-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataloader.py
145 lines (113 loc) · 4.62 KB
/
dataloader.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
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 torch
import numpy as np
from torch.utils.data import Dataset, DataLoader
import config as cfg
import soundfile as sf
from torchvision.transforms import RandomCrop
import os
# save np.load
np_load_old = np.load
# modify the default parameters of np.load
np.load = lambda *a, **k: np_load_old(*a, allow_pickle=True, **k)
def create_dataloader(mode):
if mode == 'train':
return DataLoader(
dataset=Wave_Dataset(mode),
batch_size=cfg.batch, # max 3696 * snr types
shuffle=True,
num_workers=cfg.num_workers,
pin_memory=True,
drop_last=True,
sampler=None
)
elif mode == 'valid':
return DataLoader(
dataset=Wave_Dataset(mode),
batch_size=cfg.batch, shuffle=False, num_workers=cfg.num_workers
) # max 1152
def create_dataloader_for_test(mode):
if mode == 'test':
return DataLoader(
dataset=Wave_Dataset_for_test(mode),
batch_size=1, shuffle=False, num_workers=4
) # max 192
class Wave_Dataset(Dataset):
def __init__(self, mode):
# load data
if mode == 'train':
print('<Training dataset>')
print('Load the data...')
self.input_path = './input/train_dataset.npy'
self.noisy_dir = '/datasets/wav/train/noisy'
self.clean_dir = '/datasets/wav/train/clean'
elif mode == 'valid':
self.noisy_dir = '/datasets/wav/val/noisy'
self.clean_dir = '/datasets/wav/val/clean'
print('<Validation dataset>')
print('Load the data...')
self.input_path = './input/validation_dataset.npy'
# self.input = np.load(self.input_path)
self.file_names = self.GetFilenames(self.noisy_dir)
self.crop = RandomCrop((2, 48000), pad_if_needed=True)
def GetFilenames(self, direc):
folders = os.listdir(direc)
folders_path = [os.path.join(direc, x) for x in folders]
file_names = [[os.path.join(one_folder, file_name) for file_name in os.listdir(one_folder_path)]\
for one_folder, one_folder_path in zip(folders, folders_path)]
new_file_names = []
for file_name in file_names:
new_file_names += file_name
return new_file_names
def __len__(self):
return len(self.file_names)
def __getitem__(self, idx):
file_name = self.file_names[idx]
noisy_path = os.path.join(self.noisy_dir, file_name)
clean_path = os.path.join(self.clean_dir, file_name)
inputs, _ = sf.read(noisy_path)
labels, _ = sf.read(clean_path)
# inputs = self.input[idx][0]
# labels = self.input[idx][1]
# transform to torch from numpy
inputs = torch.from_numpy(inputs).unsqueeze(0)
labels = torch.from_numpy(labels).unsqueeze(0)
# print(inputs.shape, inputs.dtype)
sounds = self.crop(torch.cat([inputs, labels], dim=0))
inputs, labels = sounds[:1], sounds[1:]
return inputs, labels
class Wave_Dataset_for_test(Dataset):
def __init__(self, mode):
# load data
if mode == 'test':
print('<Test dataset>')
print('Load the data...')
self.input_path = './input/recon_test_dataset.npy'
self.noisy_dir = '/datasets/wav/val/noisy'
self.clean_dir = '/datasets/wav/val/clean'
self.file_names = self.GetFilenames(self.noisy_dir)
# print(self.file_names[:5])
# self.input = np.load(self.input_path)
def GetFilenames(self, direc):
folders = os.listdir(direc)
folders_path = [os.path.join(direc, x) for x in folders]
file_names = [[os.path.join(one_folder, file_name) for file_name in os.listdir(one_folder_path)]\
for one_folder, one_folder_path in zip(folders, folders_path)]
new_file_names = []
for file_name in file_names:
new_file_names += file_name
return new_file_names
def __len__(self):
return len(self.file_names)
def __getitem__(self, idx):
file_name = self.file_names[idx]
noisy_path = os.path.join(self.noisy_dir, file_name)
clean_path = os.path.join(self.clean_dir, file_name)
inputs, _ = sf.read(noisy_path)
labels, _ = sf.read(clean_path)
# inputs = self.input[idx][0]
# labels = self.input[idx][1]
# transform to torch from numpy
inputs = torch.from_numpy(inputs).unsqueeze(0)
labels = torch.from_numpy(labels).unsqueeze(0)
# print(inputs.shape, inputs.dtype)
return inputs, labels, file_name