forked from xuanjihe/speech-emotion-recognition
-
Notifications
You must be signed in to change notification settings - Fork 0
/
ExtractMel.py
392 lines (359 loc) · 18.5 KB
/
ExtractMel.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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Tue Jan 9 20:32:28 2018
@author: hxj
"""
import wave
import numpy as np
import python_speech_features as ps
import os
import glob
import cPickle
#import base
#import sigproc
eps = 1e-5
def wgn(x, snr):
snr = 10**(snr/10.0)
xpower = np.sum(x**2)/len(x)
npower = xpower / snr
return np.random.randn(len(x)) * np.sqrt(npower)
def getlogspec(signal,samplerate=16000,winlen=0.02,winstep=0.01,
nfilt=26,nfft=399,lowfreq=0,highfreq=None,preemph=0.97,
winfunc=lambda x:np.ones((x,))):
highfreq= highfreq or samplerate/2
signal = ps.sigproc.preemphasis(signal,preemph)
frames = ps.sigproc.framesig(signal, winlen*samplerate, winstep*samplerate, winfunc)
pspec = ps.sigproc.logpowspec(frames,nfft)
return pspec
def read_file(filename):
file = wave.open(filename,'r')
params = file.getparams()
nchannels, sampwidth, framerate, wav_length = params[:4]
str_data = file.readframes(wav_length)
wavedata = np.fromstring(str_data, dtype = np.short)
#wavedata = np.float(wavedata*1.0/max(abs(wavedata))) # normalization)
time = np.arange(0,wav_length) * (1.0/framerate)
file.close()
return wavedata, time, framerate
def dense_to_one_hot(labels_dense, num_classes):
"""Convert class labels from scalars to one-hot vectors."""
num_labels = labels_dense.shape[0]
index_offset = np.arange(num_labels) * num_classes
labels_one_hot = np.zeros((num_labels, num_classes))
labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
return labels_one_hot
def zscore(data,mean,std):
shape = np.array(data.shape,dtype = np.int32)
for i in range(shape[0]):
data[i,:,:,0] = (data[i,:,:,0]-mean)/(std)
return data
def normalization(data):
'''
#apply zscore
mean = np.mean(data,axis=0)#axis=0纵轴方向求均值
std = np.std(data,axis=0)
train_data = zscore(train_data,mean,std)
test_data = zscore(test_data,mean,std)
'''
mean = np.mean(data,axis=0)#axis=0纵轴方向求均值
std = np.std(data,axis=0)
data = (data-mean)/std
return data
def mapminmax(data):
shape = np.array(data.shape,dtype = np.int32)
for i in range(shape[0]):
min = np.min(data[i,:,:,0])
max = np.max(data[i,:,:,0])
data[i,:,:,0] = (data[i,:,:,0] - min)/((max - min)+eps)
return data
def generate_label(emotion,classnum):
label = -1
if(emotion == 'ang'):
label = 0
elif(emotion == 'sad'):
label = 1
elif(emotion == 'hap'):
label = 2
elif(emotion == 'neu'):
label = 3
elif(emotion == 'fear'):
label = 4
else:
label = 5
return label
def load_data():
f = open('./zscore40.pkl','rb')
mean1,std1,mean2,std2,mean3,std3 = cPickle.load(f)
return mean1,std1,mean2,std2,mean3,std3
def read_IEMOCAP():
eps = 1e-5
trnum = 1723
tnum = 259
vnum = 298
test_num = 420
valid_num = 436
train_num = 2928
filter_num = 40
pernums_test = np.arange(tnum)#remerber each utterance contain how many segments
pernums_valid = np.arange(vnum)
rootdir = '/home/jamhan/hxj/datasets/IEMOCAP_full_release'
mean1,std1,mean2,std2,mean3,std3 = load_data()
#2774
hapnum = 434#2
angnum = 433#0
neunum = 1262#3
sadnum = 799#1
pernum = 300#np.min([hapnum,angnum,sadnum,neunum])
#valid_num = divmod((train_num),10)[0]
train_label = np.empty((train_num,1), dtype = np.int8)
test_label = np.empty((tnum,1), dtype = np.int8)
valid_label = np.empty((vnum,1), dtype = np.int8)
Test_label = np.empty((test_num,1), dtype = np.int8)
Valid_label = np.empty((valid_num,1), dtype = np.int8)
train_data = np.empty((train_num,300,filter_num,3),dtype = np.float32)
test_data = np.empty((test_num,300,filter_num,3),dtype = np.float32)
valid_data = np.empty((valid_num,300,filter_num,3),dtype = np.float32)
tnum = 0
vnum = 0
train_num = 0
test_num = 0
valid_num = 0
train_emt = {'hap':0,'ang':0,'neu':0,'sad':0 }
