def preprocess_ljspeech(args):
in_dir = os.path.join(args.base_dir, 'LJSpeech-1.0')
# out_dir = os.path.join(args.base_dir, args.output)
# os.makedirs(out_dir, exist_ok=True)
ljspeech.build_from_path(in_dir,
args.base_dir,
args.output,
args.num_workers,
tqdm=tqdm)
def main():
print('initializing preprocessing..')
parser = argparse.ArgumentParser()
parser.add_argument('--base_dir', default='')
parser.add_argument(
'--hparams',
default='',
help=
'Hyperparameter overrides as a comma-separated list of name=value pairs'
)
parser.add_argument('--dataset', default='MultiSets')
parser.add_argument('--output', default='training_data')
parser.add_argument('--n_jobs', type=int, default=cpu_count())
args = parser.parse_args()
# Prepare directories
# in_dir = os.path.join(args.base_dir, args.dataset)
# out_dir = os.path.join(args.base_dir, args.output)
in_dir = args.base_dir
out_dir = args.output
mel_dir = os.path.join(out_dir, 'mels')
wav_dir = os.path.join(out_dir, 'audio')
lin_dir = os.path.join(out_dir, 'linear')
os.makedirs(mel_dir, exist_ok=True)
os.makedirs(wav_dir, exist_ok=True)
os.makedirs(lin_dir, exist_ok=True)
# Process dataset
if args.dataset == 'LJSpeech-1.1':
metadata = ljspeech.build_from_path(hparams,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'DataBaker':
use_prosody = False
metadata = databaker.build_from_path(hparams,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'MultiSets':
metadata = multisets.build_from_path(hparams,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
else:
raise ValueError('Unsupported dataset provided: {} '.format(
args.dataset))
# Write metadata to 'train.txt' for training
write_metadata(metadata, out_dir)
def preprocess_ljspeech(args):
in_dir = os.path.join(args.base_dir, 'LJSpeech-1.1') # 'LJSpeech-1.1'
out_dir = os.path.join(args.base_dir, args.output) # 'training'
os.makedirs(out_dir, exist_ok=True) # 경로 존재해도 에러 보이지마마
metadata = ljspeech.build_from_path(
in_dir, out_dir, args.num_workers, tqdm=tqdm)
write_metadata(metadata, out_dir)
def preprocess_ljspeech(args):
in_dir = os.path.join(args.base_dir, 'database/LJSpeech-1.0')
out_dir = os.path.join(args.base_dir, args.output)
os.makedirs(out_dir, exist_ok=True)
metadata = ljspeech.build_from_path(in_dir,
out_dir,
args.num_workers,
tqdm=tqdm)
write_metadata(metadata, out_dir)
def preprocess_ljspeech(args, hparams):
in_dir = os.path.join(args.base_dir, 'LJSpeech-1.1')
out_dir = os.path.join(args.base_dir, args.output)
os.makedirs(out_dir, exist_ok=True)
metadata = ljspeech.build_from_path(in_dir,
out_dir,
hparams,
args.num_workers,
tqdm=tqdm)
write_metadata(metadata[:-args.validation_size - args.test_size], out_dir)
if args.validation_size > 0:
write_validation(
metadata[-args.validation_size - args.test_size:-args.test_size],
out_dir)
if args.test_size > 0:
write_validation(metadata[-args.test_size:],
out_dir,
filename='test.txt')
def test_build_from_path():
in_dir = r"D:\git\tacotron\data"
out_dir = r"D:\git\tacotron\data\specs"
build_from_path(in_dir=in_dir, out_dir=out_dir)
def preprocess_blizzard(args):
in_dir = os.path.join(args.base_dir, 'Blizzard2012')
out_dir = os.path.join(args.base_dir, args.output)
os.makedirs(out_dir, exist_ok=True)
metadata = blizzard.build_from_path(in_dir, out_dir, args.num_workers, tqdm=tqdm)
write_metadata(metadata, out_dir)
def preprocess_ljspeech(args):
in_dir = os.path.join(os.getcwd(), 'LJSpeech-1.0')
out_dir = os.path.join(os.getcwd(), args.output)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
# os.makedirs(out_dir, exist_ok=True)
metadata = ljspeech.build_from_path(in_dir, out_dir, args.num_workers, tqdm=tqdm)
write_metadata(metadata, out_dir)
def write_metadata(metadata, out_dir):
with open(os.path.join(out_dir, 'train.txt'), 'wb') as f:
for m in metadata:
f.write('|'.join([str(x) for x in m]) + '\n')
frames = sum([m[2] for m in metadata])
hours = frames * hparams.frame_shift_ms / (3600 * 1000)
print('Wrote %d utterances, %d frames (%.2f hours)' % (len(metadata), frames, hours))
print('Max input length: %d' % max(len(m[3]) for m in metadata))
print('Max output length: %d' % max(m[2] for m in metadata))
def main():
def main():
print('initializing preprocessing..')
