Merge pull request #126 from PaddlePaddle/revert-124-use_mfa
Revert "add use_mfa example"
This commit is contained in:
Feiyu Chan
GitHub
commit
9ffdd7abd8
|
|
@ -1,48 +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.
|
||||
|
||||
import argparse
|
||||
from collections import OrderedDict
|
||||
from pathlib import Path
|
||||
import logging
|
||||
|
||||
|
||||
def detect_oov(corpus_dir, lexicon_path, transcription_pattern="*.lab"):
|
||||
corpus_dir = Path(corpus_dir)
|
||||
|
||||
lexicon = OrderedDict()
|
||||
with open(lexicon_path, 'rt') as f:
|
||||
for line in f:
|
||||
syllable, phonemes = line.split(maxsplit=1)
|
||||
lexicon[syllable] = phonemes
|
||||
|
||||
for fp in corpus_dir.glob(transcription_pattern):
|
||||
syllables = fp.read_text().strip().split()
|
||||
for s in syllables:
|
||||
if s not in lexicon:
|
||||
logging.warning(f"{fp.relative_to(corpus_dir)} has OOV {s} .")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="detect oov in a corpus given a lexicon")
|
||||
parser.add_argument(
|
||||
"corpus_dir", type=str, help="corpus dir for MFA alignment.")
|
||||
parser.add_argument("lexicon_path", type=str, help="dictionary to use.")
|
||||
parser.add_argument(
|
||||
"--pattern", type=str, default="*.lab", help="dictionary to use.")
|
||||
args = parser.parse_args()
|
||||
print(args)
|
||||
|
||||
detect_oov(args.corpus_dir, args.lexicon_path, args.pattern)
|
||||
|
|
@ -1,199 +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.
|
||||
"""Generate lexicon and symbols for Mandarin Chinese phonology.
|
||||
The lexicon is used for Montreal Force Aligner.
|
||||
|
||||
Note that syllables are used as word in this lexicon. Since syllables rather
|
||||
than words are used in transcriptions produced by `reorganize_baker.py`.
|
||||
|
||||
We make this choice to better leverage other software for chinese text to
|
||||
pinyin tools like pypinyin. This is the convention for G2P in Chinese.
|
||||
"""
|
||||
|
||||
import re
|
||||
import argparse
|
||||
from collections import OrderedDict
|
||||
|
||||
INITIALS = [
|
||||
'b', 'p', 'm', 'f', 'd', 't', 'n', 'l', 'g', 'k', 'h', 'zh', 'ch', 'sh',
|
||||
'r', 'z', 'c', 's', 'j', 'q', 'x'
|
||||
]
|
||||
|
||||
FINALS = [
|
||||
'a', 'ai', 'ao', 'an', 'ang', 'e', 'er', 'ei', 'en', 'eng', 'o', 'ou',
|
||||
'ong', 'ii', 'iii', 'i', 'ia', 'iao', 'ian', 'iang', 'ie', 'io', 'iou',
|
||||
'iong', 'in', 'ing', 'u', 'ua', 'uai', 'uan', 'uang', 'uei', 'uo', 'uen',
|
||||
'ueng', 'v', 've', 'van', 'vn'
|
||||
]
|
||||
|
||||
SPECIALS = ['sil', 'sp']
|
||||
|
||||
|
||||
def rule(C, V, R, T):
|
||||
"""Generate a syllable given the initial, the final, erhua indicator, and tone.
|
||||
Orthographical rules for pinyin are applied. (special case for y, w, ui, un, iu)
|
||||
|
||||
Note that in this system, 'ü' is alway written as 'v' when appeared in phoneme, but converted to
|
||||
'u' in syllables when certain conditions are satisfied.
|
||||
|
||||
'i' is distinguished when appeared in phonemes, and separated into 3 categories, 'i', 'ii' and 'iii'.
|
||||
|
||||
Erhua is is possibly applied to every finals, except for finals that already ends with 'r'.
|
||||
|
||||
When a syllable is impossible or does not have any characters with this pronunciation, return None
|
||||
to filter it out.
