-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_io.py
322 lines (298 loc) · 13.8 KB
/
data_io.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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
import torch
import numpy as np
import os
from torch.utils.data import Dataset, DataLoader, ConcatDataset
import python_speech_features as ps
from tqdm import tqdm
import librosa
REAL=1
FAKE=0
__author__ = "JunyanWu"
__email__ = "[email protected]"
class ASVspoof2019LA(Dataset):
def __init__(self,part, path="/data/wujy/audio/asvspoof/ASVspoof2019_LA"):
self.part = part
if self.part == "train":
protocol_path=os.path.join(path, "ASVspoof2019_LA_cm_protocols","ASVspoof2019.LA.cm.train.trn.txt")
elif self.part == "dev":
protocol_path=os.path.join(path, "ASVspoof2019_LA_cm_protocols","ASVspoof2019.LA.cm.dev.trl.txt")
elif self.part == "eval":
protocol_path=os.path.join(path, "ASVspoof2019_LA_cm_protocols", "ASVspoof2019.LA.cm.eval.trl.txt")
self.label_dict={"bonafide":REAL,"spoof":FAKE}
self.protocol = np.loadtxt(protocol_path,dtype=str)
self.path = [path for i in range(len(self.protocol))]
# attack types
self.spoof2fn_dict={}
for i in range(len(self.protocol)):
spoof_label=self.protocol[i,-2] if self.protocol[i,-2]!='-' else 'bonafide'
filename=self.protocol[i,1]
if spoof_label not in self.spoof2fn_dict:
self.spoof2fn_dict[spoof_label]=[filename]
else:
self.spoof2fn_dict[spoof_label].append(filename)
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
filename = protocol_idx[1]
filepath = os.path.join(self.path[idx],"ASVspoof2019_LA_%s"%self.part, 'flac',filename ".flac")
label = torch.tensor(self.label_dict[protocol_idx[4]], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
def __len__(self):
return len(self.protocol)
def load_audio(self,filepath):
wave, fs = librosa.load(filepath, sr=16000)
return np.array(wave,dtype=float)
def extract_mel(self,x):
begin, end=0, 300
mel_spec = ps.logfbank(x, 16000, nfilt = 40)
delta1 = ps.delta(mel_spec, 2)
delta2 = ps.delta(delta1, 2)
fea=[mel_spec[begin:end,:],delta1[begin:end,:],delta2[begin:end,:]]
return np.array(fea,dtype=float)
def pad(self, x, max_len=64600):
x_len = x.shape[0]
if x_len >= max_len:
return x[:max_len]
num_repeats = int(max_len / x_len) 1
padded_x = np.tile(x, (1, num_repeats))[:, :max_len][0]
return np.array(padded_x,dtype=float)
def spoof_label_dict(self):
ans={}
count=0
spooflabels=sorted(self.spoof2fn_dict.keys())
for sl in spooflabels:
ans[sl]=self.spoof2fn_dict[sl]
count =len(self.spoof2fn_dict[sl])
return ans
class ASVspoof2021LA(ASVspoof2019LA):
def __init__(self, part,path="/data/wujy/audio/asvspoof"):
self.label_dict={"bonafide":REAL, "spoof":FAKE}
self.path=os.path.join(path,"ASVspoof2021_LA_eval")
protocol_path="./tools/evaluate/keys/LA-keys-stage-1/keys/CM/trial_metadata.txt"
#ASVspoof2021_LA_eval/keys/LA/CM/trial_metadata.txt
protocol = np.loadtxt(protocol_path,dtype=str)
if part=="eval":
protocol_mask=(protocol[:,7]=="eval")
self.protocol=protocol[protocol_mask]
attack2spooflabel={
"none":"L01",
"alaw":"L02",
"pstn":"L03",
"g722":"L04",
"ulaw":"L05",
"gsm":"L06",
"opus":"L07",
}
attacks=self.protocol[:,2]
self.spoof2fn_dict={}
for i in range(len(self.protocol)):
spoof_label=attack2spooflabel[attacks[i]]
