TianYuan
4 years ago
parent
c0ee57d400
commit
00017301c6
@ -1,2 +1,7 @@
|
||||
this is the example of MFA for thchs30 dataset
|
||||
cd a0 run run.sh to get start
|
||||
|
||||
MFA 对齐所使用的字典
|
||||
MFA 字典的格式可以参考: https://montreal-forced-aligner.readthedocs.io/en/latest/dictionary.html
|
||||
phone.lexicon 直接使用的是 THCHS-30/data_thchs30/lm_phone/lexicon.txt
|
||||
word.lexicon 是一个带概率的字典, 生成规则请参考 local/gen_word2phone.py
|
||||
|
||||
@ -1,112 +0,0 @@
|
||||
# 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.
|
||||
"""Recorganize THCHS-30 for MFA
|
||||
read manifest.train from root-dir
|
||||
Link *.wav to output-dir
|
||||
dump *.lab from manifest.train, such as: text、syllable and phone
|
||||
Manifest file is a json-format file with each line containing the
|
||||
meta data (i.e. audio filepath, transcript and audio duration)
|
||||
"""
|
||||
import argparse
|
||||
import os
|
||||
from pathlib import Path
|
||||
from typing import Union
|
||||
|
||||
from deepspeech.frontend.utility import read_manifest
|
||||
|
||||
|
||||
def link_wav(root_dir: Union[str, Path], output_dir: Union[str, Path]):
|
||||
manifest_path = root_dir / "manifest.train"
|
||||
manifest_jsons = read_manifest(manifest_path)
|
||||
for line_json in manifest_jsons:
|
||||
wav_path = line_json['feat']
|
||||
wav_name = wav_path.split("/")[-1]
|
||||
new_wav_path = output_dir / wav_name
|
||||
os.symlink(wav_path, new_wav_path)
|
||||
|
||||
|
||||
def link_lexicon(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path],
|
||||
script_type='phone'):
|
||||
manifest_path = root_dir / "manifest.train"
|
||||
manifest_jsons = read_manifest(manifest_path)
|
||||
line_json = manifest_jsons[0]
|
||||
wav_path = line_json['feat']
|
||||
|
||||
if script_type == 'phone':
|
||||
# find lexicon.txt in THCHS-30
|
||||
grader_father = os.path.abspath(
|
||||
os.path.dirname(wav_path) + os.path.sep + "..")
|
||||
grader_father = Path(grader_father).expanduser()
|
||||
lexicon_name = "lexicon.txt"
|
||||
lexicon_father_dir = "lm_phone"
|
||||
lexicon_path = grader_father / lexicon_father_dir / lexicon_name
|
||||
elif script_type == 'syllable':
|
||||
# find thchs30_pinyin2phone in dir of this py file
|
||||
py_dir_path = os.path.split(os.path.realpath(__file__))[0]
|
||||
py_dir_path = Path(py_dir_path).expanduser()
|
||||
lexicon_path = py_dir_path / "thchs30_pinyin2phone"
|
||||
else:
|
||||
# script_type == 'text'
|
||||
# find thchs30_cn2phone in dir of this py file
|
||||
py_dir_path = os.path.split(os.path.realpath(__file__))[0]
|
||||
py_dir_path = Path(py_dir_path).expanduser()
|
||||
lexicon_path = py_dir_path / "thchs30_cn2phone"
|
||||
|
||||
new_lexicon_name = script_type + ".lexicon"
|
||||
new_lexicon_path = os.path.dirname(output_dir) + "/" + new_lexicon_name
|
||||
os.symlink(lexicon_path, new_lexicon_path)
|
||||
|
||||
|
||||
def dump_lab(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path],
|
||||
script_type='phone'):
|
||||
# script_type can in {'text', 'syllable', 'phone'}
|
||||
manifest_path = root_dir / "manifest.train"
|
||||
manifest_jsons = read_manifest(manifest_path)
|
||||
for line_json in manifest_jsons:
|
||||
utt_id = line_json['utt']
|
||||
transcript_name = utt_id + ".lab"
|
||||
transcript_path = output_dir / transcript_name
|
||||
with open(transcript_path, 'wt') as wf:
|
||||
wf.write(line_json[script_type] + "\n")
|
||||
|
||||
|
||||
def reorganize_thchs30(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path]=None,
|
||||
script_type='phone'):
|
||||
root_dir = Path(root_dir).expanduser()
|
||||
output_dir = Path(output_dir).expanduser()
|
||||
output_dir.mkdir(parents=True, exist_ok=True)
|
||||
link_wav(root_dir, output_dir)
|
||||
dump_lab(root_dir, output_dir, script_type)
|
||||
link_lexicon(root_dir, output_dir, script_type)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Reorganize THCHS-30 dataset for MFA")
|
||||
parser.add_argument("--root-dir", type=str, help="path to thchs30 dataset.")
