Merge branch 'PaddlePaddle:develop' into develop

5c4df656 · WilliamZhang06 · GitHub · a2e7ccac · ae521d37 · 5c4df656
143 changed file
--- 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.

--- 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
@@ -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>
+      <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>
@@ -488,7 +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

--- 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
@@ -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>
+      <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>

--- 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
@@ -42,6 +42,7 @@ Contents
   tts/demo
   tts/demo_2
 .. toctree::
   :maxdepth: 1
   :caption: Released Models
@@ -50,12 +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
@@ -32,14 +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)
@@ -52,12 +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)

--- a/docs/source/tts/quick_start_cn.md
+++ b/docs/source/tts/quick_start_cn.md
--- 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/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:
 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).
@@ -17,7 +16,7 @@ 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 durations for aishell3_fastspeech2.
+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
@@ -117,3 +116,25 @@ ref_audio
 ```bash
 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_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
@@ -77,7 +77,7 @@ optimizer:
 ###########################################################
 #                     TRAINING SETTING                    #
 ###########################################################
-max_epoch: 200
+max_epoch: 100
 num_snapshots: 5
 ###########################################################

--- 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=new_tacotron2
+MODEL=tacotron2
 export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
--- 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).

--- 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
@@ -212,6 +212,8 @@ optional arguments:
 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 
 :-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:

--- a/examples/csmsc/tts0/local/synthesize_e2e.sh
+++ b/examples/csmsc/tts0/local/synthesize_e2e.sh
@@ -7,6 +7,7 @@ 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 \
@@ -33,7 +34,7 @@ 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 ${BIN_DIR}/../synthesize_e2e.py \
-        --am=fastspeech2_csmsc \
+        --am=tacotron2_csmsc \
        --am_config=${config_path} \
        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
        --am_stat=dump/train/speech_stats.npy \
@@ -55,7 +56,7 @@ 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 ${BIN_DIR}/../synthesize_e2e.py \
-        --am=fastspeech2_csmsc \
+        --am=tacotron2_csmsc \
        --am_config=${config_path} \
        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
        --am_stat=dump/train/speech_stats.npy \
@@ -76,7 +77,7 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
    FLAGS_allocator_strategy=naive_best_fit \
    FLAGS_fraction_of_gpu_memory_to_use=0.01 \
    python3 ${BIN_DIR}/../synthesize_e2e.py \
-        --am=fastspeech2_csmsc \
+        --am=tacotron2_csmsc \
        --am_config=${config_path} \
        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
        --am_stat=dump/train/speech_stats.npy \
@@ -90,3 +91,24 @@ 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 ${BIN_DIR}/../synthesize_e2e.py \
+        --am=tacotron2_csmsc \
+        --am_config=${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=${train_output_path}/inference
+fi
\ No newline at end of file
--- a/examples/csmsc/tts0/path.sh
+++ b/examples/csmsc/tts0/path.sh
@@ -9,5 +9,5 @@ export PYTHONDONTWRITEBYTECODE=1
 export PYTHONIOENCODING=UTF-8
 export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
-MODEL=new_tacotron2
+MODEL=tacotron2
 export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
--- a/examples/csmsc/tts0/run.sh
+++ b/examples/csmsc/tts0/run.sh
@@ -35,3 +35,8 @@ 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=${train_output_path}/inference
+fi
--- a/examples/csmsc/tts3/README.md
+++ b/examples/csmsc/tts3/README.md
@@ -243,6 +243,8 @@ fastspeech2_nosil_baker_ckpt_0.4
 └── speech_stats.npy        # statistics used to normalize spectrogram when training fastspeech2
 ```
 You can use the following scripts to synthesize for `${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/local/synthesize_e2e.sh
+++ b/examples/csmsc/tts3/local/synthesize_e2e.sh
@@ -102,9 +102,9 @@ if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
        --am_ckpt=${train_output_path}/checkpoints/${ckpt_name} \
        --am_stat=dump/train/speech_stats.npy \
        --voc=wavernn_csmsc \
-        --voc_config=wavernn_test/default.yaml \
+        --voc_config=wavernn_csmsc_ckpt_0.2.0/default.yaml \
-        --voc_ckpt=wavernn_test/snapshot_iter_5000.pdz \
+        --voc_ckpt=wavernn_csmsc_ckpt_0.2.0/snapshot_iter_400000.pdz \
-        --voc_stat=wavernn_test/feats_stats.npy \
+        --voc_stat=wavernn_csmsc_ckpt_0.2.0/feats_stats.npy \
        --lang=zh \
        --text=${BIN_DIR}/../sentences.txt \
        --output_dir=${train_output_path}/test_e2e \

--- a/examples/csmsc/tts3/run.sh
+++ b/examples/csmsc/tts3/run.sh
@@ -36,3 +36,8 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
    CUDA_VISIBLE_DEVICES=${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/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/ljspeech/tts0/README.md
+++ b/examples/ljspeech/tts0/README.md
+# Tacotron2 with LJSpeech-1.1
+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
+### 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.
+## 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 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/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 ${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_ljspeech_ckpt_0.2.0.zip](https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.2.0.zip)
+Model | Step | eval/loss | eval/l1_loss | eval/mse_loss | eval/bce_loss| eval/attn_loss 
+:-------------:| :------------:| :-----: | :-----: | :--------: |:--------:|:---------:
+default| 1(gpu) x 60300|0.554092|0.394260|0.141046|0.018747|3.8e-05|
+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
+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/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=${train_output_path}/inference
\ No newline at end of file
--- a/examples/ljspeech/tts0/path.sh
+++ b/examples/ljspeech/tts0/path.sh
@@ -9,5 +9,5 @@ export PYTHONDONTWRITEBYTECODE=1
 export PYTHONIOENCODING=UTF-8
 export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
-MODEL=new_tacotron2
+MODEL=tacotron2
 export BIN_DIR=${MAIN_ROOT}/paddlespeech/t2s/exps/${MODEL}
--- 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/voxceleb/README.md
+++ b/examples/voxceleb/README.md
+dataset info refer to [VoxCeleb](https://www.robots.ox.ac.uk/~vgg/data/voxceleb/index.html#about)
+sv0 - speaker verfication with softmax backend etc, all python code
+      more info refer to the sv0/readme.txt
+sv1 - dependence on kaldi, speaker verfication with plda/sc backend, 
+      more info refer to the sv1/readme.txt
--- a/examples/voxceleb/sv0/local/make_voxceleb_kaldi_trial.py
+++ b/examples/voxceleb/sv0/local/make_voxceleb_kaldi_trial.py
+#!/usr/bin/python3
+# 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.
+"""
+Make VoxCeleb1 trial of kaldi format
+this script creat the test trial from kaldi trial voxceleb1_test_v2.txt or official trial veri_test2.txt 
+to kaldi trial format
+"""
+import argparse
+import codecs
+import os
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument("--voxceleb_trial",
+                    default="voxceleb1_test_v2",
+                    type=str,
+                    help="VoxCeleb trial file. Default we use the kaldi trial voxceleb1_test_v2.txt")
+parser.add_argument("--trial",
+                    default="data/test/trial",
+                    type=str,
+                    help="Kaldi format trial file")
+args = parser.parse_args()
+def main(voxceleb_trial, trial):
+    """
+        VoxCeleb provide several trial file, which format is different with kaldi format.
+        VoxCeleb format's meaning is as following:
+        --------------------------------
+        target_or_nontarget path1 path2
+        --------------------------------
+        target_or_nontarget is an integer: 1 target                 path1 is equal to path2
+                                           0 nontarget              path1 is unequal to path2    
+        path1: spkr_id/rec_id/name
+        path2: spkr_id/rec_id/name
+        Kaldi format's meaning is as following:
+        ---------------------------------------
+        utt_id1 utt_id2 target_or_nontarget
+        ---------------------------------------
+        utt_id1: utterance identification or speaker identification
+        utt_id2: utterance identification or speaker identification
+        target_or_nontarget is an string: 'target' utt_id1 is equal to  utt_id2
+                                        'nontarget' utt_id2 is unequal to utt_id2
+    """
+    print("Start convert the voxceleb trial to kaldi format")
+    if not os.path.exists(voxceleb_trial):
+        raise RuntimeError("{} does not exist. Pleas input the correct file path".format(voxceleb_trial))
+    trial_dirname = os.path.dirname(trial)
+    if not os.path.exists(trial_dirname):
+        os.mkdir(trial_dirname)
+    with codecs.open(voxceleb_trial, 'r', encoding='utf-8') as f, \
+         codecs.open(trial, 'w', encoding='utf-8') as w:
+         for line in f:
+            target_or_nontarget, path1, path2 = line.strip().split()
+            utt_id1 = "-".join(path1.split("/"))
+            utt_id2 = "-".join(path2.split("/"))
+            target = "nontarget"
+            if int(target_or_nontarget):
+                target = "target"
+            w.write("{} {} {}\n".format(utt_id1, utt_id2, target))
+    print("Convert the voxceleb trial to kaldi format successfully")
+if __name__ == "__main__":
+    main(args.voxceleb_trial, args.trial)
--- a/paddleaudio/features/core.py
+++ b/paddleaudio/features/core.py
@@ -415,11 +415,11 @@ def mfcc(x,
        **kwargs)
    # librosa mfcc:
-    spect = librosa.feature.melspectrogram(x,sr=16000,n_fft=512,
+    spect = librosa.feature.melspectrogram(y=x,sr=16000,n_fft=512,
                                              win_length=512,
                                              hop_length=320,
                                              n_mels=64, fmin=50)
-    b = librosa.feature.mfcc(x,
+    b = librosa.feature.mfcc(y=x,
        sr=16000,
        S=spect,
        n_mfcc=20,

--- a/paddlespeech/cli/asr/infer.py
+++ b/paddlespeech/cli/asr/infer.py
@@ -311,8 +311,10 @@ class ASRExecutor(BaseExecutor):
                    audio = audio[:, 0]
                # pcm16 -> pcm 32
                audio = self._pcm16to32(audio)
-                audio = librosa.resample(audio, audio_sample_rate,
+                audio = librosa.resample(
-                                         self.sample_rate)
+                    audio,
+                    orig_sr=audio_sample_rate,
+                    target_sr=self.sample_rate)
                audio_sample_rate = self.sample_rate
                # pcm32 -> pcm 16
                audio = self._pcm32to16(audio)

--- a/paddlespeech/cli/cls/infer.py
+++ b/paddlespeech/cli/cls/infer.py
@@ -114,8 +114,9 @@ class CLSExecutor(BaseExecutor):
        """
            Download and returns pretrained resources path of current task.
        """
-        assert tag in pretrained_models, 'Can not find pretrained resources of {}.'.format(
+        support_models = list(pretrained_models.keys())
-            tag)
+        assert tag in pretrained_models, 'The model "{}" you want to use has not been supported, please choose other models.\nThe support models includes:\n\t\t{}\n'.format(
+            tag, '\n\t\t'.join(support_models))
        res_path = os.path.join(MODEL_HOME, tag)
        decompressed_path = download_and_decompress(pretrained_models[tag],

--- a/paddlespeech/cli/st/infer.py
+++ b/paddlespeech/cli/st/infer.py
@@ -112,8 +112,9 @@ class STExecutor(BaseExecutor):
        """
            Download and returns pretrained resources path of current task.
        """
-        assert tag in pretrained_models, "Can not find pretrained resources of {}.".format(
+        support_models = list(pretrained_models.keys())
-            tag)
+        assert tag in pretrained_models, 'The model "{}" you want to use has not been supported, please choose other models.\nThe support models includes:\n\t\t{}\n'.format(
+            tag, '\n\t\t'.join(support_models))
        res_path = os.path.join(MODEL_HOME, tag)
        decompressed_path = download_and_decompress(pretrained_models[tag],

--- a/paddlespeech/cli/text/infer.py
+++ b/paddlespeech/cli/text/infer.py
@@ -124,8 +124,9 @@ class TextExecutor(BaseExecutor):
        """
            Download and returns pretrained resources path of current task.
        """
-        assert tag in pretrained_models, 'Can not find pretrained resources of {}.'.format(
+        support_models = list(pretrained_models.keys())
-            tag)
+        assert tag in pretrained_models, 'The model "{}" you want to use has not been supported, please choose other models.\nThe support models includes:\n\t\t{}\n'.format(
+            tag, '\n\t\t'.join(support_models))
        res_path = os.path.join(MODEL_HOME, tag)
        decompressed_path = download_and_decompress(pretrained_models[tag],

--- a/paddlespeech/cli/tts/infer.py
+++ b/paddlespeech/cli/tts/infer.py
@@ -117,6 +117,36 @@ pretrained_models = {
        'speaker_dict':
        'speaker_id_map.txt',
    },
+    # tacotron2
+    "tacotron2_csmsc-zh": {
+        'url':
+        'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_csmsc_ckpt_0.2.0.zip',
+        'md5':
+        '0df4b6f0bcbe0d73c5ed6df8867ab91a',
+        'config':
+        'default.yaml',
+        'ckpt':
+        'snapshot_iter_30600.pdz',
+        'speech_stats':
+        'speech_stats.npy',
+        'phones_dict':
+        'phone_id_map.txt',
+    },
+    "tacotron2_ljspeech-en": {
+        'url':
+        'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/tacotron2/tacotron2_ljspeech_ckpt_0.2.0.zip',
+        'md5':
+        '6a5eddd81ae0e81d16959b97481135f3',
+        'config':
+        'default.yaml',
+        'ckpt':
+        'snapshot_iter_60300.pdz',
+        'speech_stats':
+        'speech_stats.npy',
+        'phones_dict':
+        'phone_id_map.txt',
+    },
    # pwgan
    "pwgan_csmsc-zh": {
        'url':
@@ -205,6 +235,20 @@ pretrained_models = {
        'speech_stats':
        'feats_stats.npy',
    },
+    # wavernn
+    "wavernn_csmsc-zh": {
+        'url':
+        'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/wavernn/wavernn_csmsc_ckpt_0.2.0.zip',
+        'md5':
+        'ee37b752f09bcba8f2af3b777ca38e13',
+        'config':
+        'default.yaml',
+        'ckpt':
+        'snapshot_iter_400000.pdz',
+        'speech_stats':
+        'feats_stats.npy',
+    }
 }
 model_alias = {
@@ -217,6 +261,10 @@ model_alias = {
    "paddlespeech.t2s.models.fastspeech2:FastSpeech2",
    "fastspeech2_inference":
    "paddlespeech.t2s.models.fastspeech2:FastSpeech2Inference",
+    "tacotron2":
+    "paddlespeech.t2s.models.tacotron2:Tacotron2",
+    "tacotron2_inference":
+    "paddlespeech.t2s.models.tacotron2:Tacotron2Inference",
    # voc
    "pwgan":
    "paddlespeech.t2s.models.parallel_wavegan:PWGGenerator",
@@ -234,6 +282,10 @@ model_alias = {
    "paddlespeech.t2s.models.hifigan:HiFiGANGenerator",
    "hifigan_inference":
    "paddlespeech.t2s.models.hifigan:HiFiGANInference",
+    "wavernn":
+    "paddlespeech.t2s.models.wavernn:WaveRNN",
+    "wavernn_inference":
+    "paddlespeech.t2s.models.wavernn:WaveRNNInference",
 }
@@ -253,9 +305,13 @@ class TTSExecutor(BaseExecutor):
            type=str,
            default='fastspeech2_csmsc',
            choices=[
-                'speedyspeech_csmsc', 'fastspeech2_csmsc',
+                'speedyspeech_csmsc',
-                'fastspeech2_ljspeech', 'fastspeech2_aishell3',
+                'fastspeech2_csmsc',
-                'fastspeech2_vctk'
+                'fastspeech2_ljspeech',
+                'fastspeech2_aishell3',
+                'fastspeech2_vctk',
+                'tacotron2_csmsc',
+                'tacotron2_ljspeech',
            ],
            help='Choose acoustic model type of tts task.')
        self.parser.add_argument(
@@ -300,8 +356,14 @@ class TTSExecutor(BaseExecutor):
            type=str,
            default='pwgan_csmsc',
            choices=[
-                'pwgan_csmsc', 'pwgan_ljspeech', 'pwgan_aishell3', 'pwgan_vctk',
+                'pwgan_csmsc',
-                'mb_melgan_csmsc', 'style_melgan_csmsc', 'hifigan_csmsc'
+                'pwgan_ljspeech',
+                'pwgan_aishell3',
+                'pwgan_vctk',
+                'mb_melgan_csmsc',
+                'style_melgan_csmsc',
+                'hifigan_csmsc',
+                'wavernn_csmsc',
            ],
            help='Choose vocoder type of tts task.')
@@ -340,8 +402,9 @@ class TTSExecutor(BaseExecutor):
        """
        Download and returns pretrained resources path of current task.
        """
-        assert tag in pretrained_models, 'Can not find pretrained resources of {}.'.format(
+        support_models = list(pretrained_models.keys())
-            tag)
+        assert tag in pretrained_models, 'The model "{}" you want to use has not been supported, please choose other models.\nThe support models includes:\n\t\t{}\n'.format(
+            tag, '\n\t\t'.join(support_models))
        res_path = os.path.join(MODEL_HOME, tag)
        decompressed_path = download_and_decompress(pretrained_models[tag],
@@ -368,7 +431,7 @@ class TTSExecutor(BaseExecutor):
        """
        Init model and other resources from a specific path.
        """
-        if hasattr(self, 'am') and hasattr(self, 'voc'):
+        if hasattr(self, 'am_inference') and hasattr(self, 'voc_inference'):
            logger.info('Models had been initialized.')
            return
        # am
@@ -488,6 +551,8 @@ class TTSExecutor(BaseExecutor):
                vocab_size=vocab_size,
                tone_size=tone_size,
                **self.am_config["model"])
+        elif am_name == 'tacotron2':
+            am = am_class(idim=vocab_size, odim=odim, **self.am_config["model"])
        am.set_state_dict(paddle.load(self.am_ckpt)["main_params"])
        am.eval()
@@ -505,10 +570,15 @@ class TTSExecutor(BaseExecutor):
        voc_class = dynamic_import(voc_name, model_alias)
        voc_inference_class = dynamic_import(voc_name + '_inference',
                                             model_alias)
+        if voc_name != 'wavernn':
            voc = voc_class(**self.voc_config["generator_params"])
            voc.set_state_dict(paddle.load(self.voc_ckpt)["generator_params"])
            voc.remove_weight_norm()
            voc.eval()
+        else:
+            voc = voc_class(**self.voc_config["model"])
+            voc.set_state_dict(paddle.load(self.voc_ckpt)["main_params"])
+            voc.eval()
        voc_mu, voc_std = np.load(self.voc_stat)
        voc_mu = paddle.to_tensor(voc_mu)
        voc_std = paddle.to_tensor(voc_std)

