Merge pull request #694 from PaddlePaddle/dataset

add more chinese speech dataset

examples/dataset/aidatatang_200zh/.gitignore

+*.tgz
+manifest.*
+*.meta
+aidatatang_200zh/
\ No newline at end of file

examples/dataset/aidatatang_200zh/README.md

+# [Aidatatang_200zh](http://www.openslr.org/62/)
+
+Aidatatang_200zh is a free Chinese Mandarin speech corpus provided by Beijing DataTang Technology Co., Ltd under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Public License.
+The contents and the corresponding descriptions of the corpus include:
+
+* The corpus contains 200 hours of acoustic data, which is mostly mobile recorded data.
+* 600 speakers from different accent areas in China are invited to participate in the recording.
+* The transcription accuracy for each sentence is larger than 98%.
+* Recordings are conducted in a quiet indoor environment.
+* The database is divided into training set, validation set, and testing set in a ratio of 7: 1: 2.
+* Detail information such as speech data coding and speaker information is preserved in the metadata file.
+* Segmented transcripts are also provided.
+
+The corpus aims to support researchers in speech recognition, machine translation, voiceprint recognition, and other speech-related fields. Therefore, the corpus is totally free for academic use.

examples/dataset/aidatatang_200zh/aidatatang_200zh.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Prepare aidatatang_200zh mandarin dataset
+
+Download, unpack and create manifest files.
+Manifest file is a json-format file with each line containing the
+meta data (i.e. audio filepath, transcript and audio duration)
+of each audio file in the data set.
+"""
+import argparse
+import codecs
+import json
+import os
+
+import soundfile
+
+from utils.utility import download
+from utils.utility import unpack
+
+DATA_HOME = os.path.expanduser('~/.cache/paddle/dataset/speech')
+
+URL_ROOT = 'http://www.openslr.org/resources/62'
+# URL_ROOT = 'https://openslr.magicdatatech.com/resources/62'
+DATA_URL = URL_ROOT + '/aidatatang_200zh.tgz'
+MD5_DATA = '6e0f4f39cd5f667a7ee53c397c8d0949'
+
+parser = argparse.ArgumentParser(description=__doc__)
+parser.add_argument(
+    "--target_dir",
+    default=DATA_HOME + "/aidatatang_200zh",
+    type=str,
+    help="Directory to save the dataset. (default: %(default)s)")
+parser.add_argument(
+    "--manifest_prefix",
+    default="manifest",
+    type=str,
+    help="Filepath prefix for output manifests. (default: %(default)s)")
+args = parser.parse_args()
+
+
+def create_manifest(data_dir, manifest_path_prefix):
+    print("Creating manifest %s ..." % manifest_path_prefix)
+    json_lines = []
+    transcript_path = os.path.join(data_dir, 'transcript',
+                                   'aidatatang_200_zh_transcript.txt')
+    transcript_dict = {}
+    for line in codecs.open(transcript_path, 'r', 'utf-8'):
+        line = line.strip()
+        if line == '':
+            continue
+        audio_id, text = line.split(' ', 1)
+        # remove withespace, charactor text
+        text = ''.join(text.split())
+        transcript_dict[audio_id] = text
+
+    data_types = ['train', 'dev', 'test']
+    for dtype in data_types:
+        del json_lines[:]
+        total_sec = 0.0
+        total_text = 0.0
+        total_num = 0
+
+        audio_dir = os.path.join(data_dir, 'corpus/', dtype)
+        for subfolder, _, filelist in sorted(os.walk(audio_dir)):
+            for fname in filelist:
+                if not fname.endswith('.wav'):
+                    continue
+
+                audio_path = os.path.abspath(os.path.join(subfolder, fname))
+                audio_id = os.path.basename(fname)[:-4]
+
+                audio_data, samplerate = soundfile.read(audio_path)
+                duration = float(len(audio_data) / samplerate)
+                text = transcript_dict[audio_id]
+                json_lines.append(
+                    json.dumps(
+                        {
+                            'utt': audio_id,
+                            'feat': audio_path,
+                            'feat_shape': (duration, ),  # second
+                            'text': text,
+                        },
+                        ensure_ascii=False))
+
+                total_sec += duration
+                total_text += len(text)
+                total_num += 1
+
+        manifest_path = manifest_path_prefix + '.' + dtype
+        with codecs.open(manifest_path, 'w', 'utf-8') as fout:
+            for line in json_lines:
+                fout.write(line + '\n')
+
+        with open(dtype + '.meta', 'w') as f:
+            print(f"{dtype}:", file=f)
+            print(f"{total_num} utts", file=f)
+            print(f"{total_sec / (60*60)} h", file=f)
+            print(f"{total_text} text", file=f)
+            print(f"{total_text / total_sec} text/sec", file=f)
+            print(f"{total_sec / total_num} sec/utt", file=f)
+
+
+def prepare_dataset(url, md5sum, target_dir, manifest_path, subset):
+    """Download, unpack and create manifest file."""
+    data_dir = os.path.join(target_dir, subset)
+    if not os.path.exists(data_dir):
+        filepath = download(url, md5sum, target_dir)
+        unpack(filepath, target_dir)
+        # unpack all audio tar files
+        audio_dir = os.path.join(data_dir, 'corpus')
+        for subfolder, dirlist, filelist in sorted(os.walk(audio_dir)):
+            for sub in dirlist:
+                print(f"unpack dir {sub}...")
+                for folder, _, filelist in sorted(
+                        os.walk(os.path.join(subfolder, sub))):
+                    for ftar in filelist:
+                        unpack(os.path.join(folder, ftar), folder, True)
+    else:
+        print("Skip downloading and unpacking. Data already exists in %s." %
+              target_dir)
+
+    create_manifest(data_dir, manifest_path)
+
+
+def main():
+    if args.target_dir.startswith('~'):
+        args.target_dir = os.path.expanduser(args.target_dir)
+
+    prepare_dataset(
+        url=DATA_URL,
+        md5sum=MD5_DATA,
+        target_dir=args.target_dir,
+        manifest_path=args.manifest_prefix,
+        subset='aidatatang_200zh')
+
+    print("Data download and manifest prepare done!")
+
+
+if __name__ == '__main__':
+    main()

