/conf/default.yaml
+# This configuration is for Paddle to train Tacotron 2. Compared to the
+# original paper, this configuration additionally use the guided attention
+# loss to accelerate the learning of the diagonal attention. It requires
+# only a single GPU with 12 GB memory and it takes ~1 days to finish the
+# training on Titan V.
+###########################################################
+#                FEATURE EXTRACTION SETTING               #
+###########################################################
+fs: 22050                # Sampling rate.
+n_fft: 1024              # FFT size (samples).
+n_shift: 256             # Hop size (samples). 11.6ms
+win_length: null         # Window length (samples).
+                         # If set to null, it will be the same as fft_size.
+window: "hann"           # Window function.
+n_mels: 80               # Number of mel basis.
+fmin: 80                 # Minimum freq in mel basis calculation. (Hz)
+fmax: 7600               # Maximum frequency in mel basis calculation. (Hz)
+###########################################################
+#                       DATA SETTING                      #
+###########################################################
+batch_size: 64
+num_workers: 2
+###########################################################
+#                       MODEL SETTING                     #
+###########################################################
+model:                          # keyword arguments for the selected model
+    embed_dim: 512               # char or phn embedding dimension
+    elayers: 1                   # number of blstm layers in encoder
+    eunits: 512                  # number of blstm units
+    econv_layers: 3              # number of convolutional layers in encoder
+    econv_chans: 512             # number of channels in convolutional layer
+    econv_filts: 5               # filter size of convolutional layer
+    atype: location              # attention function type
+    adim: 512                    # attention dimension
+    aconv_chans: 32              # number of channels in convolutional layer of attention
+    aconv_filts: 15              # filter size of convolutional layer of attention
+    cumulate_att_w: True         # whether to cumulate attention weight
+    dlayers: 2                   # number of lstm layers in decoder
+    dunits: 1024                 # number of lstm units in decoder
+    prenet_layers: 2             # number of layers in prenet
+    prenet_units: 256            # number of units in prenet
+    postnet_layers: 5            # number of layers in postnet
+    postnet_chans: 512           # number of channels in postnet
+    postnet_filts: 5             # filter size of postnet layer
+    output_activation: null      # activation function for the final output
+    use_batch_norm: True         # whether to use batch normalization in encoder
+    use_concate: True            # whether to concatenate encoder embedding with decoder outputs
+    use_residual: False          # whether to use residual connection in encoder
+    dropout_rate: 0.5            # dropout rate
+    zoneout_rate: 0.1            # zoneout rate
+    reduction_factor: 1          # reduction factor
+    spk_embed_dim: null          # speaker embedding dimension
+###########################################################
+#                       UPDATER SETTING                   #
+###########################################################
+updater:
+    use_masking: True            # whether to apply masking for padded part in loss calculation
+    bce_pos_weight: 5.0          # weight of positive sample in binary cross entropy calculation
+    use_guided_attn_loss: True   # whether to use guided attention loss
+    guided_attn_loss_sigma: 0.4  # sigma of guided attention loss
+    guided_attn_loss_lambda: 1.0 # strength of guided attention loss
+##########################################################
+#                  OPTIMIZER SETTING                     #
+##########################################################
+optimizer:
+    optim: adam              # optimizer type
+    learning_rate: 1.0e-03   # learning rate
+    epsilon: 1.0e-06         # epsilon
+    weight_decay: 0.0        # weight decay coefficient
+###########################################################
+#                     TRAINING SETTING                    #
+###########################################################
+max_epoch: 300
+num_snapshots: 5
+###########################################################
+#                       OTHER SETTING                     #
+###########################################################
+seed: 42
--- a/examples/ljspeech/tts0/local/preprocess.sh
+++ b/examples/ljspeech/tts0/local/preprocess.sh
 #!/bin/bash
-preprocess_path=$1
+stage=0
+stop_stage=100
-python3 ${BIN_DIR}/preprocess.py \
+config_path=$1
-    --input=~/datasets/LJSpeech-1.1 \
-    --output=${preprocess_path} \
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
-    -v  \
+    # get durations from MFA's result
\ No newline at end of file
+    echo "Generate durations.txt from MFA results ..."
+    python3 ${MAIN_ROOT}/utils/gen_duration_from_textgrid.py \
+        --inputdir=./ljspeech_alignment \
+        --output=durations.txt \
+        --config=${config_path}
+fi
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
+    # extract features
+    echo "Extract features ..."
+    python3 ${BIN_DIR}/preprocess.py \
+        --dataset=ljspeech \
+        --rootdir=~/datasets/LJSpeech-1.1/ \
+        --dumpdir=dump \
+        --dur-file=durations.txt \
+        --config=${config_path} \
+        --num-cpu=20 \
+        --cut-sil=True
+fi
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    # get features' stats(mean and std)
+    echo "Get features' stats ..."
+    python3 ${MAIN_ROOT}/utils/compute_statistics.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --field-name="speech"
+fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # normalize and covert phone to id, dev and test should use train's stats
+    echo "Normalize ..."