--- a/paddlespeech/s2t/exps/u2/model.py
+++ b/paddlespeech/s2t/exps/u2/model.py
@@ -175,7 +175,7 @@ class U2Trainer(Trainer):
                        observation['batch_cost'] = observation[
                            'reader_cost'] + observation['step_cost']
                        observation['samples'] = observation['batch_size']
-                        observation['ips,sent./sec'] = observation[
+                        observation['ips,samples/s'] = observation[
                            'batch_size'] / observation['batch_cost']
                        for k, v in observation.items():
                            msg += f" {k.split(',')[0]}: "

--- a/paddlespeech/s2t/io/batchfy.py
+++ b/paddlespeech/s2t/io/batchfy.py
@@ -419,7 +419,7 @@ def make_batchset(
        # sort it by input lengths (long to short)
        sorted_data = sorted(
            d.items(),
-            key=lambda data: int(data[1][batch_sort_key][batch_sort_axis]["shape"][0]),
+            key=lambda data: float(data[1][batch_sort_key][batch_sort_axis]["shape"][0]),
            reverse=not shortest_first, )
        logger.info("# utts: " + str(len(sorted_data)))

--- a/paddlespeech/s2t/io/dataloader.py
+++ b/paddlespeech/s2t/io/dataloader.py
@@ -61,7 +61,7 @@ class BatchDataLoader():
    def __init__(self,
                 json_file: str,
                 train_mode: bool,
-                 sortagrad: bool=False,
+                 sortagrad: int=0,
                 batch_size: int=0,
                 maxlen_in: float=float('inf'),
                 maxlen_out: float=float('inf'),

--- a/paddlespeech/s2t/training/trainer.py
+++ b/paddlespeech/s2t/training/trainer.py
@@ -252,8 +252,7 @@ class Trainer():
        if self.args.benchmark_max_step and self.iteration > self.args.benchmark_max_step:
            logger.info(
                f"Reach benchmark-max-step: {self.args.benchmark_max_step}")
-            sys.exit(
+            sys.exit(0)
-                f"Reach benchmark-max-step: {self.args.benchmark_max_step}")
    def do_train(self):
        """The training process control by epoch."""
@@ -282,7 +281,7 @@ class Trainer():
                        observation['batch_cost'] = observation[
                            'reader_cost'] + observation['step_cost']
                        observation['samples'] = observation['batch_size']
-                        observation['ips[sent./sec]'] = observation[
+                        observation['ips samples/s'] = observation[
                            'batch_size'] / observation['batch_cost']
                        for k, v in observation.items():
                            msg += f" {k}: "

--- a/paddlespeech/s2t/transform/perturb.py
+++ b/paddlespeech/s2t/transform/perturb.py
@@ -90,7 +90,8 @@ class SpeedPerturbation():
        # Note1: resample requires the sampling-rate of input and output,
        #        but actually only the ratio is used.
-        y = librosa.resample(x, ratio, 1, res_type=self.res_type)
+        y = librosa.resample(
+            x, orig_sr=ratio, target_sr=1, res_type=self.res_type)
        if self.keep_length:
            diff = abs(len(x) - len(y))

--- a/paddlespeech/s2t/transform/spectrogram.py
+++ b/paddlespeech/s2t/transform/spectrogram.py
@@ -38,7 +38,7 @@ def stft(x,
    x = np.stack(
        [
            librosa.stft(
-                x[:, ch],
+                y=x[:, ch],
                n_fft=n_fft,
                hop_length=n_shift,
                win_length=win_length,
@@ -67,7 +67,7 @@ def istft(x, n_shift, win_length=None, window="hann", center=True):
    x = np.stack(
        [
            librosa.istft(
-                x[:, ch].T,  # [Time, Freq] -> [Freq, Time]
+                stft_matrix=x[:, ch].T,  # [Time, Freq] -> [Freq, Time]
                hop_length=n_shift,
                win_length=win_length,
                window=window,
@@ -95,7 +95,8 @@ def stft2logmelspectrogram(x_stft,
    # spc: (Time, Channel, Freq) or (Time, Freq)
    spc = np.abs(x_stft)
    # mel_basis: (Mel_freq, Freq)
-    mel_basis = librosa.filters.mel(fs, n_fft, n_mels, fmin, fmax)
+    mel_basis = librosa.filters.mel(
+        sr=fs, n_fft=n_fft, n_mels=n_mels, fmin=fmin, fmax=fmax)
    # lmspc: (Time, Channel, Mel_freq) or (Time, Mel_freq)
    lmspc = np.log10(np.maximum(eps, np.dot(spc, mel_basis.T)))

--- a/paddlespeech/t2s/__init__.py
+++ b/paddlespeech/t2s/__init__.py
@@ -13,7 +13,6 @@
 # limitations under the License.
 import logging
-from . import data
 from . import datasets
 from . import exps
 from . import frontend

--- a/paddlespeech/t2s/audio/audio.py
+++ b/paddlespeech/t2s/audio/audio.py
@@ -53,8 +53,8 @@ class AudioProcessor(object):
    def _create_mel_filter(self):
        mel_filter = librosa.filters.mel(
-            self.sample_rate,
+            sr=self.sample_rate,
-            self.n_fft,
+            n_fft=self.n_fft,
            n_mels=self.n_mels,
            fmin=self.fmin,
            fmax=self.fmax)

--- a/paddlespeech/t2s/data/__init__.py
+++ b/paddlespeech/t2s/data/__init__.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.
-"""t2s's infrastructure for data processing.
-"""
-from .batch import *
-from .dataset import *
--- a/paddlespeech/t2s/datasets/__init__.py
+++ b/paddlespeech/t2s/datasets/__init__.py
@@ -11,5 +11,4 @@
 # 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.
-from .common import *
 from .ljspeech import *
--- a/paddlespeech/t2s/datasets/am_batch_fn.py
+++ b/paddlespeech/t2s/datasets/am_batch_fn.py
@@ -14,7 +14,7 @@
 import numpy as np
 import paddle
-from paddlespeech.t2s.data.batch import batch_sequences
+from paddlespeech.t2s.datasets.batch import batch_sequences
 def tacotron2_single_spk_batch_fn(examples):

--- a/paddlespeech/t2s/data/batch.py
+++ b/paddlespeech/t2s/data/batch.py
--- a/paddlespeech/t2s/datasets/common.py
+++ b/paddlespeech/t2s/datasets/common.py
-# Copyright (c) 2020 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.
-from pathlib import Path
-from typing import List
-import librosa
-import numpy as np
-from paddle.io import Dataset
-__all__ = ["AudioSegmentDataset", "AudioDataset", "AudioFolderDataset"]
-class AudioSegmentDataset(Dataset):
-    """A simple dataset adaptor for audio files to train vocoders.
-    Read -> trim silence -> normalize -> extract a segment
-    """
-    def __init__(self,
-                 file_paths: List[Path],
-                 sample_rate: int,
-                 length: int,
-                 top_db: float):
-        self.file_paths = file_paths
-        self.sr = sample_rate
-        self.top_db = top_db
-        self.length = length  # samples in the clip
-    def __getitem__(self, i):
-        fpath = self.file_paths[i]
-        y, sr = librosa.load(fpath, self.sr)
-        y, _ = librosa.effects.trim(y, top_db=self.top_db)
-        y = librosa.util.normalize(y)
-        y = y.astype(np.float32)
-        # pad or trim
-        if y.size <= self.length:
-            y = np.pad(y, [0, self.length - len(y)], mode='constant')
-        else:
-            start = np.random.randint(0, 1 + len(y) - self.length)
-            y = y[start:start + self.length]
-        return y
-    def __len__(self):
-        return len(self.file_paths)
-class AudioDataset(Dataset):
-    """A simple dataset adaptor for the audio files.
-    Read -> trim silence -> normalize
-    """
-    def __init__(self,
-                 file_paths: List[Path],
-                 sample_rate: int,
-                 top_db: float=60):
-        self.file_paths = file_paths
-        self.sr = sample_rate
-        self.top_db = top_db
-    def __getitem__(self, i):
-        fpath = self.file_paths[i]
-        y, sr = librosa.load(fpath, self.sr)
-        y, _ = librosa.effects.trim(y, top_db=self.top_db)
-        y = librosa.util.normalize(y)
-        y = y.astype(np.float32)
-        return y
-    def __len__(self):
-        return len(self.file_paths)
-class AudioFolderDataset(AudioDataset):
-    def __init__(
-            self,
-            root,
-            sample_rate,
-            top_db=60,
-            extension=".wav", ):
-        root = Path(root).expanduser()
-        file_paths = sorted(list(root.rglob("*{}".format(extension))))
-        super().__init__(file_paths, sample_rate, top_db)
--- a/paddlespeech/t2s/datasets/data_table.py
+++ b/paddlespeech/t2s/datasets/data_table.py
@@ -22,25 +22,16 @@ from paddle.io import Dataset
 class DataTable(Dataset):
    """Dataset to load and convert data for general purpose.
+    Args:
-    Parameters
+        data (List[Dict[str, Any]]): Metadata, a list of meta datum, each of which is composed of  several fields
-    ----------
+        fields (List[str], optional): Fields to use, if not specified, all the fields in the data are used, by default None
-    data : List[Dict[str, Any]]
+        converters (Dict[str, Callable], optional): Converters used to process each field, by default None
-        Metadata, a list of meta datum, each of which is composed of 
+        use_cache (bool, optional): Whether to use cache, by default False
-        several fields
-    fields : List[str], optional
+    Raises:
-        Fields to use, if not specified, all the fields in the data are 
+        ValueError:
-        used, by default None
-    converters : Dict[str, Callable], optional
-        Converters used to process each field, by default None
-    use_cache : bool, optional
-        Whether to use cache, by default False
-    Raises
-    ------
-    ValueError
            If there is some field that does not exist in data. 
-    ValueError
+        ValueError:
            If there is some field in converters that does not exist in fields.
    """
@@ -95,15 +86,11 @@ class DataTable(Dataset):
        """Convert a meta datum to an example by applying the corresponding 
        converters to each fields requested.
-        Parameters
+        Args:
-        ----------
+            meta_datum (Dict[str, Any]): Meta datum
-        meta_datum : Dict[str, Any]
-            Meta datum
-        Returns
+        Returns:
-        -------
+            Dict[str, Any]: Converted example
-        Dict[str, Any]
-            Converted example
        """
        example = {}
        for field in self.fields:
@@ -118,16 +105,11 @@ class DataTable(Dataset):
    def __getitem__(self, idx: int) -> Dict[str, Any]:
        """Get an example given an index.
+        Args:
+            idx (int): Index of the example to get
-        Parameters
+        Returns:
-        ----------
+            Dict[str, Any]: A converted example
-        idx : int
-            Index of the example to get
-        Returns
-        -------
-        Dict[str, Any]
-            A converted example
        """
        if self.use_cache and self.caches[idx] is not None:
            return self.caches[idx]

--- 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
@@ -18,14 +18,10 @@ import re
 def get_phn_dur(file_name):
    '''
    read MFA duration.txt
-    Parameters
+    Args:
-    ----------
+        file_name (str or Path): path of gen_duration_from_textgrid.py's result
-    file_name : str or Path
+    Returns: 
-        path of gen_duration_from_textgrid.py's result
+        Dict: sentence: {'utt': ([char], [int])}
-    Returns
-    ----------
-    Dict
-        sentence: {'utt': ([char], [int])}
    '''
    f = open(file_name, 'r')
    sentence = {}
@@ -48,10 +44,8 @@ def get_phn_dur(file_name):
 def merge_silence(sentence):
    '''
    merge silences
-    Parameters
+    Args:
-    ----------
+        sentence (Dict): sentence: {'utt': (([char], [int]), str)}
-    sentence : Dict
-        sentence: {'utt': (([char], [int]), str)}
    '''
    for utt in sentence:
        cur_phn, cur_dur, speaker = sentence[utt]
@@ -81,12 +75,9 @@ def merge_silence(sentence):
 def get_input_token(sentence, output_path, dataset="baker"):
    '''
    get phone set from training data and save it
-    Parameters
+    Args:
-    ----------
+        sentence (Dict): sentence: {'utt': ([char], [int])}
-    sentence : Dict
+        output_path (str or path):path to save phone_id_map
-        sentence: {'utt': ([char], [int])}
-    output_path : str or path
-        path to save phone_id_map
    '''
    phn_token = set()
    for utt in sentence:
@@ -112,14 +103,10 @@ def get_phones_tones(sentence,
                     dataset="baker"):
    '''
    get phone set and tone set from training data and save it
-    Parameters
+    Args:
-    ----------
+        sentence (Dict): sentence: {'utt': ([char], [int])}
-    sentence : Dict
+        phones_output_path (str or path): path to save phone_id_map
-        sentence: {'utt': ([char], [int])}
+        tones_output_path (str or path): path to save tone_id_map
-    phones_output_path : str or path
-        path to save phone_id_map
-    tones_output_path : str or path
-        path to save tone_id_map
    '''
    phn_token = set()
    tone_token = set()
@@ -162,14 +149,10 @@ def get_spk_id_map(speaker_set, output_path):
 def compare_duration_and_mel_length(sentences, utt, mel):
    '''
    check duration error, correct sentences[utt] if possible, else pop sentences[utt]
-    Parameters
+    Args:
-    ----------
+        sentences (Dict): sentences[utt] = [phones_list ,durations_list]
-    sentences : Dict
+        utt (str): utt_id
-        sentences[utt] = [phones_list ,durations_list]
+        mel (np.ndarry): features (num_frames, n_mels)
-    utt : str
-        utt_id
-    mel : np.ndarry
-        features (num_frames, n_mels)
    '''
    if utt in sentences:

--- a/paddlespeech/t2s/datasets/vocoder_batch_fn.py
+++ b/paddlespeech/t2s/datasets/vocoder_batch_fn.py
@@ -29,15 +29,11 @@ class Clip(object):
            hop_size=256,
            aux_context_window=0, ):
        """Initialize customized collater for DataLoader.
+        Args:
-        Parameters
+            batch_max_steps (int): The maximum length of input signal in batch.
-        ----------
+            hop_size (int): Hop size of auxiliary features.
-        batch_max_steps : int
+            aux_context_window (int): Context window size for auxiliary feature conv.
-            The maximum length of input signal in batch.
-        hop_size : int
-            Hop size of auxiliary features.
-        aux_context_window : int
-            Context window size for auxiliary feature conv.
        """
        if batch_max_steps % hop_size != 0:
@@ -56,17 +52,14 @@ class Clip(object):
    def __call__(self, batch):
        """Convert into batch tensors.
-        Parameters
+        Args:
-        ----------
+            batch (list): list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
-        batch : list
-            list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
-        Returns
+        Returns: 
-        ----------
+            Tensor:
-        Tensor
                Auxiliary feature batch (B, C, T'), where
                T = (T' - 2 * aux_context_window) * hop_size.
-        Tensor
+            Tensor:
                Target signal batch (B, 1, T).
        """
@@ -104,8 +97,7 @@ class Clip(object):
    def _adjust_length(self, x, c):
        """Adjust the audio and feature lengths.
-        Note
+        Note:
-        -------
            Basically we assume that the length of x and c are adjusted
            through preprocessing stage, but if we use other library processed
            features, this process will be needed.
@@ -162,22 +154,14 @@ class WaveRNNClip(Clip):
        # voc_pad = 2  this will pad the input so that the resnet can 'see' wider than input length
        # max_offsets = n_frames - 2 - (mel_win + 2 * hp.voc_pad) = n_frames - 15
        """Convert into batch tensors.
+        Args:
-        Parameters
+            batch (list): list of tuple of the pair of audio and features. Audio shape (T, ), features shape(T', C).
-        ----------
-        batch : list
+        Returns:
-            list of tuple of the pair of audio and features. 
+            Tensor: Input signal batch (B, 1, T).
-            Audio shape (T, ), features shape(T', C).
+            Tensor: Target signal batch (B, 1, T).
+            Tensor: Auxiliary feature batch (B, C, T'), 
-        Returns
+                where T = (T' - 2 * aux_context_window) * hop_size.
-        ----------
-        Tensor
-            Input signal batch (B, 1, T).
-        Tensor
-            Target signal batch (B, 1, T).
-        Tensor
-            Auxiliary feature batch (B, C, T'), where
-            T = (T' - 2 * aux_context_window) * hop_size.
        """
        # check length