examples/dataset/aishell/.gitignore

 data_aishell*
+*.meta
+manifest.*
+*.tgz
\ No newline at end of file

examples/dataset/aishell/README.md

+# [Aishell1](http://www.openslr.org/33/)
+
+This Open Source Mandarin Speech Corpus, AISHELL-ASR0009-OS1, is 178 hours long. It is a part of AISHELL-ASR0009, of which utterance contains 11 domains, including smart home, autonomous driving, and industrial production. The whole recording was put in quiet indoor environment, using 3 different devices at the same time: high fidelity microphone (44.1kHz, 16-bit,); Android-system mobile phone (16kHz, 16-bit), iOS-system mobile phone (16kHz, 16-bit). Audios in high fidelity were re-sampled to 16kHz to build AISHELL- ASR0009-OS1. 400 speakers from different accent areas in China were invited to participate in the recording. The manual transcription accuracy rate is above 95%, through professional speech annotation and strict quality inspection. The corpus is divided into training, development and testing sets. ( This database is free for academic research, not in the commerce, if without permission. )

examples/dataset/aishell/aishell.py

@@ -31,7 +31,7 @@ from utils.utility import unpack
 DATA_HOME = os.path.expanduser('~/.cache/paddle/dataset/speech')
 
 URL_ROOT = 'http://www.openslr.org/resources/33'
-URL_ROOT = 'https://openslr.magicdatatech.com/resources/33'
+# URL_ROOT = 'https://openslr.magicdatatech.com/resources/33'
 DATA_URL = URL_ROOT + '/data_aishell.tgz'
 MD5_DATA = '2f494334227864a8a8fec932999db9d8'
 