filename=self.protocol[i,1]
if spoof_label not in self.spoof2fn_dict:
self.spoof2fn_dict[spoof_label]=[filename]
else:
self.spoof2fn_dict[spoof_label].append(filename)
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
filename=protocol_idx[1]
filepath = os.path.join(self.path,'flac',filename ".flac")
label = torch.tensor(self.label_dict[protocol_idx[5]], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
def spoof_label_dict(self):
ans={}
count=0
spooflabels=sorted(self.spoof2fn_dict.keys())
for sl in spooflabels:
ans[sl]=self.spoof2fn_dict[sl]
count =len(self.spoof2fn_dict[sl])
return ans
class ASVspoof2021DF(ASVspoof2019LA):
def __init__(self, part,path="/data/wujy/audio/asvspoof"):
self.label_dict={"bonafide":REAL, "spoof":FAKE}
self.path=os.path.join(path,"ASVspoof2021_DF_eval")
protocol_path="./tools/evaluate/keys/DF-keys-stage-1/keys/CM/trial_metadata.txt"
#"ASVspoof2021_DF_eval/keys/DF/CM/trial_metadata.txt"
protocol = np.loadtxt(protocol_path,dtype=str)
if part=="eval":
protocol_mask=(protocol[:,7]=="eval")
self.protocol=protocol[protocol_mask]
attack2spooflabel={
"bonafide":"bonafide",
"nocodec":"D01",
"low_mp3":"D02",
"high_mp3":"D03",
"low_m4a":"D04",
"high_m4a":"D05",
"low_ogg":"D06",
"high_ogg":"D07",
"mp3m4a":"D08",
"oggm4a":"D09",
}
attacks=self.protocol[:,2]
self.spoof2fn_dict={}
for i in range(len(self.protocol)):
spoof_label=attack2spooflabel[attacks[i]]
filename=self.protocol[i,1]
if spoof_label not in self.spoof2fn_dict:
self.spoof2fn_dict[spoof_label]=[filename]
else:
self.spoof2fn_dict[spoof_label].append(filename)
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
filename=protocol_idx[1]
filepath = os.path.join(self.path,'flac',filename ".flac")
label = torch.tensor(self.label_dict[protocol_idx[5]], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
def spoof_label_dict(self):
count=0
ans={}
spooflabels=sorted(self.spoof2fn_dict.keys())
for sl in spooflabels:
ans[sl]=self.spoof2fn_dict[sl]
count =len(self.spoof2fn_dict[sl])
return ans
class ASVspoof2015(ASVspoof2019LA):
def __init__(self,part,path="/data/wujy/audio/asvspoof/ASVspoof2015" ):
self.part = part
if self.part == "train":
protocol_path=os.path.join(path, "CM_protocol","cm_train.trn")
elif self.part == "dev":
protocol_path=os.path.join(path, "CM_protocol","cm_develop.ndx")
elif self.part == "eval":
protocol_path=os.path.join(path, "CM_protocol", "cm_evaluation.ndx")
self.label_dict={"human":REAL, "spoof":FAKE}
self.path = path
self.protocol = np.loadtxt(protocol_path,dtype=str)
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
dirname = protocol_idx[0]
filename = protocol_idx[1]
filepath = os.path.join(self.path,'wav',dirname,filename ".wav")
label = torch.tensor(self.label_dict[protocol_idx[3]], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
class inthewild(ASVspoof2019LA):
def __init__(self,path="/data/wujy/audio/inthewild/release_in_the_wild"):
protocol_path=os.path.join(path, "meta.csv")
self.label_dict={"bona-fide":REAL, "spoof":FAKE}
self.path = path
self.protocol = np.loadtxt(protocol_path,delimiter = ",",dtype=str)[1:]
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
filename = protocol_idx[0]
filepath = os.path.join(self.path,filename)