--- a/paddlespeech/t2s/exps/fastspeech2/preprocess.py
+++ b/paddlespeech/t2s/exps/fastspeech2/preprocess.py
@@ -27,9 +27,9 @@ import tqdm
 import yaml
 from yacs.config import CfgNode
-from paddlespeech.t2s.data.get_feats import Energy
+from paddlespeech.t2s.datasets.get_feats import Energy
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
-from paddlespeech.t2s.data.get_feats import Pitch
+from paddlespeech.t2s.datasets.get_feats import Pitch
 from paddlespeech.t2s.datasets.preprocess_utils import compare_duration_and_mel_length
 from paddlespeech.t2s.datasets.preprocess_utils import get_input_token
 from paddlespeech.t2s.datasets.preprocess_utils import get_phn_dur

--- a/paddlespeech/t2s/exps/fastspeech2/train.py
+++ b/paddlespeech/t2s/exps/fastspeech2/train.py
@@ -162,7 +162,6 @@ def train_sp(args, config):
        trainer.extend(VisualDL(output_dir), trigger=(1, "iteration"))
    trainer.extend(
        Snapshot(max_size=config.num_snapshots), trigger=(1, 'epoch'))
-    # print(trainer.extensions)
    trainer.run()

--- 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
@@ -23,7 +23,7 @@ import soundfile as sf
 import yaml
 from yacs.config import CfgNode
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
 from paddlespeech.t2s.models.parallel_wavegan import PWGGenerator
 from paddlespeech.t2s.models.parallel_wavegan import PWGInference
 from paddlespeech.t2s.modules.normalizer import ZScore

--- a/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/parallelwave_gan/train.py
@@ -198,7 +198,6 @@ def train_sp(args, config):
        Snapshot(max_size=config.num_snapshots),
        trigger=(config.save_interval_steps, 'iteration'))
-    # print(trainer.extensions.keys())
    print("Trainer Done!")
    trainer.run()

--- a/paddlespeech/t2s/exps/gan_vocoder/preprocess.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/preprocess.py
@@ -27,7 +27,7 @@ import tqdm
 import yaml
 from yacs.config import CfgNode
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
 from paddlespeech.t2s.datasets.preprocess_utils import get_phn_dur
 from paddlespeech.t2s.datasets.preprocess_utils import merge_silence
 from paddlespeech.t2s.utils import str2bool

--- a/paddlespeech/t2s/exps/gan_vocoder/style_melgan/train.py
+++ b/paddlespeech/t2s/exps/gan_vocoder/style_melgan/train.py
--- a/paddlespeech/t2s/exps/speedyspeech/preprocess.py
+++ b/paddlespeech/t2s/exps/speedyspeech/preprocess.py
@@ -27,7 +27,7 @@ import tqdm
 import yaml
 from yacs.config import CfgNode
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
 from paddlespeech.t2s.datasets.preprocess_utils import compare_duration_and_mel_length
 from paddlespeech.t2s.datasets.preprocess_utils import get_phn_dur
 from paddlespeech.t2s.datasets.preprocess_utils import get_phones_tones

--- 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
@@ -38,9 +38,9 @@ model_alias = {
    "fastspeech2_inference":
    "paddlespeech.t2s.models.fastspeech2:FastSpeech2Inference",
    "tacotron2":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2",
    "tacotron2_inference":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2Inference",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2Inference",
    # voc
    "pwgan":
    "paddlespeech.t2s.models.parallel_wavegan:PWGGenerator",

--- a/paddlespeech/t2s/exps/synthesize_e2e.py
+++ b/paddlespeech/t2s/exps/synthesize_e2e.py
@@ -39,9 +39,9 @@ model_alias = {
    "fastspeech2_inference":
    "paddlespeech.t2s.models.fastspeech2:FastSpeech2Inference",
    "tacotron2":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2",
    "tacotron2_inference":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2Inference",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2Inference",
    # voc
    "pwgan":
    "paddlespeech.t2s.models.parallel_wavegan:PWGGenerator",
@@ -229,6 +229,11 @@ def evaluate(args):
    output_dir = Path(args.output_dir)
    output_dir.mkdir(parents=True, exist_ok=True)
    merge_sentences = False
+    # Avoid not stopping at the end of a sub sentence when tacotron2_ljspeech dygraph to static graph
+    # but still not stopping in the end (NOTE by yuantian01 Feb 9 2022)
+    if am_name == 'tacotron2':
+        merge_sentences = True
    for utt_id, sentence in sentences:
        get_tone_ids = False
        if am_name == 'speedyspeech':

--- a/paddlespeech/t2s/exps/new_tacotron2/__init__.py
+++ b/paddlespeech/t2s/exps/new_tacotron2/__init__.py
--- a/paddlespeech/t2s/exps/new_tacotron2/normalize.py
+++ b/paddlespeech/t2s/exps/new_tacotron2/normalize.py
--- a/paddlespeech/t2s/exps/new_tacotron2/preprocess.py
+++ b/paddlespeech/t2s/exps/new_tacotron2/preprocess.py
@@ -27,7 +27,7 @@ import tqdm
 import yaml
 from yacs.config import CfgNode
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
 from paddlespeech.t2s.datasets.preprocess_utils import compare_duration_and_mel_length
 from paddlespeech.t2s.datasets.preprocess_utils import get_input_token
 from paddlespeech.t2s.datasets.preprocess_utils import get_phn_dur

--- a/paddlespeech/t2s/exps/new_tacotron2/train.py
+++ b/paddlespeech/t2s/exps/new_tacotron2/train.py
@@ -30,9 +30,9 @@ from yacs.config import CfgNode
 from paddlespeech.t2s.datasets.am_batch_fn import tacotron2_multi_spk_batch_fn
 from paddlespeech.t2s.datasets.am_batch_fn import tacotron2_single_spk_batch_fn
 from paddlespeech.t2s.datasets.data_table import DataTable
-from paddlespeech.t2s.models.new_tacotron2 import Tacotron2
+from paddlespeech.t2s.models.tacotron2 import Tacotron2
-from paddlespeech.t2s.models.new_tacotron2 import Tacotron2Evaluator
+from paddlespeech.t2s.models.tacotron2 import Tacotron2Evaluator
-from paddlespeech.t2s.models.new_tacotron2 import Tacotron2Updater
+from paddlespeech.t2s.models.tacotron2 import Tacotron2Updater
 from paddlespeech.t2s.training.extensions.snapshot import Snapshot
 from paddlespeech.t2s.training.extensions.visualizer import VisualDL
 from paddlespeech.t2s.training.optimizer import build_optimizers
@@ -157,7 +157,6 @@ def train_sp(args, config):
        trainer.extend(VisualDL(output_dir), trigger=(1, "iteration"))
    trainer.extend(
        Snapshot(max_size=config.num_snapshots), trigger=(1, 'epoch'))
-    # print(trainer.extensions)
    trainer.run()

--- a/paddlespeech/t2s/exps/transformer_tts/preprocess.py
+++ b/paddlespeech/t2s/exps/transformer_tts/preprocess.py
@@ -26,20 +26,17 @@ import tqdm
 import yaml
 from yacs.config import CfgNode as Configuration
-from paddlespeech.t2s.data.get_feats import LogMelFBank
+from paddlespeech.t2s.datasets.get_feats import LogMelFBank
 from paddlespeech.t2s.frontend import English
 def get_lj_sentences(file_name, frontend):
-    '''
+    '''read MFA duration.txt
-    read MFA duration.txt
-    Parameters
+    Args:
-    ----------
+        file_name (str or Path)
-    file_name : str or Path
+    Returns:
-    Returns
+        Dict: sentence: {'utt': ([char], [int])}
-    ----------
-    Dict
-        sentence: {'utt': ([char], [int])}
    '''
    f = open(file_name, 'r')
    sentence = {}
@@ -59,14 +56,11 @@ def get_lj_sentences(file_name, frontend):
 def get_input_token(sentence, output_path):
-    '''
+    '''get phone set from training data and save it
-    get phone set from training data and save it
-    Parameters
+    Args:
-    ----------
+        sentence (Dict): sentence: {'utt': ([char], str)}
-    sentence : Dict
+        output_path (str or path): path to save phone_id_map
-        sentence: {'utt': ([char], str)}
-    output_path : str or path
-        path to save phone_id_map
    '''
    phn_token = set()
    for utt in sentence:

--- a/paddlespeech/t2s/exps/transformer_tts/train.py
+++ b/paddlespeech/t2s/exps/transformer_tts/train.py
@@ -150,7 +150,6 @@ def train_sp(args, config):
        trainer.extend(VisualDL(output_dir), trigger=(1, "iteration"))
    trainer.extend(
        Snapshot(max_size=config.num_snapshots), trigger=(1, 'epoch'))
-    # print(trainer.extensions)
    trainer.run()

--- a/paddlespeech/t2s/exps/voice_cloning.py
+++ b/paddlespeech/t2s/exps/voice_cloning.py
@@ -34,9 +34,9 @@ model_alias = {
    "fastspeech2_inference":
    "paddlespeech.t2s.models.fastspeech2:FastSpeech2Inference",
    "tacotron2":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2",
    "tacotron2_inference":
-    "paddlespeech.t2s.models.new_tacotron2:Tacotron2Inference",
+    "paddlespeech.t2s.models.tacotron2:Tacotron2Inference",
    # voc
    "pwgan":
    "paddlespeech.t2s.models.parallel_wavegan:PWGGenerator",

--- a/paddlespeech/t2s/exps/waveflow/ljspeech.py
+++ b/paddlespeech/t2s/exps/waveflow/ljspeech.py
@@ -17,8 +17,8 @@ import numpy as np
 import pandas
 from paddle.io import Dataset
-from paddlespeech.t2s.data.batch import batch_spec
+from paddlespeech.t2s.datasets.batch import batch_spec
-from paddlespeech.t2s.data.batch import batch_wav
+from paddlespeech.t2s.datasets.batch import batch_wav
 class LJSpeech(Dataset):

--- a/paddlespeech/t2s/exps/waveflow/train.py
+++ b/paddlespeech/t2s/exps/waveflow/train.py
@@ -19,7 +19,7 @@ from paddle import distributed as dist
 from paddle.io import DataLoader
 from paddle.io import DistributedBatchSampler
-from paddlespeech.t2s.data import dataset
+from paddlespeech.t2s.datasets import dataset
 from paddlespeech.t2s.exps.waveflow.config import get_cfg_defaults
 from paddlespeech.t2s.exps.waveflow.ljspeech import LJSpeech
 from paddlespeech.t2s.exps.waveflow.ljspeech import LJSpeechClipCollector

--- a/paddlespeech/t2s/exps/wavernn/synthesize.py
+++ b/paddlespeech/t2s/exps/wavernn/synthesize.py
@@ -31,7 +31,7 @@ from paddlespeech.t2s.models.wavernn import WaveRNN
 def main():
    parser = argparse.ArgumentParser(description="Synthesize with WaveRNN.")
-    parser.add_argument("--config", type=str, help="GANVocoder config file.")
+    parser.add_argument("--config", type=str, help="Vocoder config file.")
    parser.add_argument("--checkpoint", type=str, help="snapshot to load.")
    parser.add_argument("--test-metadata", type=str, help="dev data.")
    parser.add_argument("--output-dir", type=str, help="output dir.")

--- a/paddlespeech/t2s/exps/wavernn/train.py
+++ b/paddlespeech/t2s/exps/wavernn/train.py
@@ -179,7 +179,7 @@ def train_sp(args, config):
 def main():
    # parse args and config and redirect to train_sp
-    parser = argparse.ArgumentParser(description="Train a HiFiGAN model.")
+    parser = argparse.ArgumentParser(description="Train a WaveRNN model.")
    parser.add_argument(
        "--config", type=str, help="config file to overwrite default config.")
    parser.add_argument("--train-metadata", type=str, help="training data.")

--- a/paddlespeech/t2s/frontend/arpabet.py
+++ b/paddlespeech/t2s/frontend/arpabet.py
@@ -133,16 +133,11 @@ class ARPABET(Phonetics):
    def phoneticize(self, sentence, add_start_end=False):
        """ Normalize the input text sequence and convert it into pronunciation sequence.
+        Args:
+            sentence (str): The input text sequence.
-        Parameters
+        Returns:
-        -----------
+            List[str]: The list of pronunciation sequence.
-        sentence: str
-            The input text sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        phonemes = [
            self._remove_vowels(item) for item in self.backend(sentence)
@@ -157,15 +152,11 @@ class ARPABET(Phonetics):
    def numericalize(self, phonemes):
        """ Convert pronunciation sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            phonemes (List[str]): The list of pronunciation sequence.
-        phonemes: List[str]
-            The list of pronunciation sequence.
-        Returns
+        Returns:
-        ----------
+            List[int]: The list of pronunciation id sequence.
-        List[int]
-            The list of pronunciation id sequence.
        """
        ids = [self.vocab.lookup(item) for item in phonemes]
        return ids
@@ -173,14 +164,11 @@ class ARPABET(Phonetics):
    def reverse(self, ids):
        """ Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
-        Parameters
+        Args:
-        -----------
+            ids( List[int]): The list of pronunciation id sequence.
-        ids: List[int]
-            The list of pronunciation id sequence.
-        Returns
+        Returns: 
-        ----------
+            List[str]: 
-        List[str]
                The list of pronunciation sequence.
        """
        return [self.vocab.reverse(i) for i in ids]
@@ -188,15 +176,11 @@ class ARPABET(Phonetics):
    def __call__(self, sentence, add_start_end=False):
        """ Convert the input text sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
-            The input text sequence.
-        Returns
+        Returns:
-        ----------
+            List[str]: The list of pronunciation id sequence.
-        List[str]
-            The list of pronunciation id sequence.
        """
        return self.numericalize(
            self.phoneticize(sentence, add_start_end=add_start_end))
@@ -229,15 +213,11 @@ class ARPABETWithStress(Phonetics):
    def phoneticize(self, sentence, add_start_end=False):
        """ Normalize the input text sequence and convert it into pronunciation sequence.
-        Parameters
+        Args: 
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
-            The input text sequence.
-        Returns
+        Returns: 
-        ----------
+            List[str]: The list of pronunciation sequence.
-        List[str]
-            The list of pronunciation sequence.
        """
        phonemes = self.backend(sentence)
        if add_start_end:
@@ -250,46 +230,32 @@ class ARPABETWithStress(Phonetics):
    def numericalize(self, phonemes):
        """ Convert pronunciation sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            phonemes (List[str]): The list of pronunciation sequence.
-        phonemes: List[str]
-            The list of pronunciation sequence.
-        Returns
+        Returns:
-        ----------
+            List[int]: The list of pronunciation id sequence.
-        List[int]
-            The list of pronunciation id sequence.
        """
        ids = [self.vocab.lookup(item) for item in phonemes]
        return ids
    def reverse(self, ids):
        """ Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
+        Args:
+            ids (List[int]): The list of pronunciation id sequence.
-        Parameters
+        Returns: 
-        -----------
+            List[str]: The list of pronunciation sequence.
-        ids: List[int]
-            The list of pronunciation id sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        return [self.vocab.reverse(i) for i in ids]
    def __call__(self, sentence, add_start_end=False):
        """ Convert the input text sequence into pronunciation id sequence.
+        Args:
+            sentence (str): The input text sequence.
-        Parameters
+        Returns: 
-        -----------
+            List[str]: The list of pronunciation id sequence.
-        sentence: str
-            The input text sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation id sequence.
        """
        return self.numericalize(
            self.phoneticize(sentence, add_start_end=add_start_end))

