PyTorch-VAE/vqvae_dataloader.py at master · waNGYC55/PyTorch-VAE · 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
136
137
138
139
140
141
142
143
144
145
#%%
import os
import random
import numpy as np
import math
import torch
import soundfile as sf
from torch.utils.data import Dataset
from torch.utils.data import DataLoader
import torchaudio
import time

#%%
class MyDataLoader(DataLoader):
    def __init__(self, dataset, batch_size=1, num_workers=10):
        self.bs = batch_size
        self.n_samples = len(dataset)
        self.num_workers = num_workers
        super(self.__class__, self).__init__(batch_sampler=self.batch_gen(), num_workers=self.num_workers, dataset=dataset)

    def __len__(self):
        return math.floor(self.n_samples / self.bs)

    def batch_gen(self):
        for i in range(math.floor(self.n_samples / self.bs)):
            ind = np.arange(self.bs).reshape(self.bs, 1) + i * self.bs
            yield ind

class VQVAE_Dataset(Dataset):
    def __init__(self, utt2data, sequence_size=22000*2, need_aug=False, shuffle=True, with_output=True):
        self.utt2data = utt2data
        self.dataset_size = len(self.utt2data)
        self.shuffle = shuffle
        self.need_aug = need_aug
        self.sequence_size = sequence_size
        self.with_output=with_output

        if shuffle:
            random.shuffle(self.utt2data)

    def __len__(self):
        return self.dataset_size

    def read_feat(self, filename):

        samples = self.sequence_size
        try:
            feat, sr = torchaudio.load(filename)
            if feat.shape[1] == samples:
                new_feat = feat
            elif feat.shape[1] > samples:
                start_point = random.randrange(0, feat.shape[1] - samples)
                new_feat = np.array(feat[:,start_point:start_point + samples])
                #print(new_feat.shape)
            else:
                new_feat = np.zeros((1, samples))
                pad_beg = int((samples - feat.shape[1])/ 2)
                new_feat[:,pad_beg:pad_beg + feat.shape[1]] = feat
                #print(new_feat.shape)
            assert new_feat.shape[1] == samples
            return new_feat
        except:
            print(filename)

        return 1

    def __getitem__(self, sample_idx):
        # try:
        #     idx = int(sample_idx[0])
        # except:
        #     idx = sample_idx
        idx = int(sample_idx[0])
        assert 0 <= idx and idx < self.dataset_size, "invalid index"

        utt, filename = self.utt2data[idx]
        feat = self.read_feat(filename)
        feat = feat.astype('float32')
        #print(feat.shape)
        if self.with_output:
            return utt, feat, feat
        else:
            return utt, feat

class VQVAE_Mel_Dataset(VQVAE_Dataset):

    def __init__(self, utt2data, sequence_size=int(256), need_aug=False, shuffle=True, with_output=True):
        super().__init__(utt2data, sequence_size, need_aug, shuffle, with_output)

    def read_feat(self, filename):
        samples = self.sequence_size
        try:
            feat=np.load(filename)
            if feat.shape[1] == samples:
                new_feat = feat
            elif feat.shape[1] > samples:
                start_point = random.randrange(0, feat.shape[1] - samples)
                new_feat = np.array(feat[:,start_point:start_point + samples])
                #print(new_feat.shape)
            else:
                new_feat = np.zeros((feat.shape[0], samples))
                pad_beg = int((samples - feat.shape[1])/ 2)
                new_feat[:,pad_beg:pad_beg + feat.shape[1]] = feat
                #print(new_feat.shape)
            assert new_feat.shape[1] == samples
            return torch.tensor(new_feat, dtype=torch.float)
        except:
            print(filename)
        return 1

    def __getitem__(self, sample_idx):
        # try:
        #     idx = int(sample_idx[0])
        # except:
        #     idx = sample_idx
        idx = int(sample_idx[0])
        assert 0 <= idx and idx < self.dataset_size, "invalid index"

        utt, filename = self.utt2data[idx]
        feat = self.read_feat(filename)
        #print(feat.shape)
        if self.with_output:
            return utt, feat, feat
        else:
            return utt, feat

#%%
##############################
def test():
    batch_size = 1
    num_workers = 1
    index_root_dir = './index'
    utt2wav = [line.split() for line in open(os.path.join(index_root_dir, 'train.index'))]


    dev_dataset = VQVAE_Mel_Dataset(utt2wav, need_aug=False, with_output=True, shuffle=False)
    dear_dev_dataloader = MyDataLoader(dev_dataset, batch_size=batch_size, num_workers=1)

    print("Data loader prepared...")
    for utt, feat_in, feat_out in dear_dev_dataloader:
        # print(utt)
        print(feat_in.shape)
        time.sleep(1)
if __name__ == "__main__":
    test()
# %%