-
Notifications
You must be signed in to change notification settings - Fork 4
/
inference.py
328 lines (307 loc) · 11.7 KB
/
inference.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
import argparse
import numpy as np
from pathlib import Path
import sys
import random
from tqdm import tqdm
from math import fabs
import torch
import torch.nn.functional as F
from model import SpeakerNet
from utils import read_config, tuneThresholdfromScore
parser = argparse.ArgumentParser(description="SpeakerNet")
# YAML config
parser.add_argument('--config', type=str, default=None)
# Data loader
parser.add_argument('--max_frames',
type=int,
default=100,
help='Input length to the network for training')
parser.add_argument(
'--eval_frames',
type=int,
default=100,
help='Input length to the network for testing; 0 for whole files')
parser.add_argument('--batch_size',
type=int,
default=320,
help='Batch size, number of speakers per batch')
parser.add_argument('--max_seg_per_spk',
type=int,
default=100,
help='Maximum number of utterances per speaker per epoch')
parser.add_argument('--nDataLoaderThread',
type=int,
default=8,
help='Number of loader threads')
parser.add_argument('--augment',
action='store_true',
default=False,
help='Augment input')
# Training details
parser.add_argument('--device', type=str, default="cuda", help='cuda or cpu')
parser.add_argument('--test_interval',
type=int,
default=10,
help='Test and save every [test_interval] epochs')
parser.add_argument('--max_epoch',
type=int,
default=500,
help='Maximum number of epochs')
parser.add_argument('--trainfunc',
type=str,
default="softmaxproto",
help='Loss function')
# Optimizer
parser.add_argument('--optimizer',
type=str,
default="adam",
help='sgd or adam')
parser.add_argument('--scheduler',
type=str,
default="steplr",
help='Learning rate scheduler')
parser.add_argument('--lr', type=float, default=0.001, help='Learning rate')
parser.add_argument("--lr_decay",
type=float,
default=0.95,
help='Learning rate decay every [test_interval] epochs')
parser.add_argument('--weight_decay',
type=float,
default=0,
help='Weight decay in the optimizer')
# Loss functions
parser.add_argument(
"--hard_prob",
type=float,
default=0.5,
help='Hard negative mining probability, otherwise random, only for some loss functions'
)
parser.add_argument(
"--hard_rank",
type=int,
default=10,
help='Hard negative mining rank in the batch, only for some loss functions'
)
parser.add_argument('--margin',
type=float,
default=1,
help='Loss margin, only for some loss functions')
parser.add_argument('--scale',
type=float,
default=15,
help='Loss scale, only for some loss functions')
parser.add_argument(
'--nPerSpeaker',
type=int,
default=2,
help='Number of utterances per speaker per batch, only for metric learning based losses'
)
parser.add_argument(
'--nClasses',
type=int,
default=400,
help='Number of speakers in the softmax layer, only for softmax-based losses')
# Load and save
parser.add_argument('--initial_model',
type=str,
default="checkpoints/final_500.model",
help='Initial model weights')
parser.add_argument('--save_path',
type=str,
default="exp",
help='Path for model and logs')
# Training and test data
parser.add_argument('--train_list',
type=str,
default="dataset/train.def.txt",
help='Train list')
parser.add_argument('--test_list',
type=str,
default="dataset/val.def.txt",
help='Evaluation list')
parser.add_argument('--test_path',
type=str,
default="dataset/",
help='Absolute path to the test set')
parser.add_argument('--musan_path',
type=str,
default="dataset/musan_split",
help='Absolute path to the test set')
parser.add_argument('--rir_path',
type=str,
default="dataset/RIRS_NOISES/simulated_rirs",
help='Absolute path to the test set')
# Model definition
parser.add_argument('--n_mels',
type=int,
default=64,
help='Number of mel filterbanks')
parser.add_argument('--log_input',
type=bool,
default=True,
help='Log input features')
parser.add_argument('--model',
type=str,
default="ResNetSE34V2",
help='Name of model definition')
parser.add_argument('--encoder_type',
type=str,
default="ASP",
help='Type of encoder')