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/train/raw/metadata.jsonl \
+        --dumpdir=dump/train/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/dev/raw/metadata.jsonl \
+        --dumpdir=dump/dev/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+    python3 ${BIN_DIR}/normalize.py \
+        --metadata=dump/test/raw/metadata.jsonl \
+        --dumpdir=dump/test/norm \
+        --speech-stats=dump/train/speech_stats.npy \
+        --phones-dict=dump/phone_id_map.txt \
+        --speaker-dict=dump/speaker_id_map.txt
+fi
--- a/examples/ljspeech/tts0/local/synthesize.sh
+++ b/examples/ljspeech/tts0/local/synthesize.sh
 #!/bin/bash
-train_output_path=$1
+config_path=$1
-ckpt_name=$2
+train_output_path=$2
+ckpt_name=$3
-python3 ${BIN_DIR}/synthesize.py \
+FLAGS_allocator_strategy=naive_best_fit \
-    --config=${train_output_path}/config.yaml \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
-    --checkpoint_path=${train_output_path}/checkpoints/${ckpt_name} \
+python3 ${BIN_DIR}/../synthesize.py \
-    --input=${BIN_DIR}/../sentences_en.txt \
+    --am=tacotron2_ljspeech \
-    --output=${train_output_path}/test \
+    --am_config=${config_path} \
-    --ngpu=1
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_stat=dump/train/speech_stats.npy \
+    --voc=pwgan_ljspeech \
+    --voc_config=pwg_ljspeech_ckpt_0.5/pwg_default.yaml \
+    --voc_ckpt=pwg_ljspeech_ckpt_0.5/pwg_snapshot_iter_400000.pdz  \
+    --voc_stat=pwg_ljspeech_ckpt_0.5/pwg_stats.npy \
+    --test_metadata=dump/test/norm/metadata.jsonl \
+    --output_dir=${train_output_path}/test \
+    --phones_dict=dump/phone_id_map.txt
--- a/examples/ljspeech/tts0/local/synthesize_e2e.sh
+++ b/examples/ljspeech/tts0/local/synthesize_e2e.sh
+#!/bin/bash
+config_path=$1
+train_output_path=$2
+ckpt_name=$3
+# TODO: dygraph to static graph is not good for tacotron2_ljspeech now
+FLAGS_allocator_strategy=naive_best_fit \
+FLAGS_fraction_of_gpu_memory_to_use=0.01 \
+python3 ${BIN_DIR}/../synthesize_e2e.py \
+    --am=tacotron2_ljspeech \
+    --am_config=${config_path} \
+    --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
+    --am_stat=dump/train/speech_stats.npy \
+    --voc=pwgan_ljspeech \
+    --voc_config=pwg_ljspeech_ckpt_0.5/pwg_default.yaml \
+    --voc_ckpt=pwg_ljspeech_ckpt_0.5/pwg_snapshot_iter_400000.pdz  \
+    --voc_stat=pwg_ljspeech_ckpt_0.5/pwg_stats.npy \
+    --lang=en \
+    --text=${BIN_DIR}/../sentences_en.txt \
+    --output_dir=${train_output_path}/test_e2e \
+    --phones_dict=dump/phone_id_map.txt \
+    # --inference_dir=${train_output_path}/inference
\ No newline at end of file
--- a/examples/ljspeech/tts0/local/train.sh
+++ b/examples/ljspeech/tts0/local/train.sh
 #!/bin/bash
-preprocess_path=$1
+config_path=$1
 train_output_path=$2
 python3 ${BIN_DIR}/train.py \
-    --data=${preprocess_path} \
+    --train-metadata=dump/train/norm/metadata.jsonl \
-    --output=${train_output_path} \
+    --dev-metadata=dump/dev/norm/metadata.jsonl \
-    --ngpu=1 \
+    --config=${config_path} \
\ No newline at end of file
+    --output-dir=${train_output_path} \
+    --ngpu=1 \
+    --phones-dict=dump/phone_id_map.txt
\ No newline at end of file
--- a/examples/ljspeech/tts0/run.sh
+++ b/examples/ljspeech/tts0/run.sh
@@ -3,13 +3,13 @@
 set -e
 source path.sh
-gpus=0
+gpus=0,1
 stage=0
 stop_stage=100
-preprocess_path=preprocessed_ljspeech
+conf_path=conf/default.yaml
-train_output_path=output
+train_output_path=exp/default
-ckpt_name=step-35000
+ckpt_name=snapshot_iter_201.pdz
 # with the following command, you can choose the stage range you want to run
 # such as `./run.sh --stage 0 --stop-stage 0`
@@ -18,16 +18,20 @@ source ${MAIN_ROOT}/utils/parse_options.sh || exit 1
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    # prepare data
-    ./local/preprocess.sh ${preprocess_path} || exit -1
+    ./local/preprocess.sh ${conf_path} || exit -1
 fi
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
    # train model, all `ckpt` under `train_output_path/checkpoints/` dir
-    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${preprocess_path} ${train_output_path} || exit -1
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/train.sh ${conf_path} ${train_output_path} || exit -1
 fi
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    # train model, all `ckpt` under `train_output_path/checkpoints/` dir
+    # synthesize, vocoder is pwgan
-    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${train_output_path} ${ckpt_name} || exit -1
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
 fi
+if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
+    # synthesize_e2e, vocoder is pwgan
+    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize_e2e.sh ${conf_path} ${train_output_path} ${ckpt_name} || exit -1
+fi
--- a/examples/ljspeech/tts1/conf/default.yaml
+++ b/examples/ljspeech/tts1/conf/default.yaml
@@ -63,9 +63,9 @@ model:                     # keyword arguments for the selected model
 #                       UPDATER SETTING                   #
 ###########################################################
 updater:
-    use_masking: true                  # whether to apply masking for padded part in loss calculation
+    use_masking: True                  # whether to apply masking for padded part in loss calculation
    loss_type: L1
-    use_guided_attn_loss: true         # whether to use guided attention loss
+    use_guided_attn_loss: True         # whether to use guided attention loss
    guided_attn_loss_sigma: 0.4        # sigma in guided attention loss
    guided_attn_loss_lambda: 10.0      # lambda in guided attention loss
    modules_applied_guided_attn: ["encoder-decoder"] # modules to apply guided attention loss

--- a/examples/ljspeech/tts3/README.md
+++ b/examples/ljspeech/tts3/README.md
-# FastSpeech2 with the LJSpeech-1.1
+# FastSpeech2 with LJSpeech-1.1
 This example contains code used to train a [Fastspeech2](https://arxiv.org/abs/2006.04558) model with [LJSpeech-1.1](https://keithito.com/LJ-Speech-Dataset/).
 ## Dataset

--- a/examples/ljspeech/tts3/conf/default.yaml
+++ b/examples/ljspeech/tts3/conf/default.yaml
@@ -16,8 +16,8 @@ fmax: 7600         # Maximum frequency of Mel basis.
 n_mels: 80         # The number of mel basis.
 # Only used for the model using pitch features (e.g. FastSpeech2)
-f0min: 80          # Maximum f0 for pitch extraction.
+f0min: 80          # Minimum f0 for pitch extraction.
-f0max: 400         # Minimum f0 for pitch extraction.
+f0max: 400         # Maximum f0 for pitch extraction.