@@ -67,11 +67,15 @@ def create_manifest(data_dir, manifest_path_prefix):
     data_types = ['train', 'dev', 'test']
     for dtype in data_types:
         del json_lines[:]
+        total_sec = 0.0
+        total_text = 0.0
+        total_num = 0
+
         audio_dir = os.path.join(data_dir, 'wav', dtype)
         for subfolder, _, filelist in sorted(os.walk(audio_dir)):
             for fname in filelist:
-                audio_path = os.path.join(subfolder, fname)
-                audio_id = fname[:-4]
+                audio_path = os.path.abspath(os.path.join(subfolder, fname))
+                audio_id = os.path.basename(fname)[:-4]
                 # if no transcription for audio then skipped
                 if audio_id not in transcript_dict:
                     continue
@@ -81,20 +85,30 @@ def create_manifest(data_dir, manifest_path_prefix):
                 json_lines.append(
                     json.dumps(
                         {
-                            'utt':
-                            os.path.splitext(os.path.basename(audio_path))[0],
-                            'feat':
-                            audio_path,
+                            'utt': audio_id,
+                            'feat': audio_path,
                             'feat_shape': (duration, ),  # second
-                            'text':
-                            text
+                            'text': text
                         },
                         ensure_ascii=False))
+
+                total_sec += duration
+                total_text += len(text)
+                total_num += 1
+
         manifest_path = manifest_path_prefix + '.' + dtype
         with codecs.open(manifest_path, 'w', 'utf-8') as fout:
             for line in json_lines:
                 fout.write(line + '\n')
 
+        with open(dtype + '.meta', 'w') as f:
+            print(f"{dtype}:", file=f)
+            print(f"{total_num} utts", file=f)
+            print(f"{total_sec / (60*60)} h", file=f)
+            print(f"{total_text} text", file=f)
+            print(f"{total_text / total_sec} text/sec", file=f)
+            print(f"{total_sec / total_num} sec/utt", file=f)
+
 
 def prepare_dataset(url, md5sum, target_dir, manifest_path):
     """Download, unpack and create manifest file."""

examples/dataset/aishell3/README.md

+# [Aishell3](http://www.openslr.org/93/)
+
+AISHELL-3 is a large-scale and high-fidelity multi-speaker Mandarin speech corpus which could be used to train multi-speaker Text-to-Speech (TTS) systems. The corpus contains roughly **85 hours** of emotion-neutral recordings spoken by 218 native Chinese mandarin speakers and total 88035 utterances. Their auxiliary attributes such as gender, age group and native accents are explicitly marked and provided in the corpus. Accordingly, transcripts in Chinese character-level and pinyin-level are provided along with the recordings. The  word & tone transcription accuracy rate is above 98%, through professional speech annotation and strict quality inspection for tone and prosody. ( This database is free for academic research, not in the commerce, if without permission. )

examples/dataset/gigaspeech/.gitignore

+GigaSpeech/

examples/dataset/gigaspeech/README.md

+# [GigaSpeech](https://github.com/SpeechColab/GigaSpeech)
+
+```
+git clone https://github.com/SpeechColab/GigaSpeech.git
+
+cd GigaSpeech
+utils/gigaspeech_download.sh /disk1/audio_data/gigaspeech
+toolkits/kaldi/gigaspeech_data_prep.sh --train-subset XL /disk1/audio_data/gigaspeech ../data
+cd ..
+```

examples/dataset/gigaspeech/gigaspeech.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.

examples/dataset/gigaspeech/run.sh

+#!/bin/bash
+
+set -e
+
+curdir=$PWD
+
+test -d GigaSpeech || git clone https://github.com/SpeechColab/GigaSpeech.git
+cd GigaSpeech
+source env_vars.sh
+utils/gigaspeech_download.sh ${curdir}/

examples/dataset/librispeech/.gitignore

@@ -5,3 +5,5 @@ test-other
 train-clean-100
 train-clean-360
 train-other-500
+*.meta
+manifest.*

examples/dataset/librispeech/librispeech.py

@@ -77,6 +77,10 @@ def create_manifest(data_dir, manifest_path):
     """
     print("Creating manifest %s ..." % manifest_path)
     json_lines = []
+    total_sec = 0.0
+    total_text = 0.0
+    total_num = 0
+
     for subfolder, _, filelist in sorted(os.walk(data_dir)):
         text_filelist = [
             filename for filename in filelist if filename.endswith('trans.txt')
@@ -86,7 +90,9 @@ def create_manifest(data_dir, manifest_path):
             for line in io.open(text_filepath, encoding="utf8"):
                 segments = line.strip().split()
                 text = ' '.join(segments[1:]).lower()
-                audio_filepath = os.path.join(subfolder, segments[0] + '.flac')
+
+                audio_filepath = os.path.abspath(
+                    os.path.join(subfolder, segments[0] + '.flac'))
                 audio_data, samplerate = soundfile.read(audio_filepath)
                 duration = float(len(audio_data)) / samplerate
                 json_lines.append(
@@ -99,10 +105,24 @@ def create_manifest(data_dir, manifest_path):
                         'text':
                         text
                     }))
+
+                total_sec += duration
+                total_text += len(text)
+                total_num += 1
+
     with codecs.open(manifest_path, 'w', 'utf-8') as out_file:
         for line in json_lines:
             out_file.write(line + '\n')
 
