This repository has been archived by the owner on Dec 21, 2023. It is now read-only.
forked from Rayhane-mamah/Tacotron-2
-
Notifications
You must be signed in to change notification settings - Fork 70
/
wavenet_preprocess.py
52 lines (42 loc) · 1.82 KB
/
wavenet_preprocess.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
import argparse
import os
from multiprocessing import cpu_count
from datasets import wavenet_preprocessor
from hparams import hparams
from tqdm import tqdm
def preprocess(args, input_dir, out_dir, hparams):
mel_dir = os.path.join(out_dir, 'mels')
wav_dir = os.path.join(out_dir, 'audio')
os.makedirs(mel_dir, exist_ok=True)
os.makedirs(wav_dir, exist_ok=True)
metadata = wavenet_preprocessor.build_from_path(hparams, input_dir, mel_dir, wav_dir, args.n_jobs, tqdm=tqdm)
write_metadata(metadata, out_dir)
def write_metadata(metadata, out_dir):
with open(os.path.join(out_dir, 'map.txt'), 'w', encoding='utf-8') as f:
for m in metadata:
f.write('|'.join([str(x) for x in m]) + '\n')
mel_frames = sum([int(m[5]) for m in metadata])
timesteps = sum([int(m[4]) for m in metadata])
sr = hparams.sample_rate
hours = timesteps / sr / 3600
print('Write {} utterances, {} audio timesteps, ({:.2f} hours)'.format(
len(metadata), timesteps, hours))
print('Max mel frames length: {}'.format(max(int(m[5]) for m in metadata)))
print('Max audio timesteps length: {}'.format(max(m[4] for m in metadata)))
def run_preprocess(args, hparams):
output_folder = os.path.join(args.base_dir, args.output)
preprocess(args, args.input_dir, output_folder, hparams)
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('--input_dir', default='LJSpeech-1.1/wavs')
parser.add_argument('--output', default='tacotron_output/gta/')
parser.add_argument('--n_jobs', type=int, default=cpu_count())
args = parser.parse_args()
modified_hp = hparams.parse(args.hparams)
run_preprocess(args, modified_hp)
if __name__ == '__main__':
main()