test_emt = {'hap':0,'ang':0,'neu':0,'sad':0 }
valid_emt = {'hap':0,'ang':0,'neu':0,'sad':0 }
for speaker in os.listdir(rootdir):
if(speaker[0] == 'S'):
sub_dir = os.path.join(rootdir,speaker,'sentences/wav')
emoevl = os.path.join(rootdir,speaker,'dialog/EmoEvaluation')
for sess in os.listdir(sub_dir):
if(sess[7] == 'i'):
emotdir = emoevl+'/'+sess+'.txt'
#emotfile = open(emotdir)
emot_map = {}
with open(emotdir,'r') as emot_to_read:
while True:
line = emot_to_read.readline()
if not line:
break
if(line[0] == '['):
t = line.split()
emot_map[t[3]] = t[4]
file_dir = os.path.join(sub_dir, sess, '*.wav')
files = glob.glob(file_dir)
for filename in files:
#wavname = filename[-23:-4]
wavname = filename.split("/")[-1][:-4]
emotion = emot_map[wavname]
if(emotion in ['hap','ang','neu','sad']):
data, time, rate = read_file(filename)
mel_spec = ps.logfbank(data,rate,nfilt = filter_num)
delta1 = ps.delta(mel_spec, 2)
delta2 = ps.delta(delta1, 2)
#apply zscore
time = mel_spec.shape[0]
if(speaker in ['Session1','Session2','Session3','Session4']):
#training set
if(time <= 300):
part = mel_spec
delta11 = delta1
delta21 = delta2
part = np.pad(part,((0,300 - part.shape[0]),(0,0)),'constant',constant_values = 0)
delta11 = np.pad(delta11,((0,300 - delta11.shape[0]),(0,0)),'constant',constant_values = 0)
delta21 = np.pad(delta21,((0,300 - delta21.shape[0]),(0,0)),'constant',constant_values = 0)
train_data[train_num,:,:,0] = (part -mean1)/(std1+eps)
train_data[train_num,:,:,1] = (delta11 - mean2)/(std2+eps)
train_data[train_num,:,:,2] = (delta21 - mean3)/(std3+eps)
em = generate_label(emotion,6)
train_label[train_num] = em
train_emt[emotion] = train_emt[emotion] + 1
train_num = train_num + 1
else:
if(emotion in ['ang','neu','sad']):
for i in range(2):
if(i == 0):
begin = 0
end = begin + 300
else:
begin = time - 300
end = time
part = mel_spec[begin:end,:]
delta11 = delta1[begin:end,:]
delta21 = delta2[begin:end,:]
train_data[train_num,:,:,0] = (part -mean1)/(std1+eps)
train_data[train_num,:,:,1] = (delta11 - mean2)/(std2+eps)
train_data[train_num,:,:,2] = (delta21 - mean3)/(std3+eps)
em = generate_label(emotion,6)
train_label[train_num] = em
train_emt[emotion] = train_emt[emotion] + 1
train_num = train_num + 1
else:
frames = divmod(time-300,100)[0] + 1
for i in range(frames):
begin = 100*i
end = begin + 300
part = mel_spec[begin:end,:]
delta11 = delta1[begin:end,:]
delta21 = delta2[begin:end,:]
train_data[train_num,:,:,0] = (part -mean1)/(std1+eps)
train_data[train_num,:,:,1] = (delta11 - mean2)/(std2+eps)
train_data[train_num,:,:,2] = (delta21 - mean3)/(std3+eps)
em = generate_label(emotion,6)
train_label[train_num] = em
train_emt[emotion] = train_emt[emotion] + 1
train_num = train_num + 1
else:
em = generate_label(emotion,6)
if(wavname[-4] == 'M'):
#test_set
test_label[tnum] = em
if(time <= 300):
pernums_test[tnum] = 1
part = mel_spec
delta11 = delta1
delta21 = delta2
part = np.pad(part,((0,300 - part.shape[0]),(0,0)),'constant',constant_values = 0)
delta11 = np.pad(delta11,((0,300 - delta11.shape[0]),(0,0)),'constant',constant_values = 0)
delta21 = np.pad(delta21,((0,300 - delta21.shape[0]),(0,0)),'constant',constant_values = 0)
test_data[test_num,:,:,0] = (part -mean1)/(std1+eps)
test_data[test_num,:,:,1] = (delta11 - mean2)/(std2+eps)
test_data[test_num,:,:,2] = (delta21 - mean3)/(std3+eps)
test_emt[emotion] = test_emt[emotion] + 1
Test_label[test_num] = em
test_num = test_num + 1
tnum = tnum + 1
else:
pernums_test[tnum] = 2
tnum = tnum + 1
for i in range(2):
if(i == 0):
begin = 0
end = begin + 300
else:
end = time
begin = time - 300
part = mel_spec[begin:end,:]
delta11 = delta1[begin:end,:]
delta21 = delta2[begin:end,:]
test_data[test_num,:,:,0] = (part -mean1)/(std1+eps)
test_data[test_num,:,:,1] = (delta11 - mean2)/(std2+eps)
test_data[test_num,:,:,2] = (delta21 - mean3)/(std3+eps)