+    subset = os.path.splitext(manifest_path)[1]
+    with open(subset + '.meta', 'w') as f:
+        print(f"{subset}:", file=f)
+        print(f"{total_num} utts", file=f)
+        print(f"{total_sec / (60*60)} h", file=f)
+        print(f"{total_text} text", file=f)
+        print(f"{total_text / total_sec} text/sec", file=f)
+        print(f"{total_sec / total_num} sec/utt", file=f)
+
 
 def prepare_dataset(url, md5sum, target_dir, manifest_path):
     """Download, unpack and create summmary manifest file.

examples/dataset/magicdata/README.md

+# [MagicData](http://www.openslr.org/68/)
+
+MAGICDATA Mandarin Chinese Read Speech Corpus was developed by MAGIC DATA Technology Co., Ltd. and freely published for non-commercial use.
+The contents and the corresponding descriptions of the corpus include:
+
+* The corpus contains 755 hours of speech data, which is mostly mobile recorded data.
+* 1080 speakers from different accent areas in China are invited to participate in the recording.
+* The sentence transcription accuracy is higher than 98%.
+* Recordings are conducted in a quiet indoor environment.
+* The database is divided into training set, validation set, and testing set in a ratio of 51: 1: 2.
+* Detail information such as speech data coding and speaker information is preserved in the metadata file.
+* The domain of recording texts is diversified, including interactive Q&A, music search, SNS messages, home command and control, etc.
+* Segmented transcripts are also provided.
+
+The corpus aims to support researchers in speech recognition, machine translation, speaker recognition, and other speech-related fields. Therefore, the corpus is totally free for academic use.

examples/dataset/mini_librispeech/.gitignore

@@ -2,3 +2,4 @@ dev-clean/
 manifest.dev-clean
 manifest.train-clean
 train-clean/
+*.meta

examples/dataset/mini_librispeech/mini_librispeech.py

@@ -58,6 +58,10 @@ def create_manifest(data_dir, manifest_path):
     """
     print("Creating manifest %s ..." % manifest_path)
     json_lines = []
+    total_sec = 0.0
+    total_text = 0.0
+    total_num = 0
+
     for subfolder, _, filelist in sorted(os.walk(data_dir)):
         text_filelist = [
             filename for filename in filelist if filename.endswith('trans.txt')
@@ -80,10 +84,24 @@ def create_manifest(data_dir, manifest_path):
                         'text':
                         text
                     }))
+
+                total_sec += duration
+                total_text += len(text)
+                total_num += 1
+
     with codecs.open(manifest_path, 'w', 'utf-8') as out_file:
         for line in json_lines:
             out_file.write(line + '\n')
 
+    subset = os.path.splitext(manifest_path)[1]
+    with open(subset + '.meta', 'w') as f:
+        print(f"{subset}:", file=f)
+        print(f"{total_num} utts", file=f)
+        print(f"{total_sec / (60*60)} h", file=f)
+        print(f"{total_text} text", file=f)
+        print(f"{total_text / total_sec} text/sec", file=f)
+        print(f"{total_sec / total_num} sec/utt", file=f)
+
 
 def prepare_dataset(url, md5sum, target_dir, manifest_path):
     """Download, unpack and create summmary manifest file.

examples/dataset/multi_cn/README.md

+# multi-cn
+
+This is a Chinese speech recognition recipe that trains on all Chinese corpora on OpenSLR, including:
+
+* Aidatatang (140 hours)
+* Aishell (151 hours)
+* MagicData (712 hours)
+* Primewords (99 hours)
+* ST-CMDS (110 hours)
+* THCHS-30 (26 hours)
+* optional AISHELL2 (~1000 hours) if available

examples/dataset/primewords/README.md

+# [Primewords](http://www.openslr.org/47/)
+
+This free Chinese Mandarin speech corpus set is released by Shanghai Primewords Information Technology Co., Ltd.
+The corpus is recorded by smart mobile phones from 296 native Chinese speakers. The transcription accuracy is larger than 98%, at the confidence level of 95%. It is free for academic use.
+
+The mapping between the transcript and utterance is given in JSON format.