--- a/paddlespeech/t2s/frontend/phonectic.py
+++ b/paddlespeech/t2s/frontend/phonectic.py
@@ -65,14 +65,10 @@ class English(Phonetics):
    def phoneticize(self, sentence):
        """ Normalize the input text sequence and convert it into pronunciation sequence.
-        Parameters
+        Args:
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
+        Returns: 
-            The input text sequence.
+            List[str]: The list of pronunciation sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        start = self.vocab.start_symbol
        end = self.vocab.end_symbol
@@ -83,11 +79,6 @@ class English(Phonetics):
        return phonemes
    def _p2id(self, phonemes: List[str]) -> np.array:
-        # replace unk phone with sp
-        phonemes = [
-            phn if (phn in self.vocab_phones and phn not in self.punc) else "sp"
-            for phn in phonemes
-        ]
        phone_ids = [self.vocab_phones[item] for item in phonemes]
        return np.array(phone_ids, np.int64)
@@ -102,6 +93,12 @@ class English(Phonetics):
            # remove start_symbol and end_symbol
            phones = phones[1:-1]
            phones = [phn for phn in phones if not phn.isspace()]
+            # replace unk phone with sp
+            phones = [
+                phn
+                if (phn in self.vocab_phones and phn not in self.punc) else "sp"
+                for phn in phones
+            ]
            phones_list.append(phones)
        if merge_sentences:
@@ -122,14 +119,10 @@ class English(Phonetics):
    def numericalize(self, phonemes):
        """ Convert pronunciation sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            phonemes (List[str]): The list of pronunciation sequence.
-        phonemes: List[str]
+        Returns: 
-            The list of pronunciation sequence.
+            List[int]: The list of pronunciation id sequence.
-        Returns
-        ----------
-        List[int]
-            The list of pronunciation id sequence.
        """
        ids = [
            self.vocab.lookup(item) for item in phonemes
@@ -139,27 +132,19 @@ class English(Phonetics):
    def reverse(self, ids):
        """ Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
-        Parameters
+        Args:
-        -----------
+            ids (List[int]): The list of pronunciation id sequence.
-        ids: List[int]
+        Returns: 
-            The list of pronunciation id sequence.
+            List[str]: The list of pronunciation sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        return [self.vocab.reverse(i) for i in ids]
    def __call__(self, sentence):
        """ Convert the input text sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            sentence(str): The input text sequence.
-        sentence: str
+        Returns: 
-            The input text sequence.
+            List[str]: The list of pronunciation id sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation id sequence.
        """
        return self.numericalize(self.phoneticize(sentence))
@@ -182,27 +167,20 @@ class EnglishCharacter(Phonetics):
    def phoneticize(self, sentence):
        """ Normalize the input text sequence.
-        Parameters
+        Args:
-        -----------
+            sentence(str): The input text sequence.
-        sentence: str
+        Returns:
-            The input text sequence.
+            str: A text sequence after normalize.
-        Returns
-        ----------
-        str
-            A text sequence after normalize.
        """
        words = normalize(sentence)
        return words
    def numericalize(self, sentence):
        """ Convert a text sequence into ids.
-        Parameters
+        Args:
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
+        Returns:
-            The input text sequence.
+            List[int]:
-        Returns
-        ----------
-        List[int]
                List of a character id sequence.
        """
        ids = [
@@ -213,27 +191,19 @@ class EnglishCharacter(Phonetics):
    def reverse(self, ids):
        """ Convert a character id sequence into text.
-        Parameters
+        Args:
-        -----------
+            ids (List[int]): List of a character id sequence.
-        ids: List[int]
+        Returns:
-            List of a character id sequence.
+            str: The input text sequence.
-        Returns
-        ----------
-        str
-            The input text sequence.
        """
        return [self.vocab.reverse(i) for i in ids]
    def __call__(self, sentence):
        """ Normalize the input text sequence and convert it into character id sequence.
-        Parameters
+        Args:
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
+        Returns: 
-            The input text sequence.
+            List[int]: List of a character id sequence.
-        Returns
-        ----------
-        List[int]
-            List of a character id sequence.
        """
        return self.numericalize(self.phoneticize(sentence))
@@ -263,14 +233,10 @@ class Chinese(Phonetics):
    def phoneticize(self, sentence):
        """ Normalize the input text sequence and convert it into pronunciation sequence.
-        Parameters
+        Args:
-        -----------
+            sentence(str): The input text sequence.
-        sentence: str
+        Returns: 
-            The input text sequence.
+            List[str]: The list of pronunciation sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        # simplified = self.opencc_backend.convert(sentence)
        simplified = sentence
@@ -295,28 +261,20 @@ class Chinese(Phonetics):
    def numericalize(self, phonemes):
        """ Convert pronunciation sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            phonemes(List[str]): The list of pronunciation sequence.
-        phonemes: List[str]
+        Returns:
-            The list of pronunciation sequence.
+                List[int]: The list of pronunciation id sequence.
-        Returns
-        ----------
-        List[int]
-            The list of pronunciation id sequence.
        """
        ids = [self.vocab.lookup(item) for item in phonemes]
        return ids
    def __call__(self, sentence):
        """ Convert the input text sequence into pronunciation id sequence.
-        Parameters
+        Args:
-        -----------
+            sentence (str): The input text sequence.
-        sentence: str
+        Returns:
-            The input text sequence.
+            List[str]: The list of pronunciation id sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation id sequence.
        """
        return self.numericalize(self.phoneticize(sentence))
@@ -328,13 +286,9 @@ class Chinese(Phonetics):
    def reverse(self, ids):
        """ Reverse the list of pronunciation id sequence to a list of pronunciation sequence.
-        Parameters
+        Args:
-        -----------
+        ids (List[int]): The list of pronunciation id sequence.
-        ids: List[int]
+        Returns: 
-            The list of pronunciation id sequence.
+            List[str]: The list of pronunciation sequence.
-        Returns
-        ----------
-        List[str]
-            The list of pronunciation sequence.
        """
        return [self.vocab.reverse(i) for i in ids]
--- a/paddlespeech/t2s/frontend/vocab.py
+++ b/paddlespeech/t2s/frontend/vocab.py
@@ -20,22 +20,12 @@ __all__ = ["Vocab"]
 class Vocab(object):
    """  Vocabulary.
-    Parameters
+    Args:
-    -----------
+        symbols (Iterable[str]): Common symbols.
-    symbols: Iterable[str]
+        padding_symbol (str, optional): Symbol for pad. Defaults to "<pad>".
-        Common symbols.
+        unk_symbol (str, optional): Symbol for unknow. Defaults to "<unk>"
+        start_symbol (str, optional): Symbol for start. Defaults to "<s>"
-    padding_symbol: str, optional
+        end_symbol (str, optional): Symbol for end. Defaults to "</s>"
-        Symbol for pad. Defaults to "<pad>".
-    unk_symbol: str, optional
-        Symbol for unknow. Defaults to "<unk>"
-    start_symbol: str, optional
-        Symbol for start. Defaults to "<s>"
-    end_symbol: str, optional
-        Symbol for end. Defaults to "</s>"
    """
    def __init__(self,

--- a/paddlespeech/t2s/frontend/zh_normalization/chronology.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/chronology.py
@@ -44,11 +44,9 @@ RE_TIME_RANGE = re.compile(r'([0-1]?[0-9]|2[0-3])'
 def replace_time(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
@@ -87,11 +85,9 @@ RE_DATE = re.compile(r'(\d{4}|\d{2})年'
 def replace_date(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    year = match.group(1)
@@ -114,11 +110,9 @@ RE_DATE2 = re.compile(
 def replace_date2(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    year = match.group(1)

--- a/paddlespeech/t2s/frontend/zh_normalization/num.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/num.py
@@ -36,11 +36,9 @@ RE_FRAC = re.compile(r'(-?)(\d+)/(\d+)')
 def replace_frac(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    sign = match.group(1)
@@ -59,11 +57,9 @@ RE_PERCENTAGE = re.compile(r'(-?)(\d+(\.\d+)?)%')
 def replace_percentage(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    sign = match.group(1)
@@ -81,11 +77,9 @@ RE_INTEGER = re.compile(r'(-)' r'(\d+)')
 def replace_negative_num(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    sign = match.group(1)
@@ -103,11 +97,9 @@ RE_DEFAULT_NUM = re.compile(r'\d{3}\d*')
 def replace_default_num(match):
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    number = match.group(0)
@@ -124,11 +116,9 @@ RE_NUMBER = re.compile(r'(-?)((\d+)(\.\d+)?)' r'|(\.(\d+))')
 def replace_positive_quantifier(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    number = match.group(1)
@@ -142,11 +132,9 @@ def replace_positive_quantifier(match) -> str:
 def replace_number(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    sign = match.group(1)
@@ -169,11 +157,9 @@ RE_RANGE = re.compile(
 def replace_range(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    first, second = match.group(1), match.group(8)
@@ -222,7 +208,7 @@ def verbalize_digit(value_string: str, alt_one=False) -> str:
    result_symbols = [DIGITS[digit] for digit in value_string]
    result = ''.join(result_symbols)
    if alt_one:
-        result.replace("一", "幺")
+        result = result.replace("一", "幺")
    return result

--- a/paddlespeech/t2s/frontend/zh_normalization/phonecode.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/phonecode.py
@@ -45,11 +45,9 @@ def phone2str(phone_string: str, mobile=True) -> str:
 def replace_phone(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    return phone2str(match.group(0), mobile=False)
@@ -57,11 +55,9 @@ def replace_phone(match) -> str:
 def replace_mobile(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    return phone2str(match.group(0))
--- a/paddlespeech/t2s/frontend/zh_normalization/quantifier.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/quantifier.py
@@ -22,11 +22,9 @@ RE_TEMPERATURE = re.compile(r'(-?)(\d+(\.\d+)?)(°C|℃|度|摄氏度)')
 def replace_temperature(match) -> str:
    """
-    Parameters
+    Args:
-    ----------
+        match (re.Match)
-    match : re.Match
+    Returns:
-    Returns
-    ----------
        str
    """
    sign = match.group(1)

--- a/paddlespeech/t2s/frontend/zh_normalization/text_normlization.py
+++ b/paddlespeech/t2s/frontend/zh_normalization/text_normlization.py
@@ -55,14 +55,10 @@ class TextNormalizer():
    def _split(self, text: str, lang="zh") -> List[str]:
        """Split long text into sentences with sentence-splitting punctuations.
-        Parameters
+        Args:
-        ----------
+            text (str): The input text.
-        text : str
+        Returns:
-            The input text.
+            List[str]: Sentences.
-        Returns
-        -------
-        List[str]
-            Sentences.
        """
        # Only for pure Chinese here
        if lang == "zh":

--- a/paddlespeech/t2s/models/__init__.py
+++ b/paddlespeech/t2s/models/__init__.py
@@ -14,9 +14,9 @@
 from .fastspeech2 import *
 from .hifigan import *
 from .melgan import *
-from .new_tacotron2 import *
 from .parallel_wavegan import *
 from .speedyspeech import *
+from .tacotron2 import *
 from .transformer_tts import *
 from .waveflow import *
 from .wavernn import *
--- a/paddlespeech/t2s/models/fastspeech2/fastspeech2.py
+++ b/paddlespeech/t2s/models/fastspeech2/fastspeech2.py
--- a/paddlespeech/t2s/models/hifigan/hifigan.py
+++ b/paddlespeech/t2s/models/hifigan/hifigan.py
@@ -37,34 +37,20 @@ class HiFiGANGenerator(nn.Layer):
            use_weight_norm: bool=True,
            init_type: str="xavier_uniform", ):
        """Initialize HiFiGANGenerator module.
-        Parameters
+        Args:
-        ----------
+            in_channels (int): Number of input channels.
-        in_channels : int
+            out_channels (int): Number of output channels.
-            Number of input channels.
+            channels (int): Number of hidden representation channels.
-        out_channels : int
+            kernel_size (int): Kernel size of initial and final conv layer.
-            Number of output channels.
+            upsample_scales (list): List of upsampling scales.
-        channels : int
+            upsample_kernel_sizes (list): List of kernel sizes for upsampling layers.
-            Number of hidden representation channels.
+            resblock_kernel_sizes (list): List of kernel sizes for residual blocks.
-        kernel_size : int
+            resblock_dilations (list): List of dilation list for residual blocks.
-            Kernel size of initial and final conv layer.
+            use_additional_convs (bool): Whether to use additional conv layers in residual blocks.
-        upsample_scales : list
+            bias (bool): Whether to add bias parameter in convolution layers.
-            List of upsampling scales.
+            nonlinear_activation (str): Activation function module name.
-        upsample_kernel_sizes : list
+            nonlinear_activation_params (dict): Hyperparameters for activation function.
-            List of kernel sizes for upsampling layers.
+            use_weight_norm (bool): Whether to use weight norm.
-        resblock_kernel_sizes : list
-            List of kernel sizes for residual blocks.
-        resblock_dilations : list
-            List of dilation list for residual blocks.
-        use_additional_convs : bool
-            Whether to use additional conv layers in residual blocks.
-        bias : bool
-            Whether to add bias parameter in convolution layers.
-        nonlinear_activation : str
-            Activation function module name.
-        nonlinear_activation_params : dict
-            Hyperparameters for activation function.
-        use_weight_norm : bool
-            Whether to use weight norm.
                If set to true, it will be applied to all of the conv layers.
        """
        super().__init__()
@@ -134,14 +120,11 @@ class HiFiGANGenerator(nn.Layer):
    def forward(self, c):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        c : Tensor
+            c (Tensor): Input tensor (B, in_channels, T).
-            Input tensor (B, in_channels, T).
+        Returns:
-        Returns
+            Tensor: Output tensor (B, out_channels, T).
-        ----------
-        Tensor
-            Output tensor (B, out_channels, T).
        """
        c = self.input_conv(c)
        for i in range(self.num_upsamples):
@@ -196,14 +179,11 @@ class HiFiGANGenerator(nn.Layer):
    def inference(self, c):
        """Perform inference.
-        Parameters
+        Args:
-        ----------
+            c (Tensor): Input tensor (T, in_channels).
-        c : Tensor 
-            Input tensor (T, in_channels).
                normalize_before (bool): Whether to perform normalization.
-        Returns
+        Returns:
-        ----------
+            Tensor:
-        Tensor
                Output tensor (T ** prod(upsample_scales), out_channels).
        """
        c = self.forward(c.transpose([1, 0]).unsqueeze(0))
@@ -229,35 +209,22 @@ class HiFiGANPeriodDiscriminator(nn.Layer):
            use_spectral_norm: bool=False,
            init_type: str="xavier_uniform", ):
        """Initialize HiFiGANPeriodDiscriminator module.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input channels.
-            Number of input channels.
+            out_channels (int): Number of output channels.
-        out_channels : int
+            period (int): Period.
-            Number of output channels.
+            kernel_sizes (list): Kernel sizes of initial conv layers and the final conv layer.
-        period : int
+            channels (int): Number of initial channels.
-            Period.
+            downsample_scales (list): List of downsampling scales.
-        kernel_sizes : list
+            max_downsample_channels (int): Number of maximum downsampling channels.
-            Kernel sizes of initial conv layers and the final conv layer.
+            use_additional_convs (bool): Whether to use additional conv layers in residual blocks.
-        channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Number of initial channels.
+            nonlinear_activation (str): Activation function module name.
-        downsample_scales : list
+            nonlinear_activation_params (dict): Hyperparameters for activation function.
-            List of downsampling scales.
+            use_weight_norm (bool): Whether to use weight norm.
-        max_downsample_channels : int
-            Number of maximum downsampling channels.
-        use_additional_convs : bool
-            Whether to use additional conv layers in residual blocks.
-        bias : bool
-            Whether to add bias parameter in convolution layers.
-        nonlinear_activation : str
-            Activation function module name.
-        nonlinear_activation_params : dict
-            Hyperparameters for activation function.
-        use_weight_norm : bool
-            Whether to use weight norm.
                If set to true, it will be applied to all of the conv layers.
-        use_spectral_norm : bool
+            use_spectral_norm (bool): Whether to use spectral norm.
-            Whether to use spectral norm.
                If set to true, it will be applied to all of the conv layers.
        """
        super().__init__()
@@ -307,14 +274,11 @@ class HiFiGANPeriodDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        c : Tensor
+            c (Tensor): Input tensor (B, in_channels, T).
-            Input tensor (B, in_channels, T).
+        Returns:
-        Returns
+            list: List of each layer's tensors.
-        ----------
-        list
-            List of each layer's tensors.
        """
        # transform 1d to 2d -> (B, C, T/P, P)
        b, c, t = paddle.shape(x)
@@ -379,12 +343,10 @@ class HiFiGANMultiPeriodDiscriminator(nn.Layer):
            },
            init_type: str="xavier_uniform", ):
        """Initialize HiFiGANMultiPeriodDiscriminator module.
-        Parameters
-        ----------
+        Args:
-        periods : list
+            periods (list): List of periods.
-            List of periods.
+            discriminator_params (dict): Parameters for hifi-gan period discriminator module.
-        discriminator_params : dict
-            Parameters for hifi-gan period discriminator module.
                The period parameter will be overwritten.
        """
        super().__init__()
@@ -399,14 +361,11 @@ class HiFiGANMultiPeriodDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        x : Tensor
+            x (Tensor): Input noise signal (B, 1, T).
-            Input noise signal (B, 1, T).
+        Returns:
-        Returns
+            List: List of list of each discriminator outputs, which consists of each layer output tensors.
-        ----------
-        List
-            List of list of each discriminator outputs, which consists of each layer output tensors.
        """
        outs = []
        for f in self.discriminators:
@@ -434,32 +393,21 @@ class HiFiGANScaleDiscriminator(nn.Layer):
            use_spectral_norm: bool=False,
            init_type: str="xavier_uniform", ):
        """Initilize HiFiGAN scale discriminator module.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input channels.
-            Number of input channels.
+            out_channels (int): Number of output channels.
-        out_channels : int
+            kernel_sizes (list): List of four kernel sizes. The first will be used for the first conv layer,
-            Number of output channels.
-        kernel_sizes : list
-            List of four kernel sizes. The first will be used for the first conv layer,
                and the second is for downsampling part, and the remaining two are for output layers.
-        channels : int
+            channels (int): Initial number of channels for conv layer.
-            Initial number of channels for conv layer.
+            max_downsample_channels (int): Maximum number of channels for downsampling layers.
-        max_downsample_channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Maximum number of channels for downsampling layers.
+            downsample_scales (list): List of downsampling scales.
-        bias : bool
+            nonlinear_activation (str): Activation function module name.
-            Whether to add bias parameter in convolution layers.
+            nonlinear_activation_params (dict): Hyperparameters for activation function.
-        downsample_scales : list
+            use_weight_norm (bool): Whether to use weight norm.
-            List of downsampling scales.
-        nonlinear_activation : str
-            Activation function module name.
-        nonlinear_activation_params : dict
-            Hyperparameters for activation function.
-        use_weight_norm : bool
-            Whether to use weight norm.
                If set to true, it will be applied to all of the conv layers.
-        use_spectral_norm : bool
+            use_spectral_norm (bool): Whether to use spectral norm.
-            Whether to use spectral norm.
                If set to true, it will be applied to all of the conv layers.
        """
        super().__init__()
@@ -546,14 +494,11 @@ class HiFiGANScaleDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        x : Tensor
+            x (Tensor): Input noise signal (B, 1, T).
-            Input noise signal (B, 1, T).
+        Returns:
-        Returns
+            List: List of output tensors of each layer.
-        ----------
-        List
-            List of output tensors of each layer.
        """
        outs = []
        for f in self.layers:
@@ -613,20 +558,14 @@ class HiFiGANMultiScaleDiscriminator(nn.Layer):
            follow_official_norm: bool=False,
            init_type: str="xavier_uniform", ):
        """Initilize HiFiGAN multi-scale discriminator module.
-        Parameters
-        ----------
+        Args:
-        scales : int
+            scales (int): Number of multi-scales.
-            Number of multi-scales.
+            downsample_pooling (str): Pooling module name for downsampling of the inputs.
-        downsample_pooling : str
+            downsample_pooling_params (dict): Parameters for the above pooling module.
-            Pooling module name for downsampling of the inputs.
+            discriminator_params (dict): Parameters for hifi-gan scale discriminator module.
-        downsample_pooling_params : dict
+            follow_official_norm (bool): Whether to follow the norm setting of the official
-            Parameters for the above pooling module.
+                implementaion. The first discriminator uses spectral norm and the other discriminators use weight norm.
-        discriminator_params : dict
-            Parameters for hifi-gan scale discriminator module.
-        follow_official_norm : bool
-            Whether to follow the norm setting of the official
-            implementaion. The first discriminator uses spectral norm and the other
-            discriminators use weight norm.
        """
        super().__init__()
@@ -651,14 +590,11 @@ class HiFiGANMultiScaleDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        x : Tensor
+            x (Tensor): Input noise signal (B, 1, T).
-            Input noise signal (B, 1, T).
+        Returns:
-        Returns
+            List: List of list of each discriminator outputs, which consists of each layer output tensors.
-        ----------
-        List
-            List of list of each discriminator outputs, which consists of each layer output tensors.
        """
        outs = []
        for f in self.discriminators:
@@ -715,23 +651,16 @@ class HiFiGANMultiScaleMultiPeriodDiscriminator(nn.Layer):
            },
            init_type: str="xavier_uniform", ):
        """Initilize HiFiGAN multi-scale + multi-period discriminator module.
-        Parameters
-        ----------
+        Args:
-        scales : int
+            scales (int): Number of multi-scales.
-            Number of multi-scales.
+            scale_downsample_pooling (str): Pooling module name for downsampling of the inputs.
-        scale_downsample_pooling : str
+            scale_downsample_pooling_params (dict): Parameters for the above pooling module.
-            Pooling module name for downsampling of the inputs.
+            scale_discriminator_params (dict): Parameters for hifi-gan scale discriminator module.
-        scale_downsample_pooling_params : dict
+            follow_official_norm （bool): Whether to follow the norm setting of the official implementaion. 
-            Parameters for the above pooling module.
+                The first discriminator uses spectral norm and the other discriminators use weight norm.
-        scale_discriminator_params : dict
+            periods (list): List of periods.
-            Parameters for hifi-gan scale discriminator module.
+            period_discriminator_params (dict): Parameters for hifi-gan period discriminator module.
-        follow_official_norm : bool): Whether to follow the norm setting of the official
-            implementaion. The first discriminator uses spectral norm and the other
-            discriminators use weight norm.
-        periods : list
-            List of periods.
-        period_discriminator_params : dict
-            Parameters for hifi-gan period discriminator module.
                The period parameter will be overwritten.
        """
        super().__init__()
@@ -751,12 +680,10 @@ class HiFiGANMultiScaleMultiPeriodDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        x : Tensor
+            x (Tensor): Input noise signal (B, 1, T).
-            Input noise signal (B, 1, T).
+        Returns:
-        Returns
-        ----------
            List:
                List of list of each discriminator outputs,
                which consists of each layer output tensors.

