/
speaker_verification_MSP.py
316 lines (255 loc) · 13 KB
/
speaker_verification_MSP.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
import io
import sys
import torch
import torch.nn as nn
from torchvision import models
from PIL import Image
import torchvision.transforms as transforms
from encoder import inference as encoder
from encoder import audio
import librosa
from pathlib import Path
import json
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from itertools import combinations
from tqdm import tqdm
import pickle
tokensize=sys.argv[3]
#
# Author: Morgan Sandler (sandle20@msu.edu)
# The purpose of this file is to perform a speaker verification for E-Vector over the MSP-Podcast testing sets
# Note: this file will also perform the VAD/other preprocessing necessary if the get_speaker_embedding flag is set to preprocessing=True (which is default nature)
#
# NOTE: Here is the syntax to run: python speaker_verification_MSP.py <val/test1/test2> <dataset_root_path> <tokensize>
# example: python speaker_verification_MSP.py val /research/iprobe/datastore/datasets/speech/utd-msppodcast_v1.8/ 20token
def get_model():
#model_save_path = Path('/research/iprobe-sandle20/Playground/evector/encoder/saved_models/first_backups/first_bak_105000.pt') # NOTE: Add your own path here to your saved model
#model_save_path = Path('/research/iprobe-sandle20/Playground/evector/encoder/saved_models/20tokens_backups/20tokens_bak_105000.pt') # NOTE: Add your own path here to your saved model
model_save_path = Path('/research/iprobe-sandle20/Playground/evector/encoder/saved_models/5tokens_backups/5tokens_bak_105000.pt') # NOTE: Add your own path here to your saved model
module_name = 'model_GST'
encoder.load_model(model_save_path, module_name=module_name)
return encoder
def get_tensor(file_path, preprocess=True, sampling_rate=8000, duration=None):
if(preprocess):
ref_audio = encoder.preprocess_wav(file_path)
else:
ref_audio, sr = librosa.load(file_path, sr=sampling_rate)
if(duration is not None):
ref_audio = ref_audio[0:int(duration*sampling_rate)]
return ref_audio
encoder = get_model()
def get_speaker_embedding(file_path, preprocess=True, sampling_rate=16000, duration=None, normalize=True):
ref_audio = get_tensor(file_path, preprocess=preprocess, sampling_rate=sampling_rate, duration=duration)
#print('ref audio', ref_audio.shape)
embed, partial_embeds, _ = encoder.embed_utterance(ref_audio, return_partials=True)
if embed is None:
return None
#print(embed.shape)
if(normalize):
embed = embed / np.linalg.norm(embed)
return embed
# Main Program
testing_set = sys.argv[1]
dataset_root = sys.argv[2] + '/Labels/labels_concensus.csv'
if not dataset_root:
pass
else:
# Preprocess all speakers
import pandas as pd
msp_information = pd.read_csv(dataset_root)
train_set = msp_information[msp_information['Split_Set'] == 'Train'][msp_information['SpkrID'] != 'Unknown']
validation_set = msp_information[msp_information['Split_Set'] == 'Validation'][msp_information['SpkrID'] != 'Unknown']
test1_set = msp_information[msp_information['Split_Set'] == 'Test1'][msp_information['SpkrID'] != 'Unknown']
test2_set = msp_information[msp_information['Split_Set'] == 'Test2'][msp_information['SpkrID'] != 'Unknown']
results = []
if testing_set == 'val':
print('Validation set selected')
print(validation_set)
combos = list(combinations(validation_set['FileName'], 2))