|
||||
parser.add_argument(
|
||||
"--output-dir",
|
||||
type=str,
|
||||
help="path to save outputs(audio and transcriptions)")
|
||||
|
||||
parser.add_argument(
|
||||
"--script-type",
|
||||
type=str,
|
||||
default="phone",
|
||||
help="type of lab (text'/'syllable'/'phone')")
|
||||
args = parser.parse_args()
|
||||
reorganize_thchs30(args.root_dir, args.output_dir, args.script_type)
|
||||
@ -0,0 +1,83 @@
|
||||
# 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.
|
||||
"""Recorganize THCHS-30 for MFA
|
||||
read manifest.train from root-dir
|
||||
Link *.wav to output-dir
|
||||
dump *.lab from manifest.train, such as: text、syllable and phone
|
||||
Manifest file is a json-format file with each line containing the
|
||||
meta data (i.e. audio filepath, transcript and audio duration)
|
||||
"""
|
||||
import argparse
|
||||
import os
|
||||
from pathlib import Path
|
||||
from typing import Union
|
||||
|
||||
|
||||
def link_wav(root_dir: Union[str, Path], output_dir: Union[str, Path]):
|
||||
wav_scp_path = root_dir / 'wav.scp'
|
||||
with open(wav_scp_path, 'r') as rf:
|
||||
for line in rf:
|
||||
utt, feat = line.strip().split()
|
||||
wav_path = feat
|
||||
wav_name = wav_path.split("/")[-1]
|
||||
new_wav_path = output_dir / wav_name
|
||||
os.symlink(wav_path, new_wav_path)
|
||||
|
||||
|
||||
def write_lab(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path],
|
||||
script_type='phone'):
|
||||
# script_type can in {'word', 'syllable', 'phone'}
|
||||
json_name = 'text.' + script_type
|
||||
json_path = root_dir / json_name
|
||||
with open(json_path, 'r') as rf:
|
||||
for line in rf:
|
||||
line = line.strip().split()
|
||||
utt_id = line[0]
|
||||
context = ' '.join(line[1:])
|
||||
transcript_name = utt_id + '.lab'
|
||||
transcript_path = output_dir / transcript_name
|
||||
with open(transcript_path, 'wt') as wf:
|
||||
if script_type == 'word':
|
||||
# add space between chinese char
|
||||
context = ''.join([f + ' ' for f in context])[:-1]
|
||||
wf.write(context + "\n")
|
||||
|
||||
|
||||
def reorganize_thchs30(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path]=None,
|
||||
script_type='phone'):
|
||||
root_dir = Path(root_dir).expanduser()
|
||||
output_dir = Path(output_dir).expanduser()
|
||||
output_dir.mkdir(parents=True, exist_ok=True)
|
||||
link_wav(root_dir, output_dir)
|
||||
write_lab(root_dir, output_dir, script_type)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Reorganize THCHS-30 dataset for MFA")
|
||||
parser.add_argument("--root-dir", type=str, help="path to thchs30 dataset.")
|
||||
parser.add_argument(
|
||||
"--output-dir",
|
||||
type=str,
|
||||
help="path to save outputs(audio and transcriptions)")
|
||||
|
||||
parser.add_argument(
|
||||
"--script-type",
|
||||
type=str,
|
||||
default="phone",
|
||||
help="type of lab ('word'/'syllable'/'phone')")
|
||||
args = parser.parse_args()
|
||||
reorganize_thchs30(args.root_dir, args.output_dir, args.script_type)
|
||||
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,63 @@
|
||||
#!/usr/bin/env 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.
|
||||
"""format manifest into wav.scp text.word [text.syllable text.phone]"""
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
from typing import Union
|
||||
|
||||
from deepspeech.frontend.utility import read_manifest
|
||||
|
||||
key_whitelist = set(['feat', 'text', 'syllable', 'phone'])
|
||||
filename = {
|
||||
'text': 'text.word',
|
||||
'syllable': 'text.syllable',
|
||||
'phone': 'text.phone',
|
||||
'feat': 'wav.scp',
|
||||
}
|
||||
|
||||
|
||||
def dump_manifest(manifest_path, output_dir: Union[str, Path]):
|
||||
|
||||
output_dir = Path(output_dir).expanduser()
|
||||
manifest_path = Path(manifest_path).expanduser()
|
||||
manifest_jsons = read_manifest(manifest_path)
|
||||
first_line = manifest_jsons[0]
|
||||
file_map = {}
|
||||
|
||||
for k in first_line.keys():
|
||||
if k not in key_whitelist:
|
||||
continue
|
||||
file_map[k] = open(output_dir / filename[k], 'w')
|
||||
|
||||
for line_json in manifest_jsons:
|
||||
for k in line_json.keys():
|
||||
if k not in key_whitelist:
|
||||
continue
|
||||
file_map[k].write(line_json['utt'] + ' ' + line_json[k] + '\n')
|
||||
|
||||
for _, file in file_map.items():
|
||||
file.close()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="dump manifest to wav.scp text.word ...")
|
||||
parser.add_argument("--manifest-path", type=str, help="path to manifest")
|
||||
parser.add_argument(
|
||||
"--output-dir",
|
||||
type=str,
|
||||
help="path to save outputs(audio and transcriptions)")
|
||||
args = parser.parse_args()
|
||||
dump_manifest(args.manifest_path, args.output_dir)
|
||||
Loading…
Reference in new issue