 ###########################################################
@@ -64,14 +64,14 @@ model:
    pitch_predictor_dropout: 0.5               # dropout rate in pitch predictor
    pitch_embed_kernel_size: 1                 # kernel size of conv embedding layer for pitch
    pitch_embed_dropout: 0.0                   # dropout rate after conv embedding layer for pitch
-    stop_gradient_from_pitch_predictor: true   # whether to stop the gradient from pitch predictor to encoder
+    stop_gradient_from_pitch_predictor: True   # whether to stop the gradient from pitch predictor to encoder
    energy_predictor_layers: 2                 # number of conv layers in energy predictor
    energy_predictor_chans: 256                # number of channels of conv layers in energy predictor
    energy_predictor_kernel_size: 3            # kernel size of conv leyers in energy predictor
    energy_predictor_dropout: 0.5              # dropout rate in energy predictor
    energy_embed_kernel_size: 1                # kernel size of conv embedding layer for energy
    energy_embed_dropout: 0.0                  # dropout rate after conv embedding layer for energy
-    stop_gradient_from_energy_predictor: false # whether to stop the gradient from energy predictor to encoder
+    stop_gradient_from_energy_predictor: False # whether to stop the gradient from energy predictor to encoder

--- a/examples/ljspeech/voc0/run.sh
+++ b/examples/ljspeech/voc0/run.sh
@@ -10,7 +10,7 @@ stop_stage=100
 preprocess_path=preprocessed_ljspeech
 train_output_path=output
 # mel generated by Tacotron2
-input_mel_path=../tts0/output/test
+input_mel_path=${preprocess_path}/mel_test
 ckpt_name=step-10000
 # with the following command, you can choose the stage range you want to run
@@ -28,5 +28,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
 fi
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
+    mkdir -p ${preprocess_path}/mel_test
+    cp ${preprocess_path}/mel/LJ050-001*.npy ${preprocess_path}/mel_test/
    CUDA_VISIBLE_DEVICES=${gpus} ./local/synthesize.sh ${input_mel_path} ${train_output_path} ${ckpt_name} || exit -1
 fi
--- a/examples/ljspeech/voc1/conf/default.yaml
+++ b/examples/ljspeech/voc1/conf/default.yaml
--- a/examples/other/1xt2x/src_deepspeech2x/models/ds2/deepspeech2.py
+++ b/examples/other/1xt2x/src_deepspeech2x/models/ds2/deepspeech2.py
--- a/examples/other/1xt2x/src_deepspeech2x/test_model.py
+++ b/examples/other/1xt2x/src_deepspeech2x/test_model.py
--- a/examples/other/g2p/run.sh
+++ b/examples/other/g2p/run.sh
--- a/examples/ted_en_zh/st0/conf/tuning/decode.yaml
+++ b/examples/ted_en_zh/st0/conf/tuning/decode.yaml
--- a/examples/ted_en_zh/st1/conf/tuning/decode.yaml
+++ b/examples/ted_en_zh/st1/conf/tuning/decode.yaml
--- a/examples/thchs30/align0/README.md
+++ b/examples/thchs30/align0/README.md
--- a/examples/vctk/tts3/README.md
+++ b/examples/vctk/tts3/README.md
--- a/examples/vctk/tts3/conf/default.yaml
+++ b/examples/vctk/tts3/conf/default.yaml
--- a/examples/vctk/tts3/local/inference.sh
+++ b/examples/vctk/tts3/local/inference.sh
--- a/examples/vctk/tts3/local/synthesize_e2e.sh
+++ b/examples/vctk/tts3/local/synthesize_e2e.sh
--- a/examples/vctk/voc1/conf/default.yaml
+++ b/examples/vctk/voc1/conf/default.yaml
--- a/examples/voxceleb/README.md
+++ b/examples/voxceleb/README.md
--- a/examples/voxceleb/sv0/local/make_voxceleb_kaldi_trial.py
+++ b/examples/voxceleb/sv0/local/make_voxceleb_kaldi_trial.py
--- a/paddleaudio/CHANGELOG.md
+++ b/paddleaudio/CHANGELOG.md
--- a/paddleaudio/__init__.py
+++ b/paddleaudio/__init__.py
--- a/paddleaudio/features/core.py
+++ b/paddleaudio/features/core.py
--- a/paddlespeech/__init__.py
+++ b/paddlespeech/__init__.py
--- a/paddlespeech/cli/asr/infer.py
+++ b/paddlespeech/cli/asr/infer.py
--- a/paddlespeech/cli/cls/infer.py
+++ b/paddlespeech/cli/cls/infer.py
--- a/paddlespeech/cli/st/infer.py
+++ b/paddlespeech/cli/st/infer.py
--- a/paddlespeech/cli/text/infer.py
+++ b/paddlespeech/cli/text/infer.py
--- a/paddlespeech/cli/tts/infer.py
+++ b/paddlespeech/cli/tts/infer.py
--- a/paddlespeech/cli/utils.py
+++ b/paddlespeech/cli/utils.py
--- a/paddlespeech/s2t/decoders/ctcdecoder/__init__.py
+++ b/paddlespeech/s2t/decoders/ctcdecoder/__init__.py
--- a/paddlespeech/s2t/decoders/ctcdecoder/swig_wrapper.py
+++ b/paddlespeech/s2t/decoders/ctcdecoder/swig_wrapper.py
--- a/paddlespeech/s2t/exps/deepspeech2/model.py
+++ b/paddlespeech/s2t/exps/deepspeech2/model.py
--- a/paddlespeech/s2t/exps/u2/model.py
+++ b/paddlespeech/s2t/exps/u2/model.py
--- a/paddlespeech/s2t/exps/u2_st/model.py
+++ b/paddlespeech/s2t/exps/u2_st/model.py
--- a/paddlespeech/s2t/io/batchfy.py
+++ b/paddlespeech/s2t/io/batchfy.py
--- a/paddlespeech/s2t/io/dataloader.py
+++ b/paddlespeech/s2t/io/dataloader.py
--- a/paddlespeech/s2t/io/sampler.py
+++ b/paddlespeech/s2t/io/sampler.py
--- a/paddlespeech/s2t/models/ds2/__init__.py
+++ b/paddlespeech/s2t/models/ds2/__init__.py
--- a/paddlespeech/s2t/models/ds2/deepspeech2.py
+++ b/paddlespeech/s2t/models/ds2/deepspeech2.py
--- a/paddlespeech/s2t/models/ds2_online/__init__.py
+++ b/paddlespeech/s2t/models/ds2_online/__init__.py
--- a/paddlespeech/s2t/models/ds2_online/deepspeech2.py
+++ b/paddlespeech/s2t/models/ds2_online/deepspeech2.py
--- a/paddlespeech/s2t/models/u2/u2.py
+++ b/paddlespeech/s2t/models/u2/u2.py
--- a/paddlespeech/s2t/models/u2_st/u2_st.py
+++ b/paddlespeech/s2t/models/u2_st/u2_st.py
--- a/paddlespeech/s2t/modules/ctc.py
+++ b/paddlespeech/s2t/modules/ctc.py
--- a/paddlespeech/s2t/training/trainer.py