|
||||
"""
|
||||
|
||||
# 不可拼的音节, ii 只能和 z, c, s 拼
|
||||
if V in ["ii"] and (C not in ['z', 'c', 's']):
|
||||
return None
|
||||
# iii 只能和 zh, ch, sh, r 拼
|
||||
if V in ['iii'] and (C not in ['zh', 'ch', 'sh', 'r']):
|
||||
return None
|
||||
|
||||
# 齐齿呼或者撮口呼不能和 f, g, k, h, zh, ch, sh, r, z, c, s
|
||||
if (V not in ['ii', 'iii']) and V[0] in ['i', 'v'] and (
|
||||
C in ['f', 'g', 'k', 'h', 'zh', 'ch', 'sh', 'r', 'z', 'c', 's']):
|
||||
return None
|
||||
|
||||
# 撮口呼只能和 j, q, x l, n 拼
|
||||
if V.startswith("v"):
|
||||
# v, ve 只能和 j ,q , x, n, l 拼
|
||||
if V in ['v', 've']:
|
||||
if C not in ['j', 'q', 'x', 'n', 'l', '']:
|
||||
return None
|
||||
# 其他只能和 j, q, x 拼
|
||||
else:
|
||||
if C not in ['j', 'q', 'x', '']:
|
||||
return None
|
||||
|
||||
# j, q, x 只能和齐齿呼或者撮口呼拼
|
||||
if (C in ['j', 'q', 'x']) and not (
|
||||
(V not in ['ii', 'iii']) and V[0] in ['i', 'v']):
|
||||
return None
|
||||
|
||||
# b, p ,m, f 不能和合口呼拼，除了 u 之外
|
||||
# bm p, m, f 不能和撮口呼拼
|
||||
if (C in ['b', 'p', 'm', 'f']) and ((V[0] in ['u', 'v'] and V != "u") or
|
||||
V == 'ong'):
|
||||
return None
|
||||
|
||||
# ua, uai, uang 不能和 d, t, n, l, r, z, c, s 拼
|
||||
if V in ['ua', 'uai', 'uang'
|
||||
] and C in ['d', 't', 'n', 'l', 'r', 'z', 'c', 's']:
|
||||
return None
|
||||
|
||||
# sh 和 ong 不能拼
|
||||
if V == 'ong' and C in ['sh']:
|
||||
return None
|
||||
|
||||
# o 和 gkh, zh ch sh r z c s 不能拼
|
||||
if V == "o" and C in [
|
||||
'd', 't', 'n', 'g', 'k', 'h', 'zh', 'ch', 'sh', 'r', 'z', 'c', 's'
|
||||
]:
|
||||
return None
|
||||
|
||||
# ueng 只是 weng 这个 ad-hoc 其他情况下都是 ong
|
||||
if V == 'ueng' and C != '':
|
||||
return
|
||||
|
||||
# 非儿化的 er 只能单独存在
|
||||
if V == 'er' and C != '':
|
||||
return None
|
||||
|
||||
if C == '':
|
||||
if V in ["i", "in", "ing"]:
|
||||
C = 'y'
|
||||
elif V == 'u':
|
||||
C = 'w'
|
||||
elif V.startswith('i') and V not in ["ii", "iii"]:
|
||||
C = 'y'
|
||||
V = V[1:]
|
||||
elif V.startswith('u'):
|
||||
C = 'w'
|
||||
V = V[1:]
|
||||
elif V.startswith('v'):
|
||||
C = 'yu'
|
||||
V = V[1:]
|
||||
else:
|
||||
if C in ['j', 'q', 'x']:
|
||||
if V.startswith('v'):
|
||||
V = re.sub('v', 'u', V)
|
||||
if V == 'iou':
|
||||
V = 'iu'
|
||||
elif V == 'uei':
|
||||
V = 'ui'
|
||||
elif V == 'uen':
|
||||
V = 'un'
|
||||
result = C + V
|
||||
|
||||
# Filter er 不能再儿化
|
||||
if result.endswith('r') and R == 'r':
|
||||
return None
|
||||
|
||||
# ii and iii, change back to i
|
||||
result = re.sub(r'i+', 'i', result)
|
||||
|
||||
result = result + R + T
|
||||
return result
|
||||
|
||||
|
||||
def generate_lexicon(with_tone=False, with_erhua=False):
|
||||
"""Generate lexicon for Mandarin Chinese."""
|
||||
syllables = OrderedDict()
|
||||
|
||||
for C in [''] + INITIALS:
|
||||
for V in FINALS:
|
||||
for R in [''] if not with_erhua else ['', 'r']:
|
||||
for T in [''] if not with_tone else ['1', '2', '3', '4', '5']:
|
||||
result = rule(C, V, R, T)
|
||||
if result:
|
||||
syllables[result] = f'{C} {V}{R}{T}'
|
||||
return syllables
|
||||
|
||||
|
||||
def generate_symbols(lexicon):
|
||||
"""Generate phoneme list for a lexicon."""