label = torch.tensor(self.label_dict[protocol_idx[2]], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
class FakeAVCeleb(ASVspoof2019LA):
def __init__(self, part, path="/data/wujy/audio/fac/FakeAVCeleb_v1.2"):
protocol_path=os.path.join(path, "meta_data.csv")
self.label_dict={"RealVideo-RealAudio":REAL, "RealVideo-FakeAudio":FAKE,"FakeVideo-RealAudio":REAL, "FakeVideo-FakeAudio":FAKE}
self.path = path
self.protocol = []
lines=open(protocol_path,'r').readlines()
for line in lines[1:]:
line=line.rstrip('\n')
line_split=line.split(",")
self.protocol.append(line_split)
self.protocol=np.array(self.protocol,dtype=str)
# https://github.com/DASH-Lab/FakeAVCeleb/blob/main/train_main.py
# val_ratio=0.3, validation ratio on trainset
if part=='dev':
self.protocol=self.protocol[:int(0.3*len(self.protocol))]
elif part=='eval':
self.protocol=self.protocol
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
dirname = protocol_idx[9].replace("FakeAVCeleb/","")
filename = protocol_idx[8]
mp4filepath = os.path.join(self.path, dirname,filename)
filepath = mp4filepath.replace("mp4","wav")
label = torch.tensor(self.label_dict[protocol_idx[5]], dtype=torch.float32)
if not os.path.exists(filepath):
cmd="ffmpeg -i '%s' -acodec pcm_s16le -f s16le -ac 1 -ar 16000 -f wav '%s'"%(mp4filepath,filepath)
os.system(cmd)
featureTensor = self.load_audio(filepath)
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
class FoR(ASVspoof2019LA):
def __init__(self,part="dev", path="/data/wujy/audio/for/for-original"):
part=part.replace("eval","testing").replace("dev","validation")
self.label_dict={"real":REAL, "fake":FAKE}
self.path=os.path.join(path,part)
self.protocol=[os.path.join(i,kk) for i,j,k in os.walk(self.path) for kk in k]
def __getitem__(self, idx):
protocol_idx=self.protocol[idx]
filepath = protocol_idx
filename = os.path.basename(filepath)
label_=os.path.dirname(filepath).split("/")[-1]
label = torch.tensor(self.label_dict[label_], dtype=torch.float32)
featureTensor = self.load_audio(filepath)
if len(featureTensor.shape)==2: # more than one channel
featureTensor=featureTensor[0]
featureTensor = self.pad(featureTensor)
SpecTensor = self.extract_mel(featureTensor)
featureTensor = torch.tensor(torch.Tensor(featureTensor), dtype=torch.float32)
SpecTensor = torch.tensor(torch.Tensor(SpecTensor), dtype=torch.float32)
featureTensor = torch.squeeze(featureTensor,dim=0)
return featureTensor, SpecTensor, label, filename
def get_dataloader(if_train=False,if_dev=False,if_eval=False,ename='',batch_size=16,num_workers=8):
if if_train:
dst=ASVspoof2019LA(part='train')
elif if_dev:
dst1=ASVspoof2019LA(part='dev')
dst2=FoR(part='dev')
dst3=FakeAVCeleb(part='dev')
dst = ConcatDataset([dst1, dst2, dst3])
del dst1,dst2,dst3
else:
if ename=='asvs2019la':
dst=ASVspoof2019LA(part="eval")
elif ename=='asvs2015e':
dst=ASVspoof2015(part="eval")
elif ename=='asvs2015d':
dst=ASVspoof2015(part="dev")
elif ename=='asvs2021df':
dst=ASVspoof2021DF(part="eval")
elif ename=='asvs2021la':
dst=ASVspoof2021LA(part="eval")
elif ename=='fac':
dst=FakeAVCeleb(part="eval")
elif ename=='for':
dst=FoR(part="eval")
elif ename=='itw':
dst=inthewild()
dlr=torch.utils.data.DataLoader(dst, batch_size=batch_size,num_workers=num_workers, shuffle=if_train)
del dst
return dlr