-
Notifications
You must be signed in to change notification settings - Fork 109
/
preprocess_sr.py
100 lines (88 loc) · 3.33 KB
/
preprocess_sr.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
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
import os
import argparse
import torch
import librosa
import json
from glob import glob
from tqdm import tqdm
from scipy.io import wavfile
import utils
from mel_processing import mel_spectrogram_torch
from wavlm import WavLM, WavLMConfig
#import h5py
import logging
logging.getLogger('numba').setLevel(logging.WARNING)
def process(filename):
basename = os.path.basename(filename)
speaker = filename.split("/")[-2]#basename[:4]
wav_dir = os.path.join(args.wav_dir, speaker)
ssl_dir = os.path.join(args.ssl_dir, speaker)
os.makedirs(wav_dir, exist_ok=True)
os.makedirs(ssl_dir, exist_ok=True)
wav, _ = librosa.load(filename, sr=hps.sampling_rate)
wav = torch.from_numpy(wav).unsqueeze(0).cuda()
mel = mel_spectrogram_torch(
wav,
hps.n_fft,
hps.num_mels,
hps.sampling_rate,
hps.hop_size,
hps.win_size,
hps.fmin,
hps.fmax
)
'''
f = {}
for i in range(args.min, args.max+1):
fpath = os.path.join(ssl_dir, f"{i}.hdf5")
f[i] = h5py.File(fpath, "a")
'''
for i in range(args.min, args.max+1):
mel_rs = utils.transform(mel, i)
wav_rs = vocoder(mel_rs)[0][0].detach().cpu().numpy()
_wav_rs = librosa.resample(wav_rs, orig_sr=hps.sampling_rate, target_sr=args.sr)
wav_rs = torch.from_numpy(_wav_rs).cuda().unsqueeze(0)
c = utils.get_content(cmodel, wav_rs)
ssl_path = os.path.join(ssl_dir, basename.replace(".wav", f"_{i