--- a/paddlespeech/t2s/models/melgan/melgan.py
+++ b/paddlespeech/t2s/models/melgan/melgan.py
@@ -51,41 +51,26 @@ class MelGANGenerator(nn.Layer):
            use_causal_conv: bool=False,
            init_type: str="xavier_uniform", ):
        """Initialize MelGANGenerator module.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input channels.
-            Number of input channels.
+            out_channels (int): Number of output channels,
-        out_channels : int
-            Number of output channels,
                the number of sub-band is out_channels in multi-band melgan.
-        kernel_size : int
+            kernel_size (int): Kernel size of initial and final conv layer.
-            Kernel size of initial and final conv layer.
+            channels (int): Initial number of channels for conv layer.
-        channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Initial number of channels for conv layer.
+            upsample_scales (List[int]): List of upsampling scales.
-        bias : bool
+            stack_kernel_size (int): Kernel size of dilated conv layers in residual stack.
-            Whether to add bias parameter in convolution layers.
+            stacks (int): Number of stacks in a single residual stack.
-        upsample_scales : List[int]
+            nonlinear_activation (Optional[str], optional): Non linear activation in upsample network, by default None
-            List of upsampling scales.
+            nonlinear_activation_params (Dict[str, Any], optional): Parameters passed to the linear activation in the upsample network, 
-        stack_kernel_size : int
-            Kernel size of dilated conv layers in residual stack.
-        stacks : int
-            Number of stacks in a single residual stack.
-        nonlinear_activation : Optional[str], optional
-            Non linear activation in upsample network, by default None
-        nonlinear_activation_params : Dict[str, Any], optional
-            Parameters passed to the linear activation in the upsample network, 
                by default {}
-        pad : str
+            pad (str): Padding function module name before dilated convolution layer.
-            Padding function module name before dilated convolution layer.
+            pad_params （dict): Hyperparameters for padding function.
-        pad_params : dict
+            use_final_nonlinear_activation (nn.Layer): Activation function for the final layer.
-            Hyperparameters for padding function.
+            use_weight_norm (bool): Whether to use weight norm.
-        use_final_nonlinear_activation : nn.Layer
-            Activation function for the final layer.
-        use_weight_norm : bool
-            Whether to use weight norm.
                If set to true, it will be applied to all of the conv layers.
-        use_causal_conv : bool
+            use_causal_conv (bool): Whether to use causal convolution.
-            Whether to use causal convolution.
        """
        super().__init__()
@@ -207,14 +192,11 @@ class MelGANGenerator(nn.Layer):
    def forward(self, c):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        c : Tensor
+            c (Tensor): Input tensor (B, in_channels, T).
-            Input tensor (B, in_channels, T).
+        Returns:
-        Returns
+            Tensor: Output tensor (B, out_channels, T ** prod(upsample_scales)).
-        ----------
-        Tensor
-            Output tensor (B, out_channels, T ** prod(upsample_scales)).
        """
        out = self.melgan(c)
        return out
@@ -260,14 +242,11 @@ class MelGANGenerator(nn.Layer):
    def inference(self, c):
        """Perform inference.
-        Parameters
-        ----------
+        Args:
-        c : Union[Tensor, ndarray]
+            c (Union[Tensor, ndarray]): Input tensor (T, in_channels).
-            Input tensor (T, in_channels).
+        Returns:
-        Returns
+            Tensor: Output tensor (out_channels*T ** prod(upsample_scales), 1).
-        ----------
-        Tensor
-            Output tensor (out_channels*T ** prod(upsample_scales), 1).
        """
        # pseudo batch
        c = c.transpose([1, 0]).unsqueeze(0)
@@ -298,33 +277,22 @@ class MelGANDiscriminator(nn.Layer):
            pad_params: Dict[str, Any]={"mode": "reflect"},
            init_type: str="xavier_uniform", ):
        """Initilize MelGAN discriminator module.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input channels.
-            Number of input channels.
+            out_channels (int): Number of output channels.
-        out_channels : int
+            kernel_sizes (List[int]): List of two kernel sizes. The prod will be used for the first conv layer,
-            Number of output channels.
-        kernel_sizes : List[int]
-            List of two kernel sizes. The prod will be used for the first conv layer,
                and the first and the second kernel sizes will be used for the last two layers.
                For example if kernel_sizes = [5, 3], the first layer kernel size will be 5 * 3 = 15,
                the last two layers' kernel size will be 5 and 3, respectively.
-        channels : int
+            channels (int): Initial number of channels for conv layer.
-            Initial number of channels for conv layer.
+            max_downsample_channels (int): Maximum number of channels for downsampling layers.
-        max_downsample_channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Maximum number of channels for downsampling layers.
+            downsample_scales (List[int]): List of downsampling scales.
-        bias : bool
+            nonlinear_activation (str): Activation function module name.
-            Whether to add bias parameter in convolution layers.
+            nonlinear_activation_params (dict): Hyperparameters for activation function.
-        downsample_scales : List[int]
+            pad (str): Padding function module name before dilated convolution layer.
-            List of downsampling scales.
+            pad_params (dict): Hyperparameters for padding function.
-        nonlinear_activation : str
-            Activation function module name.
-        nonlinear_activation_params : dict
-            Hyperparameters for activation function.
-        pad : str
-            Padding function module name before dilated convolution layer.
-        pad_params : dict
-            Hyperparameters for padding function.
        """
        super().__init__()
@@ -395,14 +363,10 @@ class MelGANDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input noise signal (B, 1, T).
-        x : Tensor
+        Returns:
-            Input noise signal (B, 1, T).
+            List: List of output tensors of each layer (for feat_match_loss).
-        Returns
-        ----------
-        List
-            List of output tensors of each layer (for feat_match_loss).
        """
        outs = []
        for f in self.layers:
@@ -440,39 +404,24 @@ class MelGANMultiScaleDiscriminator(nn.Layer):
            use_weight_norm: bool=True,
            init_type: str="xavier_uniform", ):
        """Initilize MelGAN multi-scale discriminator module.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input channels.
-            Number of input channels.
+            out_channels (int): Number of output channels.
-        out_channels : int
+            scales (int): Number of multi-scales.
-            Number of output channels.
+            downsample_pooling (str): Pooling module name for downsampling of the inputs.
-        scales : int
+            downsample_pooling_params (dict): Parameters for the above pooling module.
-            Number of multi-scales.
+            kernel_sizes (List[int]): List of two kernel sizes. The sum will be used for the first conv layer,
-        downsample_pooling : str
-            Pooling module name for downsampling of the inputs.
-        downsample_pooling_params : dict
-            Parameters for the above pooling module.
-        kernel_sizes : List[int]
-            List of two kernel sizes. The sum will be used for the first conv layer,
                and the first and the second kernel sizes will be used for the last two layers.
-        channels : int
+            channels (int): Initial number of channels for conv layer.
-            Initial number of channels for conv layer.
+            max_downsample_channels (int): Maximum number of channels for downsampling layers.
-        max_downsample_channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Maximum number of channels for downsampling layers.
+            downsample_scales (List[int]): List of downsampling scales.
-        bias : bool
+            nonlinear_activation (str): Activation function module name.
-            Whether to add bias parameter in convolution layers.
+            nonlinear_activation_params (dict): Hyperparameters for activation function.
-        downsample_scales : List[int]
+            pad (str): Padding function module name before dilated convolution layer.
-            List of downsampling scales.
+            pad_params (dict): Hyperparameters for padding function.
-        nonlinear_activation : str
+            use_causal_conv (bool): Whether to use causal convolution.
-            Activation function module name.
-        nonlinear_activation_params : dict
-            Hyperparameters for activation function.
-        pad : str
-            Padding function module name before dilated convolution layer.
-        pad_params : dict
-            Hyperparameters for padding function.
-        use_causal_conv : bool
-            Whether to use causal convolution.
        """
        super().__init__()
@@ -514,14 +463,10 @@ class MelGANMultiScaleDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input noise signal (B, 1, T).
-        x : Tensor
+        Returns:
-            Input noise signal (B, 1, T).
+            List: List of list of each discriminator outputs, which consists of each layer output tensors.
-        Returns
-        ----------
-        List
-            List of list of each discriminator outputs, which consists of each layer output tensors.
        """
        outs = []
        for f in self.discriminators:

--- a/paddlespeech/t2s/models/melgan/style_melgan.py
+++ b/paddlespeech/t2s/models/melgan/style_melgan.py
@@ -52,36 +52,22 @@ class StyleMelGANGenerator(nn.Layer):
            use_weight_norm: bool=True,
            init_type: str="xavier_uniform", ):
        """Initilize Style MelGAN generator.
-        Parameters
-        ----------
+        Args:
-        in_channels : int
+            in_channels (int): Number of input noise channels.
-            Number of input noise channels.
+            aux_channels (int): Number of auxiliary input channels.
-        aux_channels : int
+            channels (int): Number of channels for conv layer.
-            Number of auxiliary input channels.
+            out_channels (int): Number of output channels.
-        channels : int
+            kernel_size (int): Kernel size of conv layers.
-            Number of channels for conv layer.
+            dilation (int): Dilation factor for conv layers.
-        out_channels : int
+            bias (bool): Whether to add bias parameter in convolution layers.
-            Number of output channels.
+            noise_upsample_scales (list): List of noise upsampling scales.
-        kernel_size : int
+            noise_upsample_activation (str): Activation function module name for noise upsampling.
-            Kernel size of conv layers.
+            noise_upsample_activation_params (dict): Hyperparameters for the above activation function.
-        dilation : int
+            upsample_scales (list): List of upsampling scales.
-            Dilation factor for conv layers.
+            upsample_mode (str): Upsampling mode in TADE layer.
-        bias : bool
+            gated_function (str): Gated function in TADEResBlock ("softmax" or "sigmoid").
-            Whether to add bias parameter in convolution layers.
+            use_weight_norm (bool): Whether to use weight norm.
-        noise_upsample_scales : list
-            List of noise upsampling scales.
-        noise_upsample_activation : str
-            Activation function module name for noise upsampling.
-        noise_upsample_activation_params : dict
-            Hyperparameters for the above activation function.
-        upsample_scales : list
-            List of upsampling scales.
-        upsample_mode : str
-            Upsampling mode in TADE layer.
-        gated_function : str
-            Gated function in TADEResBlock ("softmax" or "sigmoid").
-        use_weight_norm : bool
-            Whether to use weight norm.
                If set to true, it will be applied to all of the conv layers.
        """
        super().__init__()
@@ -147,16 +133,12 @@ class StyleMelGANGenerator(nn.Layer):
    def forward(self, c, z=None):
        """Calculate forward propagation.
-        Parameters
-        ----------
+        Args:
-        c : Tensor
+            c (Tensor): Auxiliary input tensor (B, channels, T).
-            Auxiliary input tensor (B, channels, T).
+            z (Tensor): Input noise tensor (B, in_channels, 1).
-        z : Tensor
+        Returns:
-            Input noise tensor (B, in_channels, 1).
+            Tensor: Output tensor (B, out_channels, T ** prod(upsample_scales)).
-        Returns
-        ----------
-        Tensor
-            Output tensor (B, out_channels, T ** prod(upsample_scales)).
        """
        # batch_max_steps(24000) == noise_upsample_factor(80) * upsample_factor(300)
        if z is None:
@@ -211,14 +193,10 @@ class StyleMelGANGenerator(nn.Layer):
    def inference(self, c):
        """Perform inference.
-        Parameters
+        Args:
-        ----------
+            c (Tensor): Input tensor (T, in_channels).
-        c : Tensor
+        Returns:
-            Input tensor (T, in_channels).
+            Tensor: Output tensor (T ** prod(upsample_scales), out_channels).
-        Returns
-        ----------
-        Tensor
-            Output tensor (T ** prod(upsample_scales), out_channels).
        """
        # (1, in_channels, T)
        c = c.transpose([1, 0]).unsqueeze(0)
@@ -278,18 +256,13 @@ class StyleMelGANDiscriminator(nn.Layer):
            use_weight_norm: bool=True,
            init_type: str="xavier_uniform", ):
        """Initilize Style MelGAN discriminator.
-        Parameters
-        ----------
+        Args:
-        repeats : int
+            repeats (int): Number of repititons to apply RWD.
-            Number of repititons to apply RWD.
+            window_sizes (list): List of random window sizes.
-        window_sizes : list
+            pqmf_params (list): List of list of Parameters for PQMF modules
-            List of random window sizes.
+            discriminator_params (dict): Parameters for base discriminator module.
-        pqmf_params : list
+            use_weight_nom (bool): Whether to apply weight normalization.
-            List of list of Parameters for PQMF modules
-        discriminator_params : dict
-            Parameters for base discriminator module.
-        use_weight_nom : bool
-            Whether to apply weight normalization.
        """
        super().__init__()
@@ -325,14 +298,10 @@ class StyleMelGANDiscriminator(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input tensor (B, 1, T).
-        x : Tensor
+        Returns:
-            Input tensor (B, 1, T).
+            List: List of discriminator outputs, #items in the list will be
-        Returns
-        ----------
-        List
-            List of discriminator outputs, #items in the list will be
                equal to repeats * #discriminators.
        """
        outs = []