+++ b/paddlespeech/s2t/training/trainer.py
--- a/paddlespeech/s2t/transform/perturb.py
+++ b/paddlespeech/s2t/transform/perturb.py
--- a/paddlespeech/s2t/transform/spectrogram.py
+++ b/paddlespeech/s2t/transform/spectrogram.py
--- a/paddlespeech/t2s/__init__.py
+++ b/paddlespeech/t2s/__init__.py
--- a/paddlespeech/t2s/audio/__init__.py
+++ b/paddlespeech/t2s/audio/__init__.py
--- a/paddlespeech/t2s/audio/audio.py
+++ b/paddlespeech/t2s/audio/audio.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/chinese_g2p.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/chinese_g2p.py
--- a/paddlespeech/t2s/datasets/__init__.py
+++ b/paddlespeech/t2s/datasets/__init__.py
--- a/paddlespeech/t2s/datasets/am_batch_fn.py
+++ b/paddlespeech/t2s/datasets/am_batch_fn.py
--- a/paddlespeech/t2s/data/batch.py
+++ b/paddlespeech/t2s/data/batch.py
--- a/paddlespeech/t2s/datasets/common.py
+++ b/paddlespeech/t2s/datasets/common.py
--- a/paddlespeech/t2s/datasets/data_table.py
+++ b/paddlespeech/t2s/datasets/data_table.py
--- a/paddlespeech/t2s/data/dataset.py
+++ b/paddlespeech/t2s/data/dataset.py
--- a/paddlespeech/t2s/data/get_feats.py
+++ b/paddlespeech/t2s/data/get_feats.py
--- a/paddlespeech/t2s/datasets/preprocess_utils.py
+++ b/paddlespeech/t2s/datasets/preprocess_utils.py
--- a/paddlespeech/t2s/datasets/vocoder_batch_fn.py
+++ b/paddlespeech/t2s/datasets/vocoder_batch_fn.py
--- a/paddlespeech/t2s/exps/fastspeech2/gen_gta_mel.py
+++ b/paddlespeech/t2s/exps/fastspeech2/gen_gta_mel.py
--- a/paddlespeech/t2s/exps/fastspeech2/preprocess.py
+++ b/paddlespeech/t2s/exps/fastspeech2/preprocess.py
--- a/paddlespeech/t2s/exps/fastspeech2/train.py
+++ b/paddlespeech/t2s/exps/fastspeech2/train.py
--- a/paddlespeech/t2s/exps/gan_vocoder/hifigan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/hifigan/train.py
--- a/paddlespeech/t2s/exps/gan_vocoder/multi_band_melgan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/multi_band_melgan/train.py
--- a/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/synthesize_from_wav.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/synthesize_from_wav.py
--- a/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/train.py
--- a/paddlespeech/t2s/exps/gan_vocoder/preprocess.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/preprocess.py
--- a/paddlespeech/t2s/exps/gan_vocoder/style_melgan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/style_melgan/train.py
--- a/paddlespeech/t2s/exps/inference.py
+++ b/paddlespeech/t2s/exps/inference.py
--- a/paddlespeech/t2s/exps/speedyspeech/gen_gta_mel.py
+++ b/paddlespeech/t2s/exps/speedyspeech/gen_gta_mel.py
--- a/paddlespeech/t2s/exps/speedyspeech/normalize.py
+++ b/paddlespeech/t2s/exps/speedyspeech/normalize.py
--- a/paddlespeech/t2s/exps/speedyspeech/preprocess.py
+++ b/paddlespeech/t2s/exps/speedyspeech/preprocess.py
--- a/paddlespeech/t2s/exps/speedyspeech/train.py
+++ b/paddlespeech/t2s/exps/speedyspeech/train.py
--- a/paddlespeech/t2s/exps/synthesize.py
+++ b/paddlespeech/t2s/exps/synthesize.py
--- a/paddlespeech/t2s/exps/synthesize_e2e.py
+++ b/paddlespeech/t2s/exps/synthesize_e2e.py
--- a/paddlespeech/t2s/exps/tacotron2/config.py
+++ b/paddlespeech/t2s/exps/tacotron2/config.py
--- a/paddlespeech/t2s/exps/tacotron2/ljspeech.py
+++ b/paddlespeech/t2s/exps/tacotron2/ljspeech.py
--- a/paddlespeech/t2s/exps/tacotron2/normalize.py
+++ b/paddlespeech/t2s/exps/tacotron2/normalize.py
--- a/paddlespeech/t2s/exps/tacotron2/preprocess.py
+++ b/paddlespeech/t2s/exps/tacotron2/preprocess.py
--- a/paddlespeech/t2s/exps/tacotron2/synthesize.ipynb
+++ b/paddlespeech/t2s/exps/tacotron2/synthesize.ipynb
--- a/paddlespeech/t2s/exps/tacotron2/synthesize.py
+++ b/paddlespeech/t2s/exps/tacotron2/synthesize.py
--- a/paddlespeech/t2s/exps/tacotron2/train.py
+++ b/paddlespeech/t2s/exps/tacotron2/train.py
--- a/paddlespeech/t2s/exps/transformer_tts/normalize.py
+++ b/paddlespeech/t2s/exps/transformer_tts/normalize.py
--- a/paddlespeech/t2s/exps/transformer_tts/preprocess.py
+++ b/paddlespeech/t2s/exps/transformer_tts/preprocess.py
--- a/paddlespeech/t2s/exps/transformer_tts/train.py
+++ b/paddlespeech/t2s/exps/transformer_tts/train.py
--- a/paddlespeech/t2s/exps/fastspeech2/voice_cloning.py
+++ b/paddlespeech/t2s/exps/fastspeech2/voice_cloning.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/aishell3.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/aishell3.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/config.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/config.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/extract_mel.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/extract_mel.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/lexicon.txt
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/lexicon.txt
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/preprocess_transcription.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/preprocess_transcription.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/process_wav.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/process_wav.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/train.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/train.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/voice_cloning.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/voice_cloning.py
--- a/paddlespeech/t2s/exps/waveflow/ljspeech.py
+++ b/paddlespeech/t2s/exps/waveflow/ljspeech.py
--- a/paddlespeech/t2s/exps/waveflow/train.py
+++ b/paddlespeech/t2s/exps/waveflow/train.py