examples/dataset/st-cmds/README.md

+# [FreeST](http://www.openslr.org/38/)

examples/dataset/thchs30/.gitignore

@@ -3,3 +3,4 @@ manifest.*
 data_thchs30
 resource
 test-noise
+*.meta

examples/dataset/thchs30/README.md

+# [THCHS30](http://www.openslr.org/18/)
+
+This is the *data part* of the `THCHS30 2015` acoustic data
+& scripts dataset.
+
+The dataset is described in more detail in the paper ``THCHS-30 : A Free
+Chinese Speech Corpus`` by Dong Wang, Xuewei Zhang.
+
+A paper (if it can be called a paper) 13 years ago regarding the database:
+
+Dong Wang, Dalei Wu, Xiaoyan Zhu, ``TCMSD: A new Chinese Continuous Speech Database``,
+International Conference on Chinese Computing (ICCC'01), 2001, Singapore.
+
+The layout of this data pack is the following:
+
+  ``data``
+      ``*.wav``
+        audio data
+
+      ``*.wav.trn``  
+        transcriptions
+
+  ``{train,dev,test}``
+    contain symlinks into the ``data`` directory for both audio and
+    transcription files. Contents of these directories define the
+    train/dev/test split of the data.
+
+  ``{lm_word}``
+       ``word.3gram.lm``
+         trigram LM based on word
+        ``lexicon.txt``
+         lexicon based on word
+
+   ``{lm_phone}``
+       ``phone.3gram.lm``
+         trigram LM based on phone
+        ``lexicon.txt``
+         lexicon based on phone
+
+  ``README.TXT``
+    this file
+
+
+Data statistics
+===============
+
+Statistics for the data are as follows:
+
+    ===========  ==========  ==========  ===========
+    **dataset**  **audio**   **#sents**  **#words**
+    ===========  ==========  ==========  ===========
+        train        25        10,000      198,252
+        dev         2:14         893        17,743
+        test        6:15        2,495       49,085
+    ===========  ==========  ==========  ===========

examples/dataset/thchs30/thchs30.py

@@ -69,9 +69,7 @@ def read_trn(filepath):
     """
     texts = []
     with open(filepath, 'r') as f:
-        lines = f.read().split('\n')
-        # last line is `empty`
-        lines = lines[:3]
+        lines = f.read().strip().split('\n')
         assert len(lines) == 3, lines
     # charactor text, remove withespace
     texts.append(''.join(lines[0].split()))
@@ -98,6 +96,10 @@ def create_manifest(data_dir, manifest_path_prefix):
     data_types = ['train', 'dev', 'test']
     for dtype in data_types:
         del json_lines[:]
+        total_sec = 0.0
+        total_text = 0.0
+        total_num = 0
+
         audio_dir = os.path.join(data_dir, dtype)
         for subfolder, _, filelist in sorted(os.walk(audio_dir)):
             for fname in filelist:
@@ -127,11 +129,23 @@ def create_manifest(data_dir, manifest_path_prefix):
                         },
                         ensure_ascii=False))
 
+                total_sec += duration
+                total_text += len(text)
+                total_num += 1
+
         manifest_path = manifest_path_prefix + '.' + dtype
         with codecs.open(manifest_path, 'w', 'utf-8') as fout:
             for line in json_lines:
                 fout.write(line + '\n')
 
+        with open(dtype + '.meta', 'w') as f:
+            print(f"{dtype}:", file=f)
+            print(f"{total_num} utts", file=f)
+            print(f"{total_sec / (60*60)} h", file=f)
+            print(f"{total_text} text", file=f)
+            print(f"{total_text / total_sec} text/sec", file=f)
+            print(f"{total_sec / total_num} sec/utt", file=f)
+
 
 def prepare_dataset(url, md5sum, target_dir, manifest_path, subset):
     """Download, unpack and create manifest file."""