-
Notifications
You must be signed in to change notification settings - Fork 2
/
dataset_utils.py
155 lines (122 loc) · 5.1 KB
/
dataset_utils.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
146
147
148
149
150
151
152
153
154
155
import os
import pandas as pd
from tqdm import tqdm
import torch
import torchaudio
from torch.utils.data import DataLoader, Dataset, WeightedRandomSampler
from typing import Tuple, Union, List, Callable, Optional
import pathlib
class DatasetDownloader():
def __init__(self, key_word='sheila'):
self.key_word = key_word
self.datadir = "speech_commands"
if os.path.isfile('speech_commands_v0.01.tar.gz'):
print('Data is already downloaded.')
else:
print('Downloading data...')
os.system(
'wget http://download.tensorflow.org/data/speech_commands_v0.01.tar.gz -O speech_commands_v0.01.tar.gz')
os.system('mkdir speech_commands && tar -C speech_commands -xvzf speech_commands_v0.01.tar.gz 1> log')
print("Ready!")
self.samples_by_target = {
cls: [os.path.join(self.datadir, cls, name) for name in os.listdir("./{}/{}".format(self.datadir, cls))]
for cls in os.listdir(self.datadir)
if os.path.isdir(os.path.join(self.datadir, cls))
}
print('Classes:', ', '.join(sorted(self.samples_by_target.keys())[1:]))
def generate_labeled_data(self):
if os.path.isfile('labeled_data.csv'):
print('Data is already labeled')
labeled_data = pd.read_csv('labeled_data.csv')
background_noises = pd.read_csv('background_noises.csv')
return labeled_data, background_noises
labeled_data = pd.DataFrame(columns=['name', 'word', 'label'])
background_noises = pd.DataFrame(columns=['name'])
print('Creating labeled dataframe:')
for el in tqdm(self.samples_by_target.keys()):
if el != '_background_noise_':
for name in self.samples_by_target[el]:
word = name.split('/')[1]
if word == self.key_word:
label = 1
else:
label = 0
labeled_data = labeled_data.append({'name': name, 'word': word, 'label': label}, ignore_index=True)
elif el == '_background_noise_':
for name in self.samples_by_target[el]:
if 'README' not in name:
background_noises = background_noises.append(
{'name': name}, ignore_index=True
)
labeled_data.to_csv('labeled_data.csv', index=False)
background_noises.to_csv('background_noises.csv', index=False)
return labeled_data, background_noises
class TrainDataset(torch.utils.data.Dataset):
def __init__(self, root='', df=None, kw=None, transform=None):
"""
Args:
root (string): Directory with all the images.
df (pd.DataFrame): dataframe with annotations (filename, word and label).
kw (string): keyword to spot.
transform (callable, optional): Optional transform to be applied on a sample.
"""
self.root = root
self.kw = kw
self.df = df
self.transform = transform
def __len__(self):
return self.df.shape[0]
def __getitem__(self, idx):
utt_name = self.root + self.df.loc[idx, 'name']
utt = torchaudio.load(utt_name)[0].squeeze()
word = self.df.loc[idx, 'word']
label = self.df.loc[idx, 'label']
if self.transform:
utt = self.transform(utt)
sample = {'utt': utt, 'word': word, 'label': label}
return sample
class SpeechCommandDataset(Dataset):
def __init__(
self,
transform: Optional[Callable] = None,
path2dir: str = None,
keywords: Union[str, List[str]] = None,
csv: Optional[pd.DataFrame] = None
):
self.transform = transform
if csv is None:
path2dir = pathlib.Path(path2dir)
keywords = keywords if isinstance(keywords, list) else [keywords]
all_keywords = [
p.stem for p in path2dir.glob('*')
if p.is_dir() and not p.stem.startswith('_')
]
triplets = []
for keyword in all_keywords:
paths = (path2dir / keyword).rglob('*.wav')
if keyword in keywords:
for path2wav in paths:
triplets.append((path2wav.as_posix(), keyword, 1))
else:
for path2wav in paths:
triplets.append((path2wav.as_posix(), keyword, 0))
self.csv = pd.DataFrame(
triplets,
columns=['path', 'keyword', 'label']
)
else:
self.csv = csv
def __getitem__(self, index: int):
instance = self.csv.iloc[index]
path2wav = instance['path']
wav, sr = torchaudio.load(path2wav)
wav = wav.sum(dim=0)
if self.transform:
wav = self.transform(wav)
return {
'wav': wav,
'keywors': instance['keyword'],
'label': instance['label']
}
def __len__(self):
return len(self.csv)