--- a/paddlespeech/t2s/models/parallel_wavegan/parallel_wavegan.py
+++ b/paddlespeech/t2s/models/parallel_wavegan/parallel_wavegan.py
@@ -31,51 +31,30 @@ from paddlespeech.t2s.modules.upsample import ConvInUpsampleNet
 class PWGGenerator(nn.Layer):
    """Wave Generator for Parallel WaveGAN
-    Parameters
+    Args:
-    ----------
+        in_channels (int, optional): Number of channels of the input waveform, by default 1
-    in_channels : int, optional
+        out_channels (int, optional): Number of channels of the output waveform, by default 1
-        Number of channels of the input waveform, by default 1
+        kernel_size (int, optional): Kernel size of the residual blocks inside, by default 3
-    out_channels : int, optional
+        layers (int, optional): Number of residual blocks inside, by default 30
-        Number of channels of the output waveform, by default 1
+        stacks (int, optional): The number of groups to split the residual blocks into, by default 3
-    kernel_size : int, optional
+            Within each group, the dilation of the residual block grows exponentially.
-        Kernel size of the residual blocks inside, by default 3
+        residual_channels (int, optional): Residual channel of the residual blocks, by default 64
-    layers : int, optional
+        gate_channels (int, optional): Gate channel of the residual blocks, by default 128
-        Number of residual blocks inside, by default 30
+        skip_channels (int, optional): Skip channel of the residual blocks, by default 64
-    stacks : int, optional
+        aux_channels (int, optional): Auxiliary channel of the residual blocks, by default 80
-        The number of groups to split the residual blocks into, by default 3
+        aux_context_window (int, optional): The context window size of the first convolution applied to the 
-        Within each group, the dilation of the residual block grows 
-        exponentially.
-    residual_channels : int, optional
-        Residual channel of the residual blocks, by default 64
-    gate_channels : int, optional
-        Gate channel of the residual blocks, by default 128
-    skip_channels : int, optional
-        Skip channel of the residual blocks, by default 64
-    aux_channels : int, optional
-        Auxiliary channel of the residual blocks, by default 80
-    aux_context_window : int, optional
-        The context window size of the first convolution applied to the 
            auxiliary input, by default 2
-    dropout : float, optional
+        dropout (float, optional): Dropout of the residual blocks, by default 0.
-        Dropout of the residual blocks, by default 0.
+        bias (bool, optional): Whether to use bias in residual blocks, by default True
-    bias : bool, optional
+        use_weight_norm (bool, optional): Whether to use weight norm in all convolutions, by default True
-        Whether to use bias in residual blocks, by default True
+        use_causal_conv (bool, optional): Whether to use causal padding in the upsample network and residual 
-    use_weight_norm : bool, optional
-        Whether to use weight norm in all convolutions, by default True
-    use_causal_conv : bool, optional
-        Whether to use causal padding in the upsample network and residual 
            blocks, by default False
-    upsample_scales : List[int], optional
+        upsample_scales (List[int], optional): Upsample scales of the upsample network, by default [4, 4, 4, 4]
-        Upsample scales of the upsample network, by default [4, 4, 4, 4]
+        nonlinear_activation (Optional[str], optional): Non linear activation in upsample network, by default None
-    nonlinear_activation : Optional[str], optional
+        nonlinear_activation_params (Dict[str, Any], optional): Parameters passed to the linear activation in the upsample network, 
-        Non linear activation in upsample network, by default None
-    nonlinear_activation_params : Dict[str, Any], optional
-        Parameters passed to the linear activation in the upsample network, 
            by default {}
-    interpolate_mode : str, optional
+        interpolate_mode (str, optional): Interpolation mode of the upsample network, by default "nearest"
-        Interpolation mode of the upsample network, by default "nearest"
+        freq_axis_kernel_size (int, optional): Kernel size along the frequency axis of the upsample network, by default 1
-    freq_axis_kernel_size : int, optional
-        Kernel size along the frequency axis of the upsample network, by default 1
    """
    def __init__(
@@ -167,18 +146,13 @@ class PWGGenerator(nn.Layer):
    def forward(self, x, c):
        """Generate waveform.
-        Parameters
+        Args:
-        ----------
+            x(Tensor): Shape (N, C_in, T), The input waveform.
-        x : Tensor
+            c(Tensor): Shape (N, C_aux, T'). The auxiliary input (e.g. spectrogram). It
-            Shape (N, C_in, T), The input waveform.
-        c : Tensor
-            Shape (N, C_aux, T'). The auxiliary input (e.g. spectrogram). It 
            is upsampled to match the time resolution of the input.
-        Returns
+        Returns:
-        -------
+            Tensor: Shape (N, C_out, T), the generated waveform.
-        Tensor
-            Shape (N, C_out, T), the generated waveform.
        """
        c = self.upsample_net(c)
        assert c.shape[-1] == x.shape[-1]
@@ -218,19 +192,14 @@ class PWGGenerator(nn.Layer):
        self.apply(_remove_weight_norm)
    def inference(self, c=None):
-        """Waveform generation. This function is used for single instance 
+        """Waveform generation. This function is used for single instance inference.
-        inference.
-        Parameters
+        Args:
-        ----------
+            c(Tensor, optional, optional): Shape (T', C_aux), the auxiliary input, by default None
-        c : Tensor, optional
+            x(Tensor, optional): Shape (T, C_in), the noise waveform, by default None
-            Shape (T', C_aux), the auxiliary input, by default None
-        x : Tensor, optional
+        Returns:
-            Shape (T, C_in), the noise waveform, by default None
+            Tensor: Shape (T, C_out), the generated waveform
-            If not provided, a sample is drawn from a gaussian distribution.
-        Returns
-        -------
-        Tensor
-            Shape (T, C_out), the generated waveform
        """
        # when to static, can not input x, see https://github.com/PaddlePaddle/Parakeet/pull/132/files
        x = paddle.randn(
@@ -244,31 +213,20 @@ class PWGGenerator(nn.Layer):
 class PWGDiscriminator(nn.Layer):
    """A convolutional discriminator for audio.
-    Parameters
+    Args:
-    ----------
+        in_channels (int, optional): Number of channels of the input audio, by default 1
-    in_channels : int, optional
+        out_channels (int, optional): Output feature size, by default 1
-        Number of channels of the input audio, by default 1
+        kernel_size (int, optional): Kernel size of convolutional sublayers, by default 3
-    out_channels : int, optional
+        layers (int, optional): Number of layers, by default 10
-        Output feature size, by default 1
+        conv_channels (int, optional): Feature size of the convolutional sublayers, by default 64
-    kernel_size : int, optional
+        dilation_factor (int, optional): The factor with which dilation of each convolutional sublayers grows 
-        Kernel size of convolutional sublayers, by default 3
+            exponentially if it is greater than 1, else the dilation of each convolutional sublayers grows linearly, 
-    layers : int, optional
+            by default 1
-        Number of layers, by default 10
+        nonlinear_activation (str, optional): The activation after each convolutional sublayer, by default "leakyrelu"
-    conv_channels : int, optional
+        nonlinear_activation_params (Dict[str, Any], optional): The parameters passed to the activation's initializer, by default 
-        Feature size of the convolutional sublayers, by default 64
-    dilation_factor : int, optional
-        The factor with which dilation of each convolutional sublayers grows 
-        exponentially if it is greater than 1, else the dilation of each 
-        convolutional sublayers grows linearly, by default 1
-    nonlinear_activation : str, optional
-        The activation after each convolutional sublayer, by default "leakyrelu"
-    nonlinear_activation_params : Dict[str, Any], optional
-        The parameters passed to the activation's initializer, by default 
            {"negative_slope": 0.2}
-    bias : bool, optional
+        bias (bool, optional): Whether to use bias in convolutional sublayers, by default True
-        Whether to use bias in convolutional sublayers, by default True
+        use_weight_norm (bool, optional): Whether to use weight normalization at all convolutional sublayers, 
-    use_weight_norm : bool, optional
-        Whether to use weight normalization at all convolutional sublayers, 
            by default True
    """
@@ -330,15 +288,12 @@ class PWGDiscriminator(nn.Layer):
    def forward(self, x):
        """
-        Parameters
-        ----------
+        Args:
-        x : Tensor
+            x (Tensor): Shape (N, in_channels, num_samples), the input audio.
-            Shape (N, in_channels, num_samples), the input audio.
+        Returns:
-        Returns
+            Tensor: Shape (N, out_channels, num_samples), the predicted logits.
-        -------
-        Tensor
-            Shape (N, out_channels, num_samples), the predicted logits.
        """
        return self.conv_layers(x)
@@ -362,39 +317,25 @@ class PWGDiscriminator(nn.Layer):
 class ResidualPWGDiscriminator(nn.Layer):
    """A wavenet-style discriminator for audio.
-    Parameters
+    Args:
-    ----------
+        in_channels (int, optional): Number of channels of the input audio, by default 1
-    in_channels : int, optional
+        out_channels (int, optional): Output feature size, by default 1
-        Number of channels of the input audio, by default 1
+        kernel_size (int, optional): Kernel size of residual blocks, by default 3
-    out_channels : int, optional
+        layers (int, optional): Number of residual blocks, by default 30
-        Output feature size, by default 1
+        stacks (int, optional): Number of groups of residual blocks, within which the dilation 
-    kernel_size : int, optional
-        Kernel size of residual blocks, by default 3
-    layers : int, optional
-        Number of residual blocks, by default 30
-    stacks : int, optional
-        Number of groups of residual blocks, within which the dilation 
            of each residual blocks grows exponentially, by default 3
-    residual_channels : int, optional
+        residual_channels (int, optional): Residual channels of residual blocks, by default 64
-        Residual channels of residual blocks, by default 64
+        gate_channels (int, optional): Gate channels of residual blocks, by default 128
-    gate_channels : int, optional
+        skip_channels (int, optional): Skip channels of residual blocks, by default 64
-        Gate channels of residual blocks, by default 128
+        dropout (float, optional): Dropout probability of residual blocks, by default 0.
-    skip_channels : int, optional
+        bias (bool, optional): Whether to use bias in residual blocks, by default True
-        Skip channels of residual blocks, by default 64
+        use_weight_norm (bool, optional): Whether to use weight normalization in all convolutional layers, 
-    dropout : float, optional
-        Dropout probability of residual blocks, by default 0.
-    bias : bool, optional
-        Whether to use bias in residual blocks, by default True
-    use_weight_norm : bool, optional
-        Whether to use weight normalization in all convolutional layers, 
            by default True
-    use_causal_conv : bool, optional
+        use_causal_conv (bool, optional): Whether to use causal convolution in residual blocks, by default False
-        Whether to use causal convolution in residual blocks, by default False
+        nonlinear_activation (str, optional): Activation after convolutions other than those in residual blocks, 
-    nonlinear_activation : str, optional
-        Activation after convolutions other than those in residual blocks, 
            by default "leakyrelu"
-    nonlinear_activation_params : Dict[str, Any], optional
+        nonlinear_activation_params (Dict[str, Any], optional): Parameters to pass to the activation, 
-        Parameters to pass to the activation, by default {"negative_slope": 0.2}
+            by default {"negative_slope": 0.2}
    """
    def __init__(
@@ -463,15 +404,11 @@ class ResidualPWGDiscriminator(nn.Layer):
    def forward(self, x):
        """
-        Parameters
+        Args:
-        ----------
+            x(Tensor): Shape (N, in_channels, num_samples), the input audio.↩
-        x : Tensor
-            Shape (N, in_channels, num_samples), the input audio.
+        Returns:
+            Tensor: Shape (N, out_channels, num_samples), the predicted logits.
-        Returns
-        -------
-        Tensor
-            Shape (N, out_channels, num_samples), the predicted logits.
        """
        x = self.first_conv(x)
        skip = 0

