-
Notifications
You must be signed in to change notification settings - Fork 4
/
dataset.py
42 lines (36 loc) · 1.13 KB
/
dataset.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
from glob import glob
import numpy as np
import os
import torch
import torchaudio
torchaudio.set_audio_backend("sox")
class AudioDataset(torch.utils.data.Dataset):
def __init__(self, paths, params):
super().__init__()
self.filenames = []
self.audio_length = params['audio_length']
for path in paths:
self.filenames += glob(f'{path}/**/*.wav', recursive=True)
def __len__(self):
return len(self.filenames)
def __getitem__(self, idx):
audio_filename = self.filenames[idx]
signal, _ = torchaudio.load(audio_filename)
signal = signal[0, :self.audio_length]
# renormalize the audio
scaler = max(signal.max(), -signal.min())
if scaler > 0:
signal = signal / scaler
return {
'audio': signal
}
def from_path(data_dirs, params):
dataset = AudioDataset(data_dirs, params)
return torch.utils.data.DataLoader(
dataset,
batch_size=params['batch_size'],
collate_fn=None,
shuffle=True,
num_workers=os.cpu_count(),
pin_memory=True,
drop_last=True)