--- a/paddlespeech/t2s/exps/voice_cloning/__init__.py
+++ b/paddlespeech/t2s/exps/voice_cloning/__init__.py
--- a/paddlespeech/t2s/exps/wavernn/synthesize.py
+++ b/paddlespeech/t2s/exps/wavernn/synthesize.py
--- a/paddlespeech/t2s/exps/wavernn/train.py
+++ b/paddlespeech/t2s/exps/wavernn/train.py
--- a/paddlespeech/t2s/frontend/arpabet.py
+++ b/paddlespeech/t2s/frontend/arpabet.py
--- a/paddlespeech/t2s/frontend/phonectic.py
+++ b/paddlespeech/t2s/frontend/phonectic.py
--- a/paddlespeech/t2s/frontend/vocab.py
+++ b/paddlespeech/t2s/frontend/vocab.py
--- a/paddlespeech/t2s/frontend/zh_normalization/chronology.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/chronology.py
--- a/paddlespeech/t2s/frontend/zh_normalization/num.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/num.py
--- a/paddlespeech/t2s/frontend/zh_normalization/phonecode.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/phonecode.py
--- a/paddlespeech/t2s/frontend/zh_normalization/quantifier.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/quantifier.py
--- a/paddlespeech/t2s/frontend/zh_normalization/text_normlization.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/text_normlization.py
--- a/paddlespeech/t2s/models/__init__.py
+++ b/paddlespeech/t2s/models/__init__.py
--- a/paddlespeech/t2s/models/fastspeech2/fastspeech2.py
+++ b/paddlespeech/t2s/models/fastspeech2/fastspeech2.py
--- a/paddlespeech/t2s/models/fastspeech2/fastspeech2_updater.py
+++ b/paddlespeech/t2s/models/fastspeech2/fastspeech2_updater.py
--- a/paddlespeech/t2s/models/hifigan/hifigan.py
+++ b/paddlespeech/t2s/models/hifigan/hifigan.py
--- a/paddlespeech/t2s/models/melgan/melgan.py
+++ b/paddlespeech/t2s/models/melgan/melgan.py
--- a/paddlespeech/t2s/models/melgan/style_melgan.py
+++ b/paddlespeech/t2s/models/melgan/style_melgan.py
--- a/paddlespeech/t2s/models/parallel_wavegan/parallel_wavegan.py
+++ b/paddlespeech/t2s/models/parallel_wavegan/parallel_wavegan.py
--- a/paddlespeech/t2s/models/speedyspeech/speedyspeech.py
+++ b/paddlespeech/t2s/models/speedyspeech/speedyspeech.py
--- a/paddlespeech/t2s/models/speedyspeech/speedyspeech_updater.py
+++ b/paddlespeech/t2s/models/speedyspeech/speedyspeech_updater.py
--- a/paddlespeech/t2s/models/tacotron2.py
+++ b/paddlespeech/t2s/models/tacotron2.py
--- a/paddlespeech/t2s/data/__init__.py
+++ b/paddlespeech/t2s/data/__init__.py
--- a/paddlespeech/t2s/models/tacotron2/tacotron2.py
+++ b/paddlespeech/t2s/models/tacotron2/tacotron2.py
--- a/paddlespeech/t2s/models/tacotron2/tacotron2_updater.py
+++ b/paddlespeech/t2s/models/tacotron2/tacotron2_updater.py
--- a/paddlespeech/t2s/models/transformer_tts/transformer_tts.py
+++ b/paddlespeech/t2s/models/transformer_tts/transformer_tts.py
--- a/paddlespeech/t2s/models/transformer_tts/transformer_tts_updater.py
+++ b/paddlespeech/t2s/models/transformer_tts/transformer_tts_updater.py
--- a/paddlespeech/t2s/models/waveflow.py
+++ b/paddlespeech/t2s/models/waveflow.py
--- a/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/__init__.py
+++ b/paddlespeech/t2s/exps/voice_cloning/tacotron2_ge2e/__init__.py
--- a/paddlespeech/t2s/models/wavernn/wavernn.py
+++ b/paddlespeech/t2s/models/wavernn/wavernn.py
--- a/paddlespeech/t2s/models/wavernn/wavernn_updater.py
+++ b/paddlespeech/t2s/models/wavernn/wavernn_updater.py
--- a/paddlespeech/t2s/modules/causal_conv.py
+++ b/paddlespeech/t2s/modules/causal_conv.py
--- a/paddlespeech/t2s/modules/conformer/convolution.py
+++ b/paddlespeech/t2s/modules/conformer/convolution.py
--- a/paddlespeech/t2s/modules/conformer/encoder_layer.py
+++ b/paddlespeech/t2s/modules/conformer/encoder_layer.py
--- a/paddlespeech/t2s/modules/conv.py
+++ b/paddlespeech/t2s/modules/conv.py
--- a/paddlespeech/t2s/modules/geometry.py
+++ b/paddlespeech/t2s/modules/geometry.py
--- a/paddlespeech/t2s/modules/layer_norm.py
+++ b/paddlespeech/t2s/modules/layer_norm.py
--- a/paddlespeech/t2s/modules/losses.py
+++ b/paddlespeech/t2s/modules/losses.py
--- a/paddlespeech/t2s/modules/nets_utils.py
+++ b/paddlespeech/t2s/modules/nets_utils.py
--- a/paddlespeech/t2s/modules/pqmf.py
+++ b/paddlespeech/t2s/modules/pqmf.py
--- a/paddlespeech/t2s/modules/predictor/duration_predictor.py
+++ b/paddlespeech/t2s/modules/predictor/duration_predictor.py
--- a/paddlespeech/t2s/modules/predictor/length_regulator.py
+++ b/paddlespeech/t2s/modules/predictor/length_regulator.py
--- a/paddlespeech/t2s/modules/predictor/variance_predictor.py
+++ b/paddlespeech/t2s/modules/predictor/variance_predictor.py
--- a/paddlespeech/t2s/modules/residual_block.py
+++ b/paddlespeech/t2s/modules/residual_block.py
--- a/paddlespeech/t2s/modules/residual_stack.py
+++ b/paddlespeech/t2s/modules/residual_stack.py
--- a/paddlespeech/t2s/modules/style_encoder.py
+++ b/paddlespeech/t2s/modules/style_encoder.py
--- a/paddlespeech/t2s/modules/tacotron2/attentions.py
+++ b/paddlespeech/t2s/modules/tacotron2/attentions.py
--- a/paddlespeech/t2s/modules/tacotron2/decoder.py
+++ b/paddlespeech/t2s/modules/tacotron2/decoder.py
--- a/paddlespeech/t2s/modules/tacotron2/encoder.py
+++ b/paddlespeech/t2s/modules/tacotron2/encoder.py
--- a/paddlespeech/t2s/modules/tade_res_block.py
+++ b/paddlespeech/t2s/modules/tade_res_block.py
--- a/paddlespeech/t2s/modules/transformer/attention.py
+++ b/paddlespeech/t2s/modules/transformer/attention.py
--- a/paddlespeech/t2s/modules/transformer/decoder.py
+++ b/paddlespeech/t2s/modules/transformer/decoder.py
--- a/paddlespeech/t2s/modules/transformer/decoder_layer.py
+++ b/paddlespeech/t2s/modules/transformer/decoder_layer.py