print(len(combos), 'combinations of files in the VAL set')
validation_set.set_index('FileName',inplace=True)
speaker_embed_cache = {}
#print(split_arrays)
try:
for sample_a, sample_b in tqdm(combos):
sample_a_info = validation_set.loc[sample_a]
sample_b_info = validation_set.loc[sample_b]
#print(sample_a_info, sample_b_info)
path_a = Path(sys.argv[2] + 'Audios/' +sample_a)
path_b = Path(sys.argv[2] + 'Audios/' +sample_b)
#print(path_a, path_b)
# Simple caching system for computed embeddings
if sample_a in speaker_embed_cache:
#print('using cache')
embed1 = speaker_embed_cache[sample_a]
else:
embed1 = get_speaker_embedding(path_a)
speaker_embed_cache[sample_a] = embed1
if sample_b in speaker_embed_cache:
#print('using cache')
embed2 = speaker_embed_cache[sample_b]
else:
embed2 = get_speaker_embedding(path_b)
speaker_embed_cache[sample_b] = embed2
if embed1 is None or embed2 is None:
continue # skip this test
assert embed1.shape == embed2.shape
match_score = cosine_similarity(embed1.reshape(1, -1), embed2.reshape(1, -1))
#print('Match Score', match_score)
row = {
'FileNameA': sample_a_info.name,
'EmoClassA': sample_a_info['EmoClass'],
'EmoActA': sample_a_info['EmoAct'],
'EmoValA': sample_a_info['EmoVal'],
'EmoDomA': sample_a_info['EmoDom'],
'SpkrIDA': sample_a_info['SpkrID'],
'GenderA': sample_a_info['Gender'],
'SplitSetA': sample_a_info['Split_Set'],
'FileNameB': sample_b_info.name,
'EmoClassB': sample_b_info['EmoClass'],
'EmoActB': sample_b_info['EmoAct'],
'EmoValB': sample_b_info['EmoVal'],
'EmoDomB': sample_b_info['EmoDom'],
'SpkrIDB': sample_b_info['SpkrID'],
'GenderB': sample_b_info['Gender'],
'SplitSetB': sample_b_info['Split_Set'],
'MatchScore': match_score[0][0]
}
results.append(row)
except KeyboardInterrupt:
# save early if keeb interrupt
final_df = pd.DataFrame(results)
final_df.to_csv('ExperimentData/'+tokensize+'/'+'val_results.csv')
with open('ExperimentData/'+tokensize+'/'+'val_speaker_embeddings.pickle', 'wb') as handle:
pickle.dump(speaker_embed_cache, handle, protocol=pickle.HIGHEST_PROTOCOL)
print('Data Saved')
elif testing_set == 'test1':
print('Test1 set selected')
print(test1_set)
combos = list(combinations(test1_set['FileName'], 2))
print(len(combos), 'combinations of files in the TEST1 set')
test1_set.set_index('FileName',inplace=True)
speaker_embed_cache = {}
#print(split_arrays)
try:
for sample_a, sample_b in tqdm(combos):
sample_a_info = test1_set.loc[sample_a]
sample_b_info = test1_set.loc[sample_b]
#print(sample_a_info, sample_b_info)
path_a = Path(sys.argv[2] + 'Audios/' +sample_a)
path_b = Path(sys.argv[2] + 'Audios/' +sample_b)
#print(path_a, path_b)
# Simple caching system for computed embeddings
if sample_a in speaker_embed_cache:
#print('using cache')
embed1 = speaker_embed_cache[sample_a]
else:
embed1 = get_speaker_embedding(path_a)
speaker_embed_cache[sample_a] = embed1
if sample_b in speaker_embed_cache:
#print('using cache')
embed2 = speaker_embed_cache[sample_b]
else:
embed2 = get_speaker_embedding(path_b)
speaker_embed_cache[sample_b] = embed2
if embed1 is None or embed2 is None:
continue # skip this test
assert embed1.shape == embed2.shape
match_score = cosine_similarity(embed1.reshape(1, -1), embed2.reshape(1, -1))
#print('Match Score', match_score)
row = {
'FileNameA': sample_a_info.name,
'EmoClassA': sample_a_info['EmoClass'],
'EmoActA': sample_a_info['EmoAct'],
'EmoValA': sample_a_info['EmoVal'],