--- a/paddlespeech/t2s/models/new_tacotron2/__init__.py
+++ b/paddlespeech/t2s/models/new_tacotron2/__init__.py
--- a/paddlespeech/t2s/models/new_tacotron2/tacotron2.py
+++ b/paddlespeech/t2s/models/new_tacotron2/tacotron2.py
@@ -81,69 +81,39 @@ class Tacotron2(nn.Layer):
            # training related
            init_type: str="xavier_uniform", ):
        """Initialize Tacotron2 module.
-        Parameters
+        Args:
-        ----------
+            idim (int): Dimension of the inputs.
-        idim : int
+            odim (int): Dimension of the outputs.
-            Dimension of the inputs.
+            embed_dim (int): Dimension of the token embedding.
-        odim : int
+            elayers (int): Number of encoder blstm layers.
-            Dimension of the outputs.
+            eunits (int): Number of encoder blstm units.
-        embed_dim : int
+            econv_layers (int): Number of encoder conv layers.
-            Dimension of the token embedding.
+            econv_filts (int): Number of encoder conv filter size.
-        elayers : int
+            econv_chans (int): Number of encoder conv filter channels.
-            Number of encoder blstm layers.
+            dlayers (int): Number of decoder lstm layers.
-        eunits : int
+            dunits (int): Number of decoder lstm units.
-            Number of encoder blstm units.
+            prenet_layers (int): Number of prenet layers.
-        econv_layers : int
+            prenet_units (int): Number of prenet units.
-            Number of encoder conv layers.
+            postnet_layers (int): Number of postnet layers.
-        econv_filts : int
+            postnet_filts (int): Number of postnet filter size.
-            Number of encoder conv filter size.
+            postnet_chans (int): Number of postnet filter channels.
-        econv_chans : int
+            output_activation (str): Name of activation function for outputs.
-            Number of encoder conv filter channels.
+            adim (int): Number of dimension of mlp in attention.
-        dlayers : int
+            aconv_chans (int): Number of attention conv filter channels.
-            Number of decoder lstm layers.
+            aconv_filts (int): Number of attention conv filter size.
-        dunits : int
+            cumulate_att_w (bool): Whether to cumulate previous attention weight.
-            Number of decoder lstm units.
+            use_batch_norm (bool): Whether to use batch normalization.
-        prenet_layers : int
+            use_concate (bool): Whether to concat enc outputs w/ dec lstm outputs.
-            Number of prenet layers.
+            reduction_factor (int): Reduction factor.
-        prenet_units : int
+            spk_num (Optional[int]): Number of speakers. If set to > 1, assume that the
-            Number of prenet units.
-        postnet_layers : int
-            Number of postnet layers.
-        postnet_filts : int
-            Number of postnet filter size.
-        postnet_chans : int
-            Number of postnet filter channels.
-        output_activation : str
-            Name of activation function for outputs.
-        adim : int
-            Number of dimension of mlp in attention.
-        aconv_chans : int
-            Number of attention conv filter channels.
-        aconv_filts : int
-            Number of attention conv filter size.
-        cumulate_att_w : bool
-            Whether to cumulate previous attention weight.
-        use_batch_norm : bool
-            Whether to use batch normalization.
-        use_concate : bool
-            Whether to concat enc outputs w/ dec lstm outputs.
-        reduction_factor : int
-            Reduction factor.
-        spk_num : Optional[int]
-            Number of speakers. If set to > 1, assume that the
                sids will be provided as the input and use sid embedding layer.
-        lang_num : Optional[int]
+            lang_num (Optional[int]): Number of languages. If set to > 1, assume that the
-            Number of languages. If set to > 1, assume that the
                lids will be provided as the input and use sid embedding layer.
-        spk_embed_dim : Optional[int]
+            spk_embed_dim (Optional[int]): Speaker embedding dimension. If set to > 0,
-            Speaker embedding dimension. If set to > 0,
                assume that spk_emb will be provided as the input.
-        spk_embed_integration_type : str
+            spk_embed_integration_type (str): How to integrate speaker embedding.
-            How to integrate speaker embedding.
+            dropout_rate (float): Dropout rate.
-        dropout_rate : float
+            zoneout_rate (float): Zoneout rate.
-            Dropout rate.
-        zoneout_rate : float
-            Zoneout rate.
        """
        assert check_argument_types()
        super().__init__()
@@ -258,31 +228,19 @@ class Tacotron2(nn.Layer):
    ) -> Tuple[paddle.Tensor, Dict[str, paddle.Tensor], paddle.Tensor]:
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            text (Tensor(int64)): Batch of padded character ids (B, T_text).
-        text : Tensor(int64)
+            text_lengths (Tensor(int64)): Batch of lengths of each input batch (B,).
-            Batch of padded character ids (B, T_text).
+            speech (Tensor): Batch of padded target features (B, T_feats, odim).
-        text_lengths : Tensor(int64)
+            speech_lengths (Tensor(int64)): Batch of the lengths of each target (B,).
-            Batch of lengths of each input batch (B,).
+            spk_emb (Optional[Tensor]): Batch of speaker embeddings (B, spk_embed_dim).
-        speech : Tensor
+            spk_id (Optional[Tensor]): Batch of speaker IDs (B, 1).
-            Batch of padded target features (B, T_feats, odim).
+            lang_id (Optional[Tensor]): Batch of language IDs (B, 1).
-        speech_lengths : Tensor(int64)
-            Batch of the lengths of each target (B,).
+        Returns:
-        spk_emb : Optional[Tensor]
+            Tensor: Loss scalar value.
-            Batch of speaker embeddings (B, spk_embed_dim).
+            Dict: Statistics to be monitored.
-        spk_id : Optional[Tensor]
+            Tensor: Weight value if not joint training else model outputs.
-            Batch of speaker IDs (B, 1).
-        lang_id : Optional[Tensor]
-            Batch of language IDs (B, 1).
-        Returns
-        ----------
-        Tensor
-            Loss scalar value.
-        Dict
-            Statistics to be monitored.
-        Tensor
-            Weight value if not joint training else model outputs.
        """
        text = text[:, :text_lengths.max()]
@@ -369,35 +327,21 @@ class Tacotron2(nn.Layer):
            use_teacher_forcing: bool=False, ) -> Dict[str, paddle.Tensor]:
        """Generate the sequence of features given the sequences of characters.
-        Parameters
+        Args:
-        ----------
+            text (Tensor(int64)): Input sequence of characters (T_text,).
-        text Tensor(int64)
+            speech (Optional[Tensor]): Feature sequence to extract style (N, idim).
-            Input sequence of characters (T_text,).
+            spk_emb (ptional[Tensor]): Speaker embedding (spk_embed_dim,).
-        speech : Optional[Tensor]
+            spk_id (Optional[Tensor]): Speaker ID (1,).
-            Feature sequence to extract style (N, idim).
+            lang_id (Optional[Tensor]): Language ID (1,).
-        spk_emb : ptional[Tensor]
+            threshold (float): Threshold in inference.
-            Speaker embedding (spk_embed_dim,).
+            minlenratio (float): Minimum length ratio in inference.
-        spk_id : Optional[Tensor]
+            maxlenratio (float): Maximum length ratio in inference.
-            Speaker ID (1,).
+            use_att_constraint (bool): Whether to apply attention constraint.
-        lang_id : Optional[Tensor]
+            backward_window (int): Backward window in attention constraint.
-            Language ID (1,).
+            forward_window (int): Forward window in attention constraint.
-        threshold : float
+            use_teacher_forcing (bool): Whether to use teacher forcing.
-            Threshold in inference.
-        minlenratio : float
+        Returns:
-            Minimum length ratio in inference.
-        maxlenratio : float
-            Maximum length ratio in inference.
-        use_att_constraint : bool
-            Whether to apply attention constraint.
-        backward_window : int
-            Backward window in attention constraint.
-        forward_window : int
-            Forward window in attention constraint.
-        use_teacher_forcing : bool
-            Whether to use teacher forcing.
-        Return
-        ----------
            Dict[str, Tensor]
            Output dict including the following items:
                * feat_gen (Tensor): Output sequence of features (T_feats, odim).
@@ -458,17 +402,12 @@ class Tacotron2(nn.Layer):
                                  spk_emb: paddle.Tensor) -> paddle.Tensor:
        """Integrate speaker embedding with hidden states.
-        Parameters
+        Args:
-        ----------
+            hs (Tensor): Batch of hidden state sequences (B, Tmax, eunits).
-         hs : Tensor
+            spk_emb (Tensor): Batch of speaker embeddings (B, spk_embed_dim).
-            Batch of hidden state sequences (B, Tmax, eunits).
-         spk_emb : Tensor
+        Returns:
-            Batch of speaker embeddings (B, spk_embed_dim).
+            Tensor: Batch of integrated hidden state sequences (B, Tmax, eunits) if
-        Returns
-        ----------
-         Tensor
-            Batch of integrated hidden state sequences (B, Tmax, eunits) if
                integration_type is "add" else (B, Tmax, eunits + spk_embed_dim).
        """

--- a/paddlespeech/t2s/models/new_tacotron2/tacotron2_updater.py
+++ b/paddlespeech/t2s/models/new_tacotron2/tacotron2_updater.py
--- a/paddlespeech/t2s/models/transformer_tts/transformer_tts.py
+++ b/paddlespeech/t2s/models/transformer_tts/transformer_tts.py
@@ -48,126 +48,66 @@ class TransformerTTS(nn.Layer):
    .. _`Neural Speech Synthesis with Transformer Network`:
        https://arxiv.org/pdf/1809.08895.pdf
-    Parameters
+    Args:
-    ----------
+        idim (int): Dimension of the inputs.
-    idim : int
+        odim (int): Dimension of the outputs.
-        Dimension of the inputs.
+        embed_dim (int, optional): Dimension of character embedding.
-    odim : int
+        eprenet_conv_layers (int, optional): Number of encoder prenet convolution layers.
-        Dimension of the outputs.
+        eprenet_conv_chans (int, optional): Number of encoder prenet convolution channels.
-    embed_dim : int, optional
+        eprenet_conv_filts (int, optional): Filter size of encoder prenet convolution.
-        Dimension of character embedding.
+        dprenet_layers (int, optional): Number of decoder prenet layers.
-    eprenet_conv_layers : int, optional
+        dprenet_units (int, optional): Number of decoder prenet hidden units.
-        Number of encoder prenet convolution layers.
+        elayers (int, optional): Number of encoder layers.
-    eprenet_conv_chans : int, optional
+        eunits (int, optional): Number of encoder hidden units.
-        Number of encoder prenet convolution channels.
+        adim (int, optional): Number of attention transformation dimensions.
-    eprenet_conv_filts : int, optional
+        aheads (int, optional): Number of heads for multi head attention.
-        Filter size of encoder prenet convolution.
+        dlayers (int, optional): Number of decoder layers.
-    dprenet_layers : int, optional
+        dunits (int, optional): Number of decoder hidden units.
-        Number of decoder prenet layers.
+        postnet_layers (int, optional): Number of postnet layers.
-    dprenet_units : int, optional
+        postnet_chans (int, optional): Number of postnet channels.
-        Number of decoder prenet hidden units.
+        postnet_filts (int, optional): Filter size of postnet.
-    elayers : int, optional
+        use_scaled_pos_enc (pool, optional): Whether to use trainable scaled positional encoding.
-        Number of encoder layers.
+        use_batch_norm (bool, optional): Whether to use batch normalization in encoder prenet.
-    eunits : int, optional
+        encoder_normalize_before (bool, optional): Whether to perform layer normalization before encoder block.
-        Number of encoder hidden units.
+        decoder_normalize_before (bool, optional): Whether to perform layer normalization before decoder block.
-    adim : int, optional
+        encoder_concat_after (bool, optional): Whether to concatenate attention layer's input and output in encoder.
-        Number of attention transformation dimensions.
+        decoder_concat_after (bool, optional): Whether to concatenate attention layer's input and output in decoder.
-    aheads : int, optional
+        positionwise_layer_type (str, optional): Position-wise operation type.
-        Number of heads for multi head attention.
+        positionwise_conv_kernel_size (int, optional): Kernel size in position wise conv 1d.
-    dlayers : int, optional
+        reduction_factor (int, optional): Reduction factor.
-        Number of decoder layers.
+        spk_embed_dim (int, optional): Number of speaker embedding dimenstions.
-    dunits : int, optional
+        spk_embed_integration_type (str, optional): How to integrate speaker embedding.
-        Number of decoder hidden units.
+        use_gst (str, optional): Whether to use global style token.
-    postnet_layers : int, optional
+        gst_tokens (int, optional): The number of GST embeddings.
-        Number of postnet layers.
+        gst_heads (int, optional): The number of heads in GST multihead attention.
-    postnet_chans : int, optional
+        gst_conv_layers (int, optional): The number of conv layers in GST.
-        Number of postnet channels.
+        gst_conv_chans_list (Sequence[int], optional): List of the number of channels of conv layers in GST.
-    postnet_filts : int, optional
+        gst_conv_kernel_size (int, optional): Kernal size of conv layers in GST.
-        Filter size of postnet.
+        gst_conv_stride (int, optional): Stride size of conv layers in GST.
-    use_scaled_pos_enc : pool, optional
+        gst_gru_layers (int, optional): The number of GRU layers in GST.
-        Whether to use trainable scaled positional encoding.
+        gst_gru_units (int, optional): The number of GRU units in GST.
-    use_batch_norm : bool, optional
+        transformer_lr (float, optional): Initial value of learning rate.
-        Whether to use batch normalization in encoder prenet.
+        transformer_warmup_steps (int, optional): Optimizer warmup steps.
-    encoder_normalize_before : bool, optional
+        transformer_enc_dropout_rate (float, optional): Dropout rate in encoder except attention and positional encoding.
-        Whether to perform layer normalization before encoder block.
+        transformer_enc_positional_dropout_rate (float, optional): Dropout rate after encoder positional encoding.
-    decoder_normalize_before : bool, optional
+        transformer_enc_attn_dropout_rate （float, optional): Dropout rate in encoder self-attention module.
-        Whether to perform layer normalization before decoder block.
+        transformer_dec_dropout_rate (float, optional): Dropout rate in decoder except attention & positional encoding.
-    encoder_concat_after : bool, optional
+        transformer_dec_positional_dropout_rate (float, optional): Dropout rate after decoder positional encoding.
-        Whether to concatenate attention layer's input and output in encoder.
+        transformer_dec_attn_dropout_rate （float, optional): Dropout rate in deocoder self-attention module.
-    decoder_concat_after : bool, optional
+        transformer_enc_dec_attn_dropout_rate (float, optional): Dropout rate in encoder-deocoder attention module.
-        Whether to concatenate attention layer's input and output in decoder.
+        init_type (str, optional): How to initialize transformer parameters.
-    positionwise_layer_type : str, optional
+        init_enc_alpha （float, optional）: Initial value of alpha in scaled pos encoding of the encoder.
-        Position-wise operation type.
+        init_dec_alpha (float, optional): Initial value of alpha in scaled pos encoding of the decoder.
-    positionwise_conv_kernel_size : int, optional
+        eprenet_dropout_rate (float, optional): Dropout rate in encoder prenet.
-        Kernel size in position wise conv 1d.
+        dprenet_dropout_rate (float, optional): Dropout rate in decoder prenet.
-    reduction_factor : int, optional
+        postnet_dropout_rate (float, optional): Dropout rate in postnet.
-        Reduction factor.
+        use_masking (bool, optional): Whether to apply masking for padded part in loss calculation.
-    spk_embed_dim : int, optional
+        use_weighted_masking (bool, optional): Whether to apply weighted masking in loss calculation.
-        Number of speaker embedding dimenstions.
+        bce_pos_weight (float, optional): Positive sample weight in bce calculation (only for use_masking=true).
-    spk_embed_integration_type : str, optional
+        loss_type (str, optional): How to calculate loss.
-        How to integrate speaker embedding.
+        use_guided_attn_loss (bool, optional): Whether to use guided attention loss.
-    use_gst : str, optional
+        num_heads_applied_guided_attn (int, optional): Number of heads in each layer to apply guided attention loss.
-        Whether to use global style token.
+        num_layers_applied_guided_attn (int, optional): Number of layers to apply guided attention loss.
-    gst_tokens : int, optional
-        The number of GST embeddings.
-    gst_heads : int, optional
-        The number of heads in GST multihead attention.
-    gst_conv_layers : int, optional
-        The number of conv layers in GST.
-    gst_conv_chans_list : Sequence[int], optional
-            List of the number of channels of conv layers in GST.
-    gst_conv_kernel_size : int, optional
-        Kernal size of conv layers in GST.
-    gst_conv_stride : int, optional
-        Stride size of conv layers in GST.
-    gst_gru_layers : int, optional
-        The number of GRU layers in GST.
-    gst_gru_units : int, optional
-        The number of GRU units in GST.
-    transformer_lr : float, optional
-        Initial value of learning rate.
-    transformer_warmup_steps : int, optional
-        Optimizer warmup steps.
-    transformer_enc_dropout_rate : float, optional
-        Dropout rate in encoder except attention and positional encoding.
-    transformer_enc_positional_dropout_rate : float, optional
-        Dropout rate after encoder positional encoding.
-    transformer_enc_attn_dropout_rate : float, optional
-        Dropout rate in encoder self-attention module.
-    transformer_dec_dropout_rate : float, optional
-        Dropout rate in decoder except attention & positional encoding.
-    transformer_dec_positional_dropout_rate : float, optional
-        Dropout rate after decoder positional encoding.
-    transformer_dec_attn_dropout_rate : float, optional
-        Dropout rate in deocoder self-attention module.
-    transformer_enc_dec_attn_dropout_rate : float, optional
-        Dropout rate in encoder-deocoder attention module.
-    init_type : str, optional
-        How to initialize transformer parameters.
-    init_enc_alpha : float, optional
-        Initial value of alpha in scaled pos encoding of the encoder.
-    init_dec_alpha : float, optional
-        Initial value of alpha in scaled pos encoding of the decoder.
-    eprenet_dropout_rate : float, optional
-        Dropout rate in encoder prenet.
-    dprenet_dropout_rate : float, optional
-        Dropout rate in decoder prenet.
-    postnet_dropout_rate : float, optional
-        Dropout rate in postnet.
-    use_masking : bool, optional
-        Whether to apply masking for padded part in loss calculation.
-    use_weighted_masking : bool, optional
-        Whether to apply weighted masking in loss calculation.
-    bce_pos_weight : float, optional
-        Positive sample weight in bce calculation (only for use_masking=true).
-    loss_type : str, optional
-        How to calculate loss.
-    use_guided_attn_loss : bool, optional
-        Whether to use guided attention loss.
-    num_heads_applied_guided_attn : int, optional
-        Number of heads in each layer to apply guided attention loss.
-    num_layers_applied_guided_attn : int, optional
-        Number of layers to apply guided attention loss.
            List of module names to apply guided attention loss.
    """
@@ -398,25 +338,16 @@ class TransformerTTS(nn.Layer):
    ) -> Tuple[paddle.Tensor, Dict[str, paddle.Tensor], paddle.Tensor]:
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            text(Tensor(int64)): Batch of padded character ids (B, Tmax).
-        text : Tensor(int64)
+            text_lengths(Tensor(int64)): Batch of lengths of each input batch (B,).
-            Batch of padded character ids (B, Tmax).
+            speech(Tensor): Batch of padded target features (B, Lmax, odim).
-        text_lengths : Tensor(int64)
+            speech_lengths(Tensor(int64)): Batch of the lengths of each target (B,).
-            Batch of lengths of each input batch (B,).
+            spk_emb(Tensor, optional): Batch of speaker embeddings (B, spk_embed_dim).
-        speech : Tensor
-            Batch of padded target features (B, Lmax, odim).
+        Returns:
-        speech_lengths : Tensor(int64)
+            Tensor: Loss scalar value.
-            Batch of the lengths of each target (B,).
+            Dict: Statistics to be monitored.
-        spk_emb : Tensor, optional
-            Batch of speaker embeddings (B, spk_embed_dim).
-        Returns
-        ----------
-        Tensor
-            Loss scalar value.
-        Dict
-            Statistics to be monitored.
        """
        # input of embedding must be int64
@@ -525,31 +456,19 @@ class TransformerTTS(nn.Layer):
    ) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
        """Generate the sequence of features given the sequences of characters.
-        Parameters
+        Args:
-        ----------
+            text(Tensor(int64)): Input sequence of characters (T,).
-        text : Tensor(int64)
+            speech(Tensor, optional): Feature sequence to extract style (N, idim).
-            Input sequence of characters (T,).
+            spk_emb(Tensor, optional): Speaker embedding vector (spk_embed_dim,).
-        speech : Tensor, optional
+            threshold(float, optional): Threshold in inference.
-            Feature sequence to extract style (N, idim).
+            minlenratio(float, optional): Minimum length ratio in inference.
-        spk_emb : Tensor, optional
+            maxlenratio(float, optional): Maximum length ratio in inference.
-            Speaker embedding vector (spk_embed_dim,).
+            use_teacher_forcing(bool, optional): Whether to use teacher forcing.
-        threshold : float, optional
-            Threshold in inference.
+        Returns:
-        minlenratio : float, optional
+            Tensor: Output sequence of features (L, odim).
-            Minimum length ratio in inference.
+            Tensor: Output sequence of stop probabilities (L,).
-        maxlenratio : float, optional
+            Tensor: Encoder-decoder (source) attention weights (#layers, #heads, L, T).
-            Maximum length ratio in inference.
-        use_teacher_forcing : bool, optional
-            Whether to use teacher forcing.
-        Returns
-        ----------
-        Tensor
-            Output sequence of features (L, odim).
-        Tensor
-            Output sequence of stop probabilities (L,).
-        Tensor
-            Encoder-decoder (source) attention weights (#layers, #heads, L, T).
        """
        # input of embedding must be int64
@@ -671,19 +590,13 @@ class TransformerTTS(nn.Layer):
    def _source_mask(self, ilens: paddle.Tensor) -> paddle.Tensor:
        """Make masks for self-attention.
-        Parameters
+        Args:
-        ----------
+            ilens(Tensor): Batch of lengths (B,).
-        ilens : Tensor
-            Batch of lengths (B,).
-        Returns
+        Returns:
-        -------
+            Tensor: Mask tensor for self-attention. dtype=paddle.bool
-        Tensor
-            Mask tensor for self-attention.
-            dtype=paddle.bool
-        Examples
+        Examples:
-        -------
            >>> ilens = [5, 3]
            >>> self._source_mask(ilens)
            tensor([[[1, 1, 1, 1, 1],
@@ -696,18 +609,13 @@ class TransformerTTS(nn.Layer):
    def _target_mask(self, olens: paddle.Tensor) -> paddle.Tensor:
        """Make masks for masked self-attention.
-        Parameters
+        Args:
-        ----------
+            olens (Tensor(int64)): Batch of lengths (B,).
-            olens : LongTensor
-                Batch of lengths (B,).
-        Returns
+        Returns:
-        ----------
+            Tensor: Mask tensor for masked self-attention.
-        Tensor
-            Mask tensor for masked self-attention.
-        Examples
+        Examples:
-        ----------
            >>> olens = [5, 3]
            >>> self._target_mask(olens)
            tensor([[[1, 0, 0, 0, 0],
@@ -731,17 +639,12 @@ class TransformerTTS(nn.Layer):
                                  spk_emb: paddle.Tensor) -> paddle.Tensor:
        """Integrate speaker embedding with hidden states.
-        Parameters
+        Args:
-        ----------
+            hs(Tensor): Batch of hidden state sequences (B, Tmax, adim).
-        hs : Tensor
+            spk_emb(Tensor): Batch of speaker embeddings (B, spk_embed_dim).
-            Batch of hidden state sequences (B, Tmax, adim).
-        spk_emb : Tensor
+        Returns:
-            Batch of speaker embeddings (B, spk_embed_dim).
+            Tensor: Batch of integrated hidden state sequences (B, Tmax, adim).
-        Returns
-        ----------
-        Tensor
-            Batch of integrated hidden state sequences (B, Tmax, adim).
        """
        if self.spk_embed_integration_type == "add":