|
||||
symbols = set()
|
||||
for p in SPECIALS:
|
||||
symbols.add(p)
|
||||
for syllable, phonemes in lexicon.items():
|
||||
phonemes = phonemes.split()
|
||||
for p in phonemes:
|
||||
symbols.add(p)
|
||||
return sorted(list(symbols))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Generate lexicon for Chinese pinyin to phoneme for MFA")
|
||||
parser.add_argument("output", type=str, help="Path to save lexicon.")
|
||||
parser.add_argument(
|
||||
"--with-tone", action="store_true", help="whether to consider tone.")
|
||||
parser.add_argument(
|
||||
"--with-r", action="store_true", help="whether to consider erhua.")
|
||||
args = parser.parse_args()
|
||||
|
||||
lexicon = generate_lexicon(args.with_tone, args.with_r)
|
||||
symbols = generate_symbols(lexicon)
|
||||
|
||||
with open(args.output + ".lexicon", 'wt') as f:
|
||||
for k, v in lexicon.items():
|
||||
f.write(f"{k} {v}\n")
|
||||
|
||||
with open(args.output + ".symbols", 'wt') as f:
|
||||
for s in symbols:
|
||||
f.write(s + "\n")
|
||||
|
||||
print("Done!")
|
||||
|
|
@ -1,113 +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.
|
||||
"""Script to reorganize Baker dataset so as to use Montreal Force
|
||||
Aligner to align transcription and audio.
|
||||
|
||||
Please refer to https://montreal-forced-aligner.readthedocs.io/en/latest/data_prep.html
|
||||
for more details about Montreal Force Aligner's requirements on cotpus.
|
||||
|
||||
For scripts to reorganize other corpus, please refer to
|
||||
https://github.com/MontrealCorpusTools/MFA-reorganization-scripts
|
||||
for more details.
|
||||
"""
|
||||
|
||||
import os
|
||||
import shutil
|
||||
import argparse
|
||||
from typing import Union
|
||||
from pathlib import Path
|
||||
from concurrent.futures import ThreadPoolExecutor
|
||||
|
||||
import librosa
|
||||
import soundfile as sf
|
||||
from tqdm import tqdm
|
||||
|
||||
|
||||
def get_transcripts(path: Union[str, Path]):
|
||||
transcripts = {}
|
||||
|
||||
with open(path) as f:
|
||||
lines = f.readlines()
|
||||
|
||||
for i in range(0, len(lines), 2):
|
||||
sentence_id = lines[i].split()[0]
|
||||
transcription = lines[i + 1].strip()
|
||||
# tones are dropped here
|
||||
# since the lexicon does not consider tones, too
|
||||
transcription = " ".join([item[:-1] for item in transcription.split()])
|
||||
transcripts[sentence_id] = transcription
|
||||
|
||||
return transcripts
|
||||
|
||||
|
||||
def resample_and_save(source, target, sr=16000):
|
||||
wav, _ = librosa.load(str(source), sr=sr)
|
||||
sf.write(str(target), wav, samplerate=sr, subtype='PCM_16')
|
||||
return target
|
||||
|
||||
|
||||
def reorganize_baker(root_dir: Union[str, Path],
|
||||
output_dir: Union[str, Path]=None,
|
||||
resample_audio=False):
|
||||
root_dir = Path(root_dir).expanduser()
|
||||
transcript_path = root_dir / "ProsodyLabeling" / "000001-010000.txt"
|
||||
transcriptions = get_transcripts(transcript_path)
|
||||
|
||||
wave_dir = root_dir / "Wave"
|
||||
wav_paths = sorted(list(wave_dir.glob("*.wav")))
|
||||
output_dir = Path(output_dir).expanduser()
|
||||
assert wave_dir != output_dir, "Don't use an the original wav's directory as output_dir"
|
||||
|
||||
output_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
if resample_audio:
|
||||
with ThreadPoolExecutor(os.cpu_count()) as pool:
|
||||
with tqdm(total=len(wav_paths), desc="resampling") as pbar:
|
||||
futures = []
|
||||
for wav_path in wav_paths:
|
||||
future = pool.submit(resample_and_save, wav_path,
|
||||
output_dir / wav_path.name)
|
||||
future.add_done_callback(lambda p: pbar.update())
|
||||
futures.append(future)
|
||||
|
||||
results = []
|
||||
for ft in futures:
|
||||
results.append(ft.result())
|
||||
else:
|
||||
for wav_path in tqdm(wav_paths, desc="copying"):
|
||||
shutil.copyfile(wav_path, output_dir / wav_path.name)
|
||||
|
||||
for sentence_id, transcript in tqdm(
|
||||
transcriptions.items(), desc="transcription process"):
|
||||
with open(output_dir / (sentence_id + ".lab"), 'wt') as f:
|
||||
f.write(transcript)
|
||||
f.write('\n')
|
||||
print("Done!")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Reorganize Baker dataset for MFA")
|
||||
parser.add_argument("--root-dir", type=str, help="path to baker dataset.")