'EmoDomA': sample_a_info['EmoDom'],
'SpkrIDA': sample_a_info['SpkrID'],
'GenderA': sample_a_info['Gender'],
'SplitSetA': sample_a_info['Split_Set'],
'FileNameB': sample_b_info.name,
'EmoClassB': sample_b_info['EmoClass'],
'EmoActB': sample_b_info['EmoAct'],
'EmoValB': sample_b_info['EmoVal'],
'EmoDomB': sample_b_info['EmoDom'],
'SpkrIDB': sample_b_info['SpkrID'],
'GenderB': sample_b_info['Gender'],
'SplitSetB': sample_b_info['Split_Set'],
'MatchScore': match_score[0][0]
}
results.append(row)
except KeyboardInterrupt:
# save early if keeb interrupt
final_df = pd.DataFrame(results)
final_df.to_csv('ExperimentData/'+tokensize+'/'+'test1_results.csv')
with open('ExperimentData/'+tokensize+'/'+'test1_speaker_embeddings.pickle', 'wb') as handle:
pickle.dump(speaker_embed_cache, handle, protocol=pickle.HIGHEST_PROTOCOL)
print('Data Saved')
elif testing_set == 'test2':
print('Test2 set selected')
print(test2_set)
combos = list(combinations(test2_set['FileName'], 2))
print(len(combos), 'combinations of files in the TEST2 set')
test2_set.set_index('FileName',inplace=True)
speaker_embed_cache = {}
#print(split_arrays)
try:
for sample_a, sample_b in tqdm(combos):
sample_a_info = test2_set.loc[sample_a]
sample_b_info = test2_set.loc[sample_b]
#print(sample_a_info, sample_b_info)
path_a = Path(sys.argv[2] + 'Audios/' +sample_a)
path_b = Path(sys.argv[2] + 'Audios/' +sample_b)
#print(path_a, path_b)
# Simple caching system for computed embeddings
if sample_a in speaker_embed_cache:
#print('using cache')
embed1 = speaker_embed_cache[sample_a]
else:
embed1 = get_speaker_embedding(path_a)
speaker_embed_cache[sample_a] = embed1
if sample_b in speaker_embed_cache:
#print('using cache')
embed2 = speaker_embed_cache[sample_b]
else:
embed2 = get_speaker_embedding(path_b)
speaker_embed_cache[sample_b] = embed2
if embed1 is None or embed2 is None:
continue # skip this test
assert embed1.shape == embed2.shape
match_score = cosine_similarity(embed1.reshape(1, -1), embed2.reshape(1, -1))
#print('Match Score', match_score)
row = {
'FileNameA': sample_a_info.name,
'EmoClassA': sample_a_info['EmoClass'],
'EmoActA': sample_a_info['EmoAct'],
'EmoValA': sample_a_info['EmoVal'],
'EmoDomA': sample_a_info['EmoDom'],
'SpkrIDA': sample_a_info['SpkrID'],
'GenderA': sample_a_info['Gender'],
'SplitSetA': sample_a_info['Split_Set'],
'FileNameB': sample_b_info.name,
'EmoClassB': sample_b_info['EmoClass'],
'EmoActB': sample_b_info['EmoAct'],
'EmoValB': sample_b_info['EmoVal'],
'EmoDomB': sample_b_info['EmoDom'],
'SpkrIDB': sample_b_info['SpkrID'],
'GenderB': sample_b_info['Gender'],
'SplitSetB': sample_b_info['Split_Set'],
'MatchScore': match_score[0][0]
}
results.append(row)
except KeyboardInterrupt:
# save early if keeb interrupt
final_df = pd.DataFrame(results)
final_df.to_csv('ExperimentData/'+tokensize+'/'+'test2_results.csv')
with open('ExperimentData/'+tokensize+'/'+'test2_speaker_embeddings.pickle', 'wb') as handle:
pickle.dump(speaker_embed_cache, handle, protocol=pickle.HIGHEST_PROTOCOL)
print('Data Saved')
else:
print('oops that set doesnt exist. try again')
final_df = pd.DataFrame(results)
final_df.to_csv('ExperimentData/'+tokensize+'/'+testing_set+'_results.csv')
with open('ExperimentData/'+tokensize+'/'+testing_set+'_speaker_embeddings.pickle', 'wb') as handle:
pickle.dump(speaker_embed_cache, handle, protocol=pickle.HIGHEST_PROTOCOL)
print('Data Saved')