--- a/paddlespeech/t2s/models/waveflow.py
+++ b/paddlespeech/t2s/models/waveflow.py
--- a/paddlespeech/t2s/models/wavernn/wavernn.py
+++ b/paddlespeech/t2s/models/wavernn/wavernn.py
@@ -67,14 +67,10 @@ class MelResNet(nn.Layer):
    def forward(self, x):
        '''
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input tensor (B, in_dims, T).
-        x : Tensor
+        Returns:
-            Input tensor (B, in_dims, T).
+            Tensor: Output tensor (B, res_out_dims, T).
-        Returns
-        ----------
-        Tensor
-            Output tensor (B, res_out_dims, T).
        '''
        x = self.conv_in(x)
@@ -121,16 +117,11 @@ class UpsampleNetwork(nn.Layer):
    def forward(self, m):
        '''
-        Parameters
+        Args:
-        ----------
+            c (Tensor): Input tensor (B, C_aux, T).
-        c : Tensor
+        Returns:
-            Input tensor (B, C_aux, T).
+            Tensor: Output tensor (B, (T - 2 * pad) *  prob(upsample_scales), C_aux).
-        Returns
+            Tensor: Output tensor (B, (T - 2 * pad) *  prob(upsample_scales), res_out_dims).
-        ----------
-        Tensor
-            Output tensor (B, (T - 2 * pad) *  prob(upsample_scales), C_aux).
-        Tensor
-            Output tensor (B, (T - 2 * pad) *  prob(upsample_scales), res_out_dims).
        '''
        # aux: [B, C_aux, T] 
        # -> [B, res_out_dims, T - 2 * aux_context_window]
@@ -172,32 +163,20 @@ class WaveRNN(nn.Layer):
            mode='RAW',
            init_type: str="xavier_uniform", ):
        '''
-        Parameters
+        Args:
-        ----------
+            rnn_dims (int, optional): Hidden dims of RNN Layers.
-        rnn_dims : int, optional
+            fc_dims (int, optional): Dims of FC Layers.
-            Hidden dims of RNN Layers.
+            bits (int, optional): bit depth of signal.
-        fc_dims : int, optional
+            aux_context_window (int, optional): The context window size of the first convolution applied to the 
-             Dims of FC Layers.
-        bits : int, optional
-            bit depth of signal.
-        aux_context_window : int, optional
-            The context window size of the first convolution applied to the 
                auxiliary input, by default 2
-        upsample_scales : List[int], optional
+            upsample_scales (List[int], optional): Upsample scales of the upsample network.
-            Upsample scales of the upsample network.
+            aux_channels (int, optional): Auxiliary channel of the residual blocks.
-        aux_channels : int, optional
+            compute_dims (int, optional): Dims of Conv1D in MelResNet.
-            Auxiliary channel of the residual blocks.
+            res_out_dims (int, optional): Dims of output in MelResNet.
-        compute_dims : int, optional
+            res_blocks (int, optional): Number of residual blocks.
-            Dims of Conv1D in MelResNet.
+            mode (str, optional): Output mode of the WaveRNN vocoder. 
-        res_out_dims : int, optional
+                `MOL` for Mixture of Logistic Distribution, and `RAW` for quantized bits as the model's output.
-            Dims of output in MelResNet.
+            init_type (str): How to initialize parameters.
-        res_blocks : int, optional
-            Number of residual blocks.
-        mode : str, optional
-            Output mode of the WaveRNN vocoder. `MOL` for Mixture of Logistic Distribution,
-            and `RAW` for quantized bits as the model's output.
-        init_type : str
-            How to initialize parameters.
        '''
        super().__init__()
        self.mode = mode
@@ -245,18 +224,13 @@ class WaveRNN(nn.Layer):
    def forward(self, x, c):
        '''
-        Parameters
+        Args:
-        ----------
+            x (Tensor): wav sequence, [B, T]
-        x : Tensor
+            c (Tensor): mel spectrogram [B, C_aux, T']
-            wav sequence, [B, T]
-        c : Tensor
-            mel spectrogram [B, C_aux, T']
            T = (T' - 2 * aux_context_window ) * hop_length
-        Returns
+        Returns:
-        ----------
+            Tensor: [B, T, n_classes]
-        Tensor
-            [B, T, n_classes]
        '''
        # Although we `_flatten_parameters()` on init, when using DataParallel
        # the model gets replicated, making it no longer guaranteed that the
@@ -304,22 +278,14 @@ class WaveRNN(nn.Layer):
                 mu_law: bool=True,
                 gen_display: bool=False):
        """
-        Parameters
+        Args:
-        ----------
+            c(Tensor): input mels, (T', C_aux)
-        c : Tensor
+            batched(bool): generate in batch or not
-            input mels, (T', C_aux)
+            target(int): target number of samples to be generated in each batch entry
-        batched : bool
+            overlap(int): number of samples for crossfading between batches
-            generate in batch or not
+            mu_law(bool)
-        target : int
+        Returns: 
-            target number of samples to be generated in each batch entry
+            wav sequence: Output (T' * prod(upsample_scales), out_channels, C_out).
-        overlap : int
-            number of samples for crossfading between batches
-        mu_law : bool
-            use mu law or not
-        Returns
-        ----------
-        wav sequence
-            Output (T' * prod(upsample_scales), out_channels, C_out).
        """
        self.eval()
@@ -434,15 +400,12 @@ class WaveRNN(nn.Layer):
    def pad_tensor(self, x, pad, side='both'):
        '''
-        Parameters
+        Args:
-        ----------
+            x(Tensor): mel, [1, n_frames, 80]
-        x : Tensor
+            pad(int): 
-            mel, [1, n_frames, 80]
+            side(str, optional):  (Default value = 'both')
-        pad : int
-        side : str 
+        Returns:
-            'both', 'before' or 'after'
-        Returns
-        ----------
            Tensor
        '''
        b, t, _ = paddle.shape(x)
@@ -461,33 +424,24 @@ class WaveRNN(nn.Layer):
        Fold the tensor with overlap for quick batched inference.
        Overlap will be used for crossfading in xfade_and_unfold()
-        Parameters
+        Args:
-        ----------
+            x(Tensor): Upsampled conditioning features. mels or aux
-        x : Tensor
-            Upsampled conditioning features. mels or aux
                shape=(1, T, features)
                mels: [1, T, 80]
                aux: [1, T, 128]
-        target : int
+            target(int): Target timesteps for each index of batch
-            Target timesteps for each index of batch
+            overlap(int): Timesteps for both xfade and rnn warmup
-        overlap : int
-            Timesteps for both xfade and rnn warmup
+        Returns:
-            overlap = hop_length * 2
+            Tensor: 
-        Returns
-        ----------
-        Tensor 
                shape=(num_folds, target + 2 * overlap, features)
                num_flods = (time_seq - overlap) // (target + overlap)
                mel: [num_folds, target + 2 * overlap, 80]
                aux: [num_folds, target + 2 * overlap, 128]
-        Details
+        Details:
-        ----------
            x = [[h1, h2, ... hn]]
            Where each h is a vector of conditioning features
            Eg: target=2, overlap=1 with x.size(1)=10
            folded = [[h1, h2, h3, h4],
@@ -520,24 +474,20 @@ class WaveRNN(nn.Layer):
    def xfade_and_unfold(self, y, target: int=12000, overlap: int=600):
        ''' Applies a crossfade and unfolds into a 1d array.
-        Parameters
+        Args:
-        ----------
+            y (Tensor): 
-        y : Tensor
                Batched sequences of audio samples
                shape=(num_folds, target + 2 * overlap)
                dtype=paddle.float32
-        overlap : int
+            overlap (int): Timesteps for both xfade and rnn warmup
-            Timesteps for both xfade and rnn warmup
-        Returns
+        Returns:
-        ----------
            Tensor
                audio samples in a 1d array
                shape=(total_len)
                dtype=paddle.float32
-        Details
+        Details:
-        ----------
            y = [[seq1],
                [seq2],
                [seq3]]

--- a/paddlespeech/t2s/modules/causal_conv.py
+++ b/paddlespeech/t2s/modules/causal_conv.py
@@ -41,14 +41,10 @@ class CausalConv1D(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input tensor (B, in_channels, T).
-        x : Tensor
+        Returns: 
-            Input tensor (B, in_channels, T).
+            Tensor: Output tensor (B, out_channels, T).
-        Returns
-        ----------
-        Tensor
-            Output tensor (B, out_channels, T).
        """
        return self.conv(self.pad(x))[:, :, :x.shape[2]]
@@ -70,13 +66,9 @@ class CausalConv1DTranspose(nn.Layer):
    def forward(self, x):
        """Calculate forward propagation.
-        Parameters
+        Args:
-        ----------
+            x (Tensor): Input tensor (B, in_channels, T_in).
-        x : Tensor
+        Returns:
-            Input tensor (B, in_channels, T_in).
+            Tensor: Output tensor (B, out_channels, T_out).
-        Returns
-        ----------
-        Tensor
-            Output tensor (B, out_channels, T_out).
        """
        return self.deconv(x)[:, :, :-self.stride]
--- a/paddlespeech/t2s/modules/conformer/convolution.py
+++ b/paddlespeech/t2s/modules/conformer/convolution.py
@@ -18,12 +18,10 @@ from paddle import nn
 class ConvolutionModule(nn.Layer):
    """ConvolutionModule in Conformer model.
-    Parameters
-    ----------
+    Args:
-    channels : int
+        channels (int): The number of channels of conv layers.
-        The number of channels of conv layers.
+        kernel_size (int): Kernerl size of conv layers.
-    kernel_size : int
-        Kernerl size of conv layers.
    """
    def __init__(self, channels, kernel_size, activation=nn.ReLU(), bias=True):
@@ -59,14 +57,11 @@ class ConvolutionModule(nn.Layer):
    def forward(self, x):
        """Compute convolution module.
-        Parameters
-        ----------
+        Args:
-        x : paddle.Tensor
+            x (Tensor): Input tensor (#batch, time, channels).
-            Input tensor (#batch, time, channels).
+        Returns:
-        Returns
+            Tensor: Output tensor (#batch, time, channels).
-        ----------
-        paddle.Tensor
-            Output tensor (#batch, time, channels).
        """
        # exchange the temporal dimension and the feature dimension
        x = x.transpose([0, 2, 1])

--- a/paddlespeech/t2s/modules/conformer/encoder_layer.py
+++ b/paddlespeech/t2s/modules/conformer/encoder_layer.py
@@ -21,36 +21,27 @@ from paddlespeech.t2s.modules.layer_norm import LayerNorm
 class EncoderLayer(nn.Layer):
    """Encoder layer module.
-    Parameters
-    ----------
+    Args:
-    size : int
+        size (int): Input dimension.
-        Input dimension.
+        self_attn (nn.Layer): Self-attention module instance.
-    self_attn : nn.Layer
-        Self-attention module instance.
            `MultiHeadedAttention` or `RelPositionMultiHeadedAttention` instance
            can be used as the argument.
-    feed_forward : nn.Layer
+        feed_forward (nn.Layer): Feed-forward module instance.
-        Feed-forward module instance.
            `PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
            can be used as the argument.
-    feed_forward_macaron : nn.Layer
+        feed_forward_macaron (nn.Layer): Additional feed-forward module instance.
-        Additional feed-forward module instance.
            `PositionwiseFeedForward`, `MultiLayeredConv1d`, or `Conv1dLinear` instance
            can be used as the argument.
-    conv_module : nn.Layer
+        conv_module (nn.Layer): Convolution module instance.
-        Convolution module instance.
            `ConvlutionModule` instance can be used as the argument.
-    dropout_rate : float
+        dropout_rate (float): Dropout rate.
-        Dropout rate.
+        normalize_before (bool): Whether to use layer_norm before the first block.
-    normalize_before : bool
+        concat_after (bool): Whether to concat attention layer's input and output.
-        Whether to use layer_norm before the first block.
-    concat_after : bool
-        Whether to concat attention layer's input and output.
            if True, additional linear will be applied.
            i.e. x -> x + linear(concat(x, att(x)))
            if False, no additional linear will be applied. i.e. x -> x + att(x)
-    stochastic_depth_rate : float
+        stochastic_depth_rate (float): Proability to skip this layer.
-        Proability to skip this layer.
            During training, the layer may skip residual computation and return input
            as-is with given probability.
    """
@@ -93,22 +84,17 @@ class EncoderLayer(nn.Layer):
    def forward(self, x_input, mask, cache=None):
        """Compute encoded features.
-        Parameters
-        ----------
+        Args:
-        x_input : Union[Tuple, paddle.Tensor]
+            x_input(Union[Tuple, Tensor]): Input tensor w/ or w/o pos emb.
-            Input tensor w/ or w/o pos emb.
                - w/ pos emb: Tuple of tensors [(#batch, time, size), (1, time, size)].
                - w/o pos emb: Tensor (#batch, time, size).
-        mask : paddle.Tensor
+            mask(Tensor): Mask tensor for the input (#batch, time).
-            Mask tensor for the input (#batch, time).
+            cache (Tensor): 
-        cache paddle.Tensor
-            Cache tensor of the input (#batch, time - 1, size).
+        Returns:
-        Returns
+            Tensor: Output tensor (#batch, time, size).
-        ----------
+            Tensor: Mask tensor (#batch, time).
-        paddle.Tensor
-            Output tensor (#batch, time, size).
-        paddle.Tensor
-            Mask tensor (#batch, time).
        """
        if isinstance(x_input, tuple):
            x, pos_emb = x_input[0], x_input[1]

--- 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/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/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