AutoVC.pytorch/preprocess.py at master · peisuke/AutoVC.pytorch · GitHub

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
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
import argparse
import math
import json
import tqdm
import os
import glob
import sys
import tqdm
import numpy as np
import librosa
import random
from hparams import hparams as hp
from resemblyzer import VoiceEncoder, preprocess_wav

def compute_embed(files, encoder):
    emb = []

    files = random.sample(files, min(len(files), 20))
    for f in files:
        wav = preprocess_wav(f)
        e = encoder.embed_utterance(wav)
        emb.append(e)
    emb = np.array(emb)
    emb = emb.mean(axis=0)
    return emb

def create_data(data, output_dir, phase, seq_len=None):
    ret = []

    for k, v in tqdm.tqdm(data.items()):
        speaker_dir = os.path.join(output_dir, phase, k)
        os.makedirs(speaker_dir, exist_ok=True)

        emb_file = os.path.join(speaker_dir, 'emb.npy')
        np.save(emb_file, v['emb'])

        for fname in v['files']:
            n = os.path.splitext(os.path.basename(fname))[0]
            target_dir = os.path.join(speaker_dir, n)
            os.makedirs(target_dir, exist_ok=True)
            x, _ = librosa.load(fname, sr=hp.sample_rate)

            files = []
            if seq_len is not None:
                for i, p in enumerate(range(0, len(x), seq_len)):
                    wav = x[p:p+seq_len*2]
                    if len(wav) < seq_len:
                        wav = np.pad(wav, (0, seq_len - len(wav)), mode='constant')

                    filename = os.path.join(target_dir, '{0:04d}.wav'.format(i))
                    librosa.output.write_wav(filename, wav, hp.sample_rate)
                    files.append(filename)
            else:
                filename = os.path.join(target_dir, '{0:04d}.wav'.format(0))
                librosa.output.write_wav(filename, x, hp.sample_rate)
                files.append(filename)

            ret.append({'wav': files, 'emb': emb_file})

    return ret

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description='')
    parser.add_argument('--wav-dir', type=str, required=True, help='The directory that contains wave files [./wavs]')
    parser.add_argument('--output-dir', default='./data', type=str, help='Output dir [./data]')
    args = parser.parse_args()

    wav_dir = args.wav_dir
    output_dir = args.output_dir

    os.makedirs(output_dir, exist_ok=True)

    speakers = sorted(glob.glob(os.path.join(wav_dir, '*')))
    speakers = list(filter(lambda x: os.path.isdir(x), speakers))

    #### DEBUG
    #speakers = speakers[:5]

    random.seed(1234)
    random.shuffle(speakers)

    nb_speakers = len(speakers)
    nb_unseen_speakers = nb_speakers // 5
    nb_seen_speakers = nb_speakers - nb_unseen_speakers

    seen_speakers = speakers[:nb_seen_speakers]
    unseen_speakers = speakers[nb_seen_speakers:]

    encoder = VoiceEncoder()

    train_data = {}
    test_data = {}
    unseen_data = {}

    for s in seen_speakers:
        speaker = os.path.basename(s)
        wavfiles = sorted(glob.glob(os.path.join(s, '*.wav')))

        #### DEBUG
        #wavfiles = wavfiles[:10]

        emb = compute_embed(wavfiles, encoder)

        nb_files = len(wavfiles)
        nb_test = nb_files // 10

        trainfiles = wavfiles[:-nb_test]
        testfiles = wavfiles[-nb_test:]

        train_data[speaker] = {'files': trainfiles, 'emb': emb}
        test_data[speaker] = {'files': testfiles, 'emb': emb}

    for s in unseen_speakers:
        speaker = os.path.basename(s)
        wavfiles = sorted(glob.glob(os.path.join(s, '*.wav')))

        emb = compute_embed(wavfiles, encoder)

        #### DEBUG
        #wavfiles = wavfiles[:10]

        unseen_data[speaker] = {'files': wavfiles, 'emb': emb}

    train_data = create_data(train_data, output_dir, 'train', hp.seq_len)
    test_data = create_data(test_data, output_dir, 'test')
    unseen_data = create_data(unseen_data, output_dir, 'unseen')

    with open(os.path.join(output_dir, 'train_data.json'), 'w') as f:
        json.dump(train_data, f)

    with open(os.path.join(output_dir, 'test_data.json'), 'w') as f:
        json.dump(test_data, f)

    with open(os.path.join(output_dir, 'unseen_data.json'), 'w') as f:
        json.dump(unseen_data, f)