--- a/paddlespeech/t2s/modules/transformer/embedding.py
+++ b/paddlespeech/t2s/modules/transformer/embedding.py
--- a/paddlespeech/t2s/modules/transformer/encoder.py
+++ b/paddlespeech/t2s/modules/transformer/encoder.py
--- a/paddlespeech/t2s/modules/transformer/encoder_layer.py
+++ b/paddlespeech/t2s/modules/transformer/encoder_layer.py
--- a/paddlespeech/t2s/modules/transformer/lightconv.py
+++ b/paddlespeech/t2s/modules/transformer/lightconv.py
--- a/paddlespeech/t2s/modules/transformer/mask.py
+++ b/paddlespeech/t2s/modules/transformer/mask.py
--- a/paddlespeech/t2s/modules/transformer/multi_layer_conv.py
+++ b/paddlespeech/t2s/modules/transformer/multi_layer_conv.py
--- a/paddlespeech/t2s/modules/transformer/positionwise_feed_forward.py
+++ b/paddlespeech/t2s/modules/transformer/positionwise_feed_forward.py
--- a/paddlespeech/t2s/modules/transformer/repeat.py
+++ b/paddlespeech/t2s/modules/transformer/repeat.py
--- a/paddlespeech/t2s/modules/transformer/subsampling.py
+++ b/paddlespeech/t2s/modules/transformer/subsampling.py
--- a/paddlespeech/t2s/modules/upsample.py
+++ b/paddlespeech/t2s/modules/upsample.py
--- a/paddlespeech/t2s/training/experiment.py
+++ b/paddlespeech/t2s/training/experiment.py
--- a/paddlespeech/t2s/training/extensions/snapshot.py
+++ b/paddlespeech/t2s/training/extensions/snapshot.py
--- a/paddlespeech/t2s/training/optimizer.py
+++ b/paddlespeech/t2s/training/optimizer.py
--- a/paddlespeech/t2s/utils/__init__.py
+++ b/paddlespeech/t2s/utils/__init__.py
--- a/paddlespeech/t2s/utils/error_rate.py
+++ b/paddlespeech/t2s/utils/error_rate.py
--- a/paddlespeech/t2s/utils/h5_utils.py
+++ b/paddlespeech/t2s/utils/h5_utils.py
--- a/paddlespeech/text/exps/ernie_linear/train.py
+++ b/paddlespeech/text/exps/ernie_linear/train.py
--- a/paddlespeech/vector/exps/ge2e/audio_processor.py
+++ b/paddlespeech/vector/exps/ge2e/audio_processor.py
--- a/paddlespeech/vector/exps/ge2e/speaker_verification_dataset.py
+++ b/paddlespeech/vector/exps/ge2e/speaker_verification_dataset.py
--- a/setup.py
+++ b/setup.py
--- a/setup_audio.py
+++ b/setup_audio.py
--- a/speechx/CMakeLists.txt
+++ b/speechx/CMakeLists.txt
--- a/speechx/docker/.gitkeep
+++ b/speechx/docker/.gitkeep
--- a/speechx/examples/.gitkeep
+++ b/speechx/examples/.gitkeep
--- a/speechx/speechx/CMakeLists.txt
+++ b/speechx/speechx/CMakeLists.txt
--- a/speechx/speechx/base/basic_types.h
+++ b/speechx/speechx/base/basic_types.h
--- a/speechx/speechx/base/macros.h
+++ b/speechx/speechx/base/macros.h
--- a/speechx/speechx/codelab/README.md
+++ b/speechx/speechx/codelab/README.md
--- a/speechx/speechx/codelab/feat_test/feature-mfcc-test.cc
+++ b/speechx/speechx/codelab/feat_test/feature-mfcc-test.cc
--- a/speechx/speechx/common/CMakeLists.txt
+++ b/speechx/speechx/common/CMakeLists.txt
--- a/speechx/speechx/decoder/CMakeLists.txt
+++ b/speechx/speechx/decoder/CMakeLists.txt
--- a/speechx/speechx/frontend/CMakeLists.txt
+++ b/speechx/speechx/frontend/CMakeLists.txt
--- a/speechx/speechx/frontend/audio/CMakeLists.txt
+++ b/speechx/speechx/frontend/audio/CMakeLists.txt
--- a/speechx/speechx/frontend/text/CMakeLists.txt
+++ b/speechx/speechx/frontend/text/CMakeLists.txt
--- a/speechx/speechx/kaldi/.gitkeep
+++ b/speechx/speechx/kaldi/.gitkeep
--- a/speechx/speechx/kaldi/CMakeLists.txt
+++ b/speechx/speechx/kaldi/CMakeLists.txt
--- a/speechx/speechx/kaldi/base/CMakeLists.txt
+++ b/speechx/speechx/kaldi/base/CMakeLists.txt
--- a/speechx/speechx/kaldi/base/io-funcs-inl.h
+++ b/speechx/speechx/kaldi/base/io-funcs-inl.h
--- a/speechx/speechx/kaldi/base/io-funcs.cc
+++ b/speechx/speechx/kaldi/base/io-funcs.cc
--- a/speechx/speechx/kaldi/base/io-funcs.h
+++ b/speechx/speechx/kaldi/base/io-funcs.h
--- a/speechx/speechx/kaldi/base/kaldi-common.h
+++ b/speechx/speechx/kaldi/base/kaldi-common.h
--- a/speechx/speechx/kaldi/base/kaldi-error.cc
+++ b/speechx/speechx/kaldi/base/kaldi-error.cc
--- a/speechx/speechx/kaldi/base/kaldi-error.h
+++ b/speechx/speechx/kaldi/base/kaldi-error.h
--- a/speechx/speechx/kaldi/base/kaldi-math.cc
+++ b/speechx/speechx/kaldi/base/kaldi-math.cc
--- a/speechx/speechx/kaldi/base/kaldi-math.h
+++ b/speechx/speechx/kaldi/base/kaldi-math.h
--- a/speechx/speechx/kaldi/base/kaldi-types.h
+++ b/speechx/speechx/kaldi/base/kaldi-types.h
--- a/speechx/speechx/kaldi/base/kaldi-utils.cc
+++ b/speechx/speechx/kaldi/base/kaldi-utils.cc
--- a/speechx/speechx/kaldi/base/kaldi-utils.h
+++ b/speechx/speechx/kaldi/base/kaldi-utils.h
--- a/speechx/speechx/kaldi/base/timer.cc
+++ b/speechx/speechx/kaldi/base/timer.cc
--- a/speechx/speechx/kaldi/base/timer.h
+++ b/speechx/speechx/kaldi/base/timer.h
--- a/speechx/speechx/kaldi/base/version.h
+++ b/speechx/speechx/kaldi/base/version.h
--- a/speechx/speechx/kaldi/feat/CMakeLists.txt
+++ b/speechx/speechx/kaldi/feat/CMakeLists.txt
--- a/speechx/speechx/kaldi/feat/feature-common-inl.h
+++ b/speechx/speechx/kaldi/feat/feature-common-inl.h
--- a/speechx/speechx/kaldi/feat/feature-common.h
+++ b/speechx/speechx/kaldi/feat/feature-common.h
--- a/speechx/speechx/kaldi/feat/feature-fbank.cc
+++ b/speechx/speechx/kaldi/feat/feature-fbank.cc
--- a/speechx/speechx/kaldi/feat/feature-fbank.h
+++ b/speechx/speechx/kaldi/feat/feature-fbank.h
--- a/speechx/speechx/kaldi/feat/feature-functions.cc
+++ b/speechx/speechx/kaldi/feat/feature-functions.cc
--- a/speechx/speechx/kaldi/feat/feature-functions.h
+++ b/speechx/speechx/kaldi/feat/feature-functions.h
--- a/speechx/speechx/kaldi/feat/feature-mfcc.cc
+++ b/speechx/speechx/kaldi/feat/feature-mfcc.cc