test_emt[emotion] = test_emt[emotion] + 1
Test_label[test_num] = em
test_num = test_num + 1
else:
#valid_set
em = generate_label(emotion,6)
valid_label[vnum] = em
if(time <= 300):
pernums_valid[vnum] = 1
part = mel_spec
delta11 = delta1
delta21 = delta2
part = np.pad(part,((0,300 - part.shape[0]),(0,0)),'constant',constant_values = 0)
delta11 = np.pad(delta11,((0,300 - delta11.shape[0]),(0,0)),'constant',constant_values = 0)
delta21 = np.pad(delta21,((0,300 - delta21.shape[0]),(0,0)),'constant',constant_values = 0)
valid_data[valid_num,:,:,0] = (part -mean1)/(std1+eps)
valid_data[valid_num,:,:,1] = (delta11 - mean2)/(std2+eps)
valid_data[valid_num,:,:,2] = (delta21 - mean3)/(std3+eps)
valid_emt[emotion] = valid_emt[emotion] + 1
Valid_label[valid_num] = em
valid_num = valid_num + 1
vnum = vnum + 1
else:
pernums_valid[vnum] = 2
vnum = vnum + 1
for i in range(2):
if(i == 0):
begin = 0
end = begin + 300
else:
end = time
begin = time - 300
part = mel_spec[begin:end,:]
delta11 = delta1[begin:end,:]
delta21 = delta2[begin:end,:]
valid_data[valid_num,:,:,0] = (part -mean1)/(std1+eps)
valid_data[valid_num,:,:,1] = (delta11 - mean2)/(std2+eps)
valid_data[valid_num,:,:,2] = (delta21 - mean3)/(std3+eps)
valid_emt[emotion] = valid_emt[emotion] + 1
Valid_label[valid_num] = em
valid_num = valid_num + 1
else:
pass
hap_index = np.arange(hapnum)
neu_index = np.arange(neunum)
sad_index = np.arange(sadnum)
ang_index = np.arange(angnum)
h2 = 0
a0 = 0
n3 = 0
s1 = 0
for l in range(train_num):
if(train_label[l] == 0):
ang_index[a0] = l
a0 = a0 + 1
elif (train_label[l] == 1):
sad_index[s1] = l
s1 = s1 + 1
elif (train_label[l] == 2):
hap_index[h2] = l
h2 = h2 + 1
else:
neu_index[n3] = l
n3 = n3 + 1
for m in range(8):
np.random.shuffle(neu_index)
np.random.shuffle(hap_index)
np.random.shuffle(sad_index)
np.random.shuffle(ang_index)
#define emotional array
hap_label = np.empty((pernum,1), dtype = np.int8)
ang_label = np.empty((pernum,1), dtype = np.int8)
sad_label = np.empty((pernum,1), dtype = np.int8)
neu_label = np.empty((pernum,1), dtype = np.int8)
hap_data = np.empty((pernum,300,filter_num,3),dtype = np.float32)
neu_data = np.empty((pernum,300,filter_num,3),dtype = np.float32)
sad_data = np.empty((pernum,300,filter_num,3),dtype = np.float32)
ang_data = np.empty((pernum,300,filter_num,3),dtype = np.float32)
hap_data = train_data[hap_index[0:pernum]].copy()
hap_label = train_label[hap_index[0:pernum]].copy()
ang_data = train_data[ang_index[0:pernum]].copy()
ang_label = train_label[ang_index[0:pernum]].copy()
sad_data = train_data[sad_index[0:pernum]].copy()
sad_label = train_label[sad_index[0:pernum]].copy()
neu_data = train_data[neu_index[0:pernum]].copy()
neu_label = train_label[neu_index[0:pernum]].copy()
train_num = 4*pernum
Train_label = np.empty((train_num,1), dtype = np.int8)
Train_data = np.empty((train_num,300,filter_num,3),dtype = np.float32)
Train_data[0:pernum] = hap_data
Train_label[0:pernum] = hap_label
Train_data[pernum:2*pernum] = sad_data
Train_label[pernum:2*pernum] = sad_label
Train_data[2*pernum:3*pernum] = neu_data
Train_label[2*pernum:3*pernum] = neu_label
Train_data[3*pernum:4*pernum] = ang_data
Train_label[3*pernum:4*pernum] = ang_label
arr = np.arange(train_num)
np.random.shuffle(arr)
Train_data = Train_data[arr[0:]]
Train_label = Train_label[arr[0:]]
print train_label.shape
print train_emt
print test_emt
print valid_emt
#print test_label[0:500,:]
#f=open('./CASIA_40_delta.pkl','wb')
#output = './IEMOCAP40.pkl'
output = './IEMOCAP'+str(m)+'.pkl'
f=open(output,'wb')
cPickle.dump((Train_data,Train_label,test_data,test_label,valid_data,valid_label,Valid_label,Test_label,pernums_test,pernums_valid),f)
f.close()
return
if __name__=='__main__':
read_IEMOCAP()
#print "test_num:", test_num
#print "train_num:", train_num
# n = wgn(x, 6)
# xn = x+n # 增加了6dBz信噪比噪声的信号