|
||||
parser.add_argument(
|
||||
"--output-dir",
|
||||
type=str,
|
||||
help="path to save outputs(audio and transcriptions)")
|
||||
parser.add_argument(
|
||||
"--resample-audio",
|
||||
action="store_true",
|
||||
help="To resample audio files or just copy them")
|
||||
args = parser.parse_args()
|
||||
|
||||
reorganize_baker(args.root_dir, args.output_dir, args.resample_audio)
|
||||
|
|
@ -1,49 +0,0 @@
|
|||
EXP_DIR=exp
|
||||
|
||||
mkdir -p $EXP_DIR
|
||||
LEXICON_NAME='simple'
|
||||
if [ ! -f "$EXP_DIR/$LEXICON_NAME.lexicon" ]; then
|
||||
echo "generating lexicon..."
|
||||
python local/generate_lexicon.py "$EXP_DIR/$LEXICON_NAME" --with-r
|
||||
echo "lexicon done"
|
||||
fi
|
||||
|
||||
if [ ! -d $EXP_DIR/baker_corpus ]; then
|
||||
echo "reorganizing baker corpus..."
|
||||
python local/recorganize_baker.py --root-dir=~/datasets/BZNSYP --output-dir=$EXP_DIR/baker_corpus --resample-audio
|
||||
echo "reorganization done. Check output in $EXP_DIR/baker_corpus."
|
||||
echo "audio files are resampled to 16kHz"
|
||||
echo "transcription for each audio file is saved with the same namd in $EXP_DIR/baker_corpus "
|
||||
fi
|
||||
|
||||
echo "detecting oov..."
|
||||
python local/detect_oov.py $EXP_DIR/baker_corpus $EXP_DIR/"$LEXICON_NAME.lexicon"
|
||||
echo "detecting oov done. you may consider regenerate lexicon if there is unexpected OOVs."
|
||||
|
||||
|
||||
MFA_DOWNLOAD_DIR=local/
|
||||
|
||||
if [ ! -f "$MFA_DOWNLOAD_DIR/montreal-forced-aligner_linux.tar.gz" ]; then
|
||||
echo "downloading mfa..."
|
||||
(cd $MFA_DOWNLOAD_DIR && wget https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner/releases/download/v1.0.1/montreal-forced-aligner_linux.tar.gz)
|
||||
echo "download mfa done!"
|
||||
fi
|
||||
|
||||
if [ ! -d "$MFA_DOWNLOAD_DIR/montreal-forced-aligner" ]; then
|
||||
echo "extracting mfa..."
|
||||
(cd $MFA_DOWNLOAD_DIR && tar xvf "montreal-forced-aligner_linux.tar.gz")
|
||||
echo "extraction done!"
|
||||
fi
|
||||
|
||||
export PATH="$MFA_DOWNLOAD_DIR/montreal-forced-aligner/bin"
|
||||
if [ ! -d "$EXP_DIR/baker_alignment" ]; then
|
||||
echo "Start MFA training..."
|
||||
mfa_train_and_align $EXP_DIR/baker_corpus "$EXP_DIR/$LEXICON_NAME.lexicon" $EXP_DIR/baker_alignment -o $EXP_DIR/baker_model --clean --verbose --temp_directory exp/.mfa_train_and_align
|
||||
echo "training done!"
|
||||
echo "results: $EXP_DIR/baker_alignment"
|
||||
echo "model: $EXP_DIR/baker_model"
|
||||
fi
|
||||
|
||||
|
||||
|
||||
|
||||
Loading…
Reference in New Issue