parser = argparse.ArgumentParser()
parser.add_argument('--base_dir', default='')
parser.add_argument(
'--hparams',
default='',
help=
'Hyperparameter overrides as a comma-separated list of name=value pairs'
)
parser.add_argument('--dataset', default='chunchun')
parser.add_argument('--output', default='training_data')
parser.add_argument('--n_jobs', type=int, default=cpu_count())
args = parser.parse_args()
modified_hp = hparams.parse(args.hparams)
# Prepare directories
in_dir = os.path.join(args.base_dir, args.dataset)
out_dir = os.path.join(args.base_dir, args.output)
mel_dir = os.path.join(out_dir, 'mels')
wav_dir = os.path.join(out_dir, 'audio')
lin_dir = os.path.join(out_dir, 'linear')
os.makedirs(mel_dir, exist_ok=True)
os.makedirs(wav_dir, exist_ok=True)
os.makedirs(lin_dir, exist_ok=True)
# Process dataset
metadata = []
if args.dataset == 'chunchun':
use_prosody = True
in_dir = os.path.join(args.base_dir,
'chunchun/english/chunchun_english_lj')
print('processing chunchun CN.../n')
metadata_1 = chunchun_EN.build_from_path(modified_hp,
0,
0,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir,
'chunchun/chinese/chunchun_8k_all_v4')
print('processing chunchun EN.../n')
metadata_2 = chunchun_CN.build_from_path_CN(modified_hp,
0,
1,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
metadata = metadata_1 + metadata_2
elif args.dataset == 'all':
use_prosody = False
in_dir = os.path.join(args.base_dir, 'LJSpeech-1.1')
print('processing LJSpeech-1.1.../n')
metadata_1 = ljspeech.build_from_path(modified_hp,
0,
0,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'DataBaker')
print('processing DataBaker.../n')
metadata_2 = databaker.build_from_path_CN(modified_hp,
1,
1,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'TTS.HUawei.zhcmn.F.Deng')
print('processing TTS.HUawei.zhcmn.F.Deng.../n')
metadata_3 = Huawei.build_from_path_CN(modified_hp,
2,
1,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'TTS.Huawei.enus.F.XuYue')
print('processing TTS.Huawei.enus.F.XuYue.../n')
metadata_4 = Huawei.build_from_path_EN(modified_hp,
3,
0,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir,
'TTS.THCoSS.zhcmn.F.M/TH-CoSS/data/03FR00')
print('processing TTS.THCoSS.zhcmn.F.M 03FR00.../n')
metadata_5 = thcoss.build_from_path(modified_hp,
4,
1,
in_dir,
'a',
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir,
'TTS.THCoSS.zhcmn.F.M/TH-CoSS/data/03MR00')
print('processing TTS.THCoSS.zhcmn.F.M 03MR00.../n')
metadata_6 = thcoss.build_from_path(modified_hp,
5,
1,
in_dir,
'b',
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir,
'TTS.Pachira.zhcmn.enus.F.DB1/zh-cmn')
print('processing TTS.Pachira.zhcmn.enus.F.DB1/zh-cmn.../n')
metadata_7 = thcoss.build_from_path_simple(modified_hp,
6,
1,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'TTS.DataBaker.enus.M.DB1')
print('processing TTS.DataBaker.enus.M.DB1.../n')
metadata_8 = databaker.build_from_path_EN(modified_hp,
7,
0,
in_dir,
'x',
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'TTS.DataBaker.enus.F.DB1')
print('processing TTS.DataBaker.enus.F.DB1.../n')
metadata_9 = databaker.build_from_path_EN(modified_hp,
8,
0,
in_dir,
'y',
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
in_dir = os.path.join(args.base_dir, 'TTS.DataBaker.enus.F.DB2')
print('processing TTS.DataBaker.enus.F.DB2.../n')
metadata_10 = databaker.build_from_path_EN(modified_hp,
9,
0,
in_dir,
'z',
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
metadata = metadata_1 + metadata_2 + metadata_3 + metadata_4 + metadata_5 + metadata_6 + metadata_7 + metadata_8 + metadata_9 + metadata_10
elif args.dataset == 'LJSpeech-1.1':
metadata = ljspeech.build_from_path(modified_hp,
in_dir,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'DataBaker':
use_prosody = False
metadata = databaker.build_from_path_CN(modified_hp,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'THCoSS':
use_prosody = True
metadata = thcoss.build_from_path(modified_hp,
in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.n_jobs,
tqdm=tqdm)
else:
raise ValueError('Unsupported dataset provided: {} '.format(
args.dataset))
# Write metadata to 'train.txt' for training
write_metadata(metadata, out_dir)
def main():
print('Initializing preprocessing..')
parser = argparse.ArgumentParser()
parser.add_argument('--indir', required=True)
parser.add_argument('--dataset', default='MultiSets')
parser.add_argument('--outdir', default='training_data')
parser.add_argument('--config_file', default='./datasets/config16k.json')
parser.add_argument('--n_jobs', type=int, default=cpu_count())
args = parser.parse_args()
# Prepare directories
in_dir = args.indir
out_dir = args.outdir
mel_dir = os.path.join(out_dir, 'mels')
wav_dir = os.path.join(out_dir, 'audio')
lin_dir = os.path.join(out_dir, 'linear')
os.makedirs(mel_dir, exist_ok=True)
os.makedirs(wav_dir, exist_ok=True)
os.makedirs(lin_dir, exist_ok=True)
# Process dataset
if args.dataset == 'LJSpeech':
metadata = ljspeech.build_from_path(in_dir,
mel_dir,
lin_dir,
wav_dir,
args.config_file,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'DataBaker':
use_prosody = True
metadata = databaker.build_from_path(in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.config_file,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'MultiSets':
metadata = multisets.build_from_path(in_dir,
mel_dir,
lin_dir,
wav_dir,
args.config_file,
args.n_jobs,
tqdm=tqdm)
elif args.dataset == 'AIShell-3':
use_prosody = False
metadata = aishell3.build_from_path(in_dir,
use_prosody,
mel_dir,
lin_dir,
wav_dir,
args.config_file,
args.n_jobs,
tqdm=tqdm)
else:
raise ValueError('Unsupported dataset provided: {} '.format(
args.dataset))
# Write metadata to 'train.txt' for training
with open(args.config_file, 'r') as f:
sr = json.load(f)["sr"]
write_metadata(metadata, out_dir, sr)