--- a/speechx/speechx/kaldi/feat/feature-mfcc.h
+++ b/speechx/speechx/kaldi/feat/feature-mfcc.h
--- a/speechx/speechx/kaldi/feat/feature-plp.cc
+++ b/speechx/speechx/kaldi/feat/feature-plp.cc
--- a/speechx/speechx/kaldi/feat/feature-plp.h
+++ b/speechx/speechx/kaldi/feat/feature-plp.h
--- a/speechx/speechx/kaldi/feat/feature-spectrogram.cc
+++ b/speechx/speechx/kaldi/feat/feature-spectrogram.cc
--- a/speechx/speechx/kaldi/feat/feature-spectrogram.h
+++ b/speechx/speechx/kaldi/feat/feature-spectrogram.h
--- a/speechx/speechx/kaldi/feat/feature-window.cc
+++ b/speechx/speechx/kaldi/feat/feature-window.cc
--- a/speechx/speechx/kaldi/feat/feature-window.h
+++ b/speechx/speechx/kaldi/feat/feature-window.h
--- a/speechx/speechx/kaldi/feat/mel-computations.cc
+++ b/speechx/speechx/kaldi/feat/mel-computations.cc
--- a/speechx/speechx/kaldi/feat/mel-computations.h
+++ b/speechx/speechx/kaldi/feat/mel-computations.h
--- a/speechx/speechx/kaldi/feat/online-feature.cc
+++ b/speechx/speechx/kaldi/feat/online-feature.cc
--- a/speechx/speechx/kaldi/feat/online-feature.h
+++ b/speechx/speechx/kaldi/feat/online-feature.h
--- a/speechx/speechx/kaldi/feat/pitch-functions.cc
+++ b/speechx/speechx/kaldi/feat/pitch-functions.cc
--- a/speechx/speechx/kaldi/feat/pitch-functions.h
+++ b/speechx/speechx/kaldi/feat/pitch-functions.h
--- a/speechx/speechx/kaldi/feat/resample.cc
+++ b/speechx/speechx/kaldi/feat/resample.cc
--- a/speechx/speechx/kaldi/feat/resample.h
+++ b/speechx/speechx/kaldi/feat/resample.h
--- a/speechx/speechx/kaldi/feat/signal.cc
+++ b/speechx/speechx/kaldi/feat/signal.cc
--- a/speechx/speechx/kaldi/feat/signal.h
+++ b/speechx/speechx/kaldi/feat/signal.h
--- a/speechx/speechx/kaldi/feat/wave-reader.cc
+++ b/speechx/speechx/kaldi/feat/wave-reader.cc
--- a/speechx/speechx/kaldi/feat/wave-reader.h
+++ b/speechx/speechx/kaldi/feat/wave-reader.h
--- a/speechx/speechx/kaldi/matrix/BUILD
+++ b/speechx/speechx/kaldi/matrix/BUILD
--- a/speechx/speechx/kaldi/matrix/CMakeLists.txt
+++ b/speechx/speechx/kaldi/matrix/CMakeLists.txt
--- a/speechx/speechx/kaldi/matrix/cblas-wrappers.h
+++ b/speechx/speechx/kaldi/matrix/cblas-wrappers.h
--- a/speechx/speechx/kaldi/matrix/compressed-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/compressed-matrix.cc
--- a/speechx/speechx/kaldi/matrix/compressed-matrix.h
+++ b/speechx/speechx/kaldi/matrix/compressed-matrix.h
--- a/speechx/speechx/kaldi/matrix/jama-eig.h
+++ b/speechx/speechx/kaldi/matrix/jama-eig.h
--- a/speechx/speechx/kaldi/matrix/jama-svd.h
+++ b/speechx/speechx/kaldi/matrix/jama-svd.h
--- a/speechx/speechx/kaldi/matrix/kaldi-blas.h
+++ b/speechx/speechx/kaldi/matrix/kaldi-blas.h
--- a/speechx/speechx/kaldi/matrix/kaldi-matrix-inl.h
+++ b/speechx/speechx/kaldi/matrix/kaldi-matrix-inl.h
--- a/speechx/speechx/kaldi/matrix/kaldi-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/kaldi-matrix.cc
--- a/speechx/speechx/kaldi/matrix/kaldi-matrix.h
+++ b/speechx/speechx/kaldi/matrix/kaldi-matrix.h
--- a/speechx/speechx/kaldi/matrix/kaldi-vector-inl.h
+++ b/speechx/speechx/kaldi/matrix/kaldi-vector-inl.h
--- a/speechx/speechx/kaldi/matrix/kaldi-vector.cc
+++ b/speechx/speechx/kaldi/matrix/kaldi-vector.cc
--- a/speechx/speechx/kaldi/matrix/kaldi-vector.h
+++ b/speechx/speechx/kaldi/matrix/kaldi-vector.h
--- a/speechx/speechx/kaldi/matrix/matrix-common.h
+++ b/speechx/speechx/kaldi/matrix/matrix-common.h
--- a/speechx/speechx/kaldi/matrix/matrix-functions-inl.h
+++ b/speechx/speechx/kaldi/matrix/matrix-functions-inl.h
--- a/speechx/speechx/kaldi/matrix/matrix-functions.cc
+++ b/speechx/speechx/kaldi/matrix/matrix-functions.cc
--- a/speechx/speechx/kaldi/matrix/matrix-functions.h
+++ b/speechx/speechx/kaldi/matrix/matrix-functions.h
--- a/speechx/speechx/kaldi/matrix/matrix-lib.h
+++ b/speechx/speechx/kaldi/matrix/matrix-lib.h
--- a/speechx/speechx/kaldi/matrix/optimization.cc
+++ b/speechx/speechx/kaldi/matrix/optimization.cc
--- a/speechx/speechx/kaldi/matrix/optimization.h
+++ b/speechx/speechx/kaldi/matrix/optimization.h
--- a/speechx/speechx/kaldi/matrix/packed-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/packed-matrix.cc
--- a/speechx/speechx/kaldi/matrix/packed-matrix.h
+++ b/speechx/speechx/kaldi/matrix/packed-matrix.h
--- a/speechx/speechx/kaldi/matrix/qr.cc
+++ b/speechx/speechx/kaldi/matrix/qr.cc
--- a/speechx/speechx/kaldi/matrix/sp-matrix-inl.h
+++ b/speechx/speechx/kaldi/matrix/sp-matrix-inl.h
--- a/speechx/speechx/kaldi/matrix/sp-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/sp-matrix.cc
--- a/speechx/speechx/kaldi/matrix/sp-matrix.h
+++ b/speechx/speechx/kaldi/matrix/sp-matrix.h
--- a/speechx/speechx/kaldi/matrix/sparse-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/sparse-matrix.cc
--- a/speechx/speechx/kaldi/matrix/sparse-matrix.h
+++ b/speechx/speechx/kaldi/matrix/sparse-matrix.h
--- a/speechx/speechx/kaldi/matrix/srfft.cc
+++ b/speechx/speechx/kaldi/matrix/srfft.cc
--- a/speechx/speechx/kaldi/matrix/srfft.h
+++ b/speechx/speechx/kaldi/matrix/srfft.h
--- a/speechx/speechx/kaldi/matrix/tp-matrix.cc
+++ b/speechx/speechx/kaldi/matrix/tp-matrix.cc
--- a/speechx/speechx/kaldi/matrix/tp-matrix.h
+++ b/speechx/speechx/kaldi/matrix/tp-matrix.h
--- a/speechx/speechx/kaldi/util/CMakeLists.txt
+++ b/speechx/speechx/kaldi/util/CMakeLists.txt
--- a/speechx/speechx/kaldi/util/basic-filebuf.h
+++ b/speechx/speechx/kaldi/util/basic-filebuf.h
--- a/speechx/speechx/kaldi/util/common-utils.h
+++ b/speechx/speechx/kaldi/util/common-utils.h
--- a/speechx/speechx/kaldi/util/const-integer-set-inl.h
+++ b/speechx/speechx/kaldi/util/const-integer-set-inl.h
--- a/speechx/speechx/kaldi/util/const-integer-set.h
+++ b/speechx/speechx/kaldi/util/const-integer-set.h
--- a/speechx/speechx/kaldi/util/edit-distance-inl.h
+++ b/speechx/speechx/kaldi/util/edit-distance-inl.h
--- a/speechx/speechx/kaldi/util/edit-distance.h
+++ b/speechx/speechx/kaldi/util/edit-distance.h
--- a/speechx/speechx/kaldi/util/hash-list-inl.h
+++ b/speechx/speechx/kaldi/util/hash-list-inl.h
--- a/speechx/speechx/kaldi/util/hash-list.h
+++ b/speechx/speechx/kaldi/util/hash-list.h
--- a/speechx/speechx/kaldi/util/kaldi-cygwin-io-inl.h
+++ b/speechx/speechx/kaldi/util/kaldi-cygwin-io-inl.h
--- a/speechx/speechx/kaldi/util/kaldi-holder-inl.h
+++ b/speechx/speechx/kaldi/util/kaldi-holder-inl.h
--- a/speechx/speechx/kaldi/util/kaldi-holder.cc
+++ b/speechx/speechx/kaldi/util/kaldi-holder.cc
--- a/speechx/speechx/kaldi/util/kaldi-holder.h
+++ b/speechx/speechx/kaldi/util/kaldi-holder.h
--- a/speechx/speechx/kaldi/util/kaldi-io-inl.h
+++ b/speechx/speechx/kaldi/util/kaldi-io-inl.h
--- a/speechx/speechx/kaldi/util/kaldi-io.cc
+++ b/speechx/speechx/kaldi/util/kaldi-io.cc
--- a/speechx/speechx/kaldi/util/kaldi-io.h
+++ b/speechx/speechx/kaldi/util/kaldi-io.h
--- a/speechx/speechx/kaldi/util/kaldi-pipebuf.h
+++ b/speechx/speechx/kaldi/util/kaldi-pipebuf.h
--- a/speechx/speechx/kaldi/util/kaldi-semaphore.cc
+++ b/speechx/speechx/kaldi/util/kaldi-semaphore.cc
--- a/speechx/speechx/kaldi/util/kaldi-semaphore.h
+++ b/speechx/speechx/kaldi/util/kaldi-semaphore.h
--- a/speechx/speechx/kaldi/util/kaldi-table-inl.h
+++ b/speechx/speechx/kaldi/util/kaldi-table-inl.h
--- a/speechx/speechx/kaldi/util/kaldi-table.cc
+++ b/speechx/speechx/kaldi/util/kaldi-table.cc
--- a/speechx/speechx/kaldi/util/kaldi-table.h
+++ b/speechx/speechx/kaldi/util/kaldi-table.h
--- a/speechx/speechx/kaldi/util/kaldi-thread.cc
+++ b/speechx/speechx/kaldi/util/kaldi-thread.cc
--- a/speechx/speechx/kaldi/util/kaldi-thread.h
+++ b/speechx/speechx/kaldi/util/kaldi-thread.h
--- a/speechx/speechx/kaldi/util/options-itf.h
+++ b/speechx/speechx/kaldi/util/options-itf.h
--- a/speechx/speechx/kaldi/util/parse-options.cc
+++ b/speechx/speechx/kaldi/util/parse-options.cc
--- a/speechx/speechx/kaldi/util/parse-options.h
+++ b/speechx/speechx/kaldi/util/parse-options.h
--- a/speechx/speechx/kaldi/util/simple-io-funcs.cc
+++ b/speechx/speechx/kaldi/util/simple-io-funcs.cc
--- a/speechx/speechx/kaldi/util/simple-io-funcs.h
+++ b/speechx/speechx/kaldi/util/simple-io-funcs.h
--- a/speechx/speechx/kaldi/util/simple-options.cc
+++ b/speechx/speechx/kaldi/util/simple-options.cc
--- a/speechx/speechx/kaldi/util/simple-options.h
+++ b/speechx/speechx/kaldi/util/simple-options.h
--- a/speechx/speechx/kaldi/util/stl-utils.h
+++ b/speechx/speechx/kaldi/util/stl-utils.h
--- a/speechx/speechx/kaldi/util/table-types.h
+++ b/speechx/speechx/kaldi/util/table-types.h
--- a/speechx/speechx/kaldi/util/text-utils.cc
+++ b/speechx/speechx/kaldi/util/text-utils.cc
--- a/speechx/speechx/kaldi/util/text-utils.h
+++ b/speechx/speechx/kaldi/util/text-utils.h
--- a/speechx/speechx/model/CMakeLists.txt
+++ b/speechx/speechx/model/CMakeLists.txt
--- a/speechx/speechx/protocol/CMakeLists.txt
+++ b/speechx/speechx/protocol/CMakeLists.txt
--- a/speechx/speechx/third_party/CMakeLists.txt
+++ b/speechx/speechx/third_party/CMakeLists.txt
--- a/speechx/speechx/third_party/README.md
+++ b/speechx/speechx/third_party/README.md
--- a/speechx/speechx/utils/CMakeLists.txt
+++ b/speechx/speechx/utils/CMakeLists.txt
--- a/tests/benchmark/conformer/README.md
+++ b/tests/benchmark/conformer/README.md
--- a/tests/test_tipc/benchmark_train.sh
+++ b/tests/test_tipc/benchmark_train.sh
--- a/tests/test_tipc/common_func.sh
+++ b/tests/test_tipc/common_func.sh
--- a/tests/test_tipc/configs/conformer/train_benchmark.txt
+++ b/tests/test_tipc/configs/conformer/train_benchmark.txt
--- a/tests/test_tipc/configs/pwgan/train_benchmark.txt
+++ b/tests/test_tipc/configs/pwgan/train_benchmark.txt
--- a/tests/test_tipc/conformer/scripts/aishell_tiny.py
+++ b/tests/test_tipc/conformer/scripts/aishell_tiny.py
--- a/tests/test_tipc/docs/benchmark_train.md
+++ b/tests/test_tipc/docs/benchmark_train.md
--- a/tests/test_tipc/prepare.sh
+++ b/tests/test_tipc/prepare.sh
--- a/tests/test_tipc/test_train_inference_python.sh
+++ b/tests/test_tipc/test_train_inference_python.sh
--- a/tests/unit/cli/test_cli.sh
+++ b/tests/unit/cli/test_cli.sh
--- a/third_party/ctc_decoders/ctc_beam_search_decoder.cpp
+++ b/third_party/ctc_decoders/ctc_beam_search_decoder.cpp
--- a/third_party/ctc_decoders/ctc_beam_search_decoder.h
+++ b/third_party/ctc_decoders/ctc_beam_search_decoder.h
--- a/third_party/ctc_decoders/ctc_greedy_decoder.cpp
+++ b/third_party/ctc_decoders/ctc_greedy_decoder.cpp
--- a/third_party/ctc_decoders/ctc_greedy_decoder.h
+++ b/third_party/ctc_decoders/ctc_greedy_decoder.h
--- a/third_party/ctc_decoders/decoders.i
+++ b/third_party/ctc_decoders/decoders.i
--- a/third_party/ctc_decoders/path_trie.cpp
+++ b/third_party/ctc_decoders/path_trie.cpp
--- a/third_party/ctc_decoders/scorer.cpp
+++ b/third_party/ctc_decoders/scorer.cpp
--- a/third_party/ctc_decoders/scorer.h
+++ b/third_party/ctc_decoders/scorer.h
--- a/third_party/ctc_decoders/setup.py
+++ b/third_party/ctc_decoders/setup.py
--- a/utils/DER.py
+++ b/utils/DER.py
--- a/utils/compute_statistics.py
+++ b/utils/compute_statistics.py
--- a/utils/generate_infer_yaml.py
+++ b/utils/generate_infer_yaml.py
--- a/utils/md-eval.pl
+++ b/utils/md-eval.pl