parser.add_argument('--nOut',
type=int,
default=512,
help='Embedding size in the last FC layer')
# For test only
parser.add_argument('--eval',
dest='eval',
action='store_true',
help='Eval only')
parser.add_argument('--test',
dest='test',
action='store_true',
help='Test only')
parser.add_argument('--prepare',
dest='prepare',
action='store_true',
help='Prepare embeddings')
parser.add_argument('-t',
'--prepare_type',
type=str,
default='cohorts',
help='embed / cohorts')
parser.add_argument('--predict',
dest='predict',
action='store_true',
help='Predict')
parser.add_argument('--cohorts_path',
type=str,
default="checkpoints/cohorts_final_500_f100.npy",
help='Cohorts path')
parser.add_argument('--test_threshold',
type=float,
default=1.7206447124481201,
help='Test threshold')
args = parser.parse_args()
if args.config is not None:
args = read_config(args.config, args)
# Load models
model = SpeakerNet(**vars(args))
model.loadParameters(args.initial_model)
model.eval()
cohorts = np.load('checkpoints/cohorts_final_500_f100.npy')
top_cohorts = 200
threshold = 1.7206447124481201
eval_frames = 100
num_eval = 10
if __name__ == '__main__':
# Evaluation code
if args.eval is True:
sc, lab, trials = model.evaluateFromList(
args.test_list,
cohorts_path=args.cohorts_path,
print_interval=100,
eval_frames=args.eval_frames)
target_fa = np.linspace(10, 0, num=50)
result = tuneThresholdfromScore(sc, lab, target_fa)
print('tfa [thre, fpr, fnr]')
best_sum_rate = 999
best_tfa = None
for i, tfa in enumerate(target_fa):
print(tfa, result[0][i])
sum_rate = result[0][i][1] + result[0][i][2]
if sum_rate < best_sum_rate:
best_sum_rate = sum_rate
best_tfa = result[0][i]
print(f'Best sum rate {best_sum_rate} at {best_tfa}')
print(f'EER {result[1]} at threshold {result[2]}')
print(f'AUC {result[3]}')
sys.exit(1)
# Test code
if args.test is True:
model.testFromList(args.test_path,
cohorts_path=args.cohorts_path,
thre_score=args.test_threshold,
print_interval=100,
eval_frames=args.eval_frames)
sys.exit(1)
# Prepare embeddings for cohorts/verification
if args.prepare is True:
model.prepare(eval_frames=args.eval_frames,
from_path=args.test_list,
save_path=args.save_path,
num_eval=num_eval,
prepare_type=args.prepare_type)
sys.exit(1)
# Predict
if args.predict is True:
"""
Predict new utterance based on distance between its embedding and saved embeddings.
"""
embeds_path = Path(args.save_path, 'embeds.pt')
classes_path = Path(args.save_path, 'classes.npy')
embeds = torch.load(embeds_path).to(torch.device(args.device))
classes = np.load(classes_path, allow_pickle=True).item()
if args.test_list.endswith('.txt'):
files = []
with open(args.test_list) as listfile:
while True:
line = listfile.readline()
if (not line):
break
data = line.split()
# Append random label if missing
if len(data) == 2:
data = [random.randint(0, 1)] + data
files.append(Path(data[1]))
files.append(Path(data[2]))
files = list(set(files))
else:
files = list(Path(args.test_list).glob('*/*.wav'))
files.sort()
same_smallest_score = 1
diff_biggest_score = 0
for f in tqdm(files):
embed = model.embed_utterance(f,
eval_frames=args.eval_frames,
num_eval=num_eval,
normalize=model.__L__.test_normalize)
embed = embed.unsqueeze(-1)
dist = F.pairwise_distance(embed, embeds).detach().cpu().numpy()
dist = np.mean(dist, axis=0)
score = 1 - np.min(dist)**2 / 2
if classes[np.argmin(dist)] == f.parent.stem:
if score < same_smallest_score:
same_smallest_score = score
indexes = np.argsort(dist)[:2]
if fabs((1 - dist[indexes[0]]**2 / 2) - (1 - dist[indexes[1]]**2 / 2)) < 0.001:
for i, idx in enumerate(indexes):
score = 1 - dist[idx]**2 / 2
if i == 0:
tqdm.write(f'+ {f}, {score} - {classes[idx]}',
end='; ')
else:
tqdm.write(f'{score} - {classes[idx]}', end='; ')
tqdm.write('***')
else:
tqdm.write(f'+ {f}, {score}', end='')
if score < args.test_threshold:
tqdm.write(' ***', end='')
tqdm.write('')
else:
if score > diff_biggest_score:
diff_biggest_score = score
if score > args.test_threshold:
indexes = np.argsort(dist)[:3]
for i, idx in enumerate(indexes):
score = 1 - dist[idx]**2 / 2
if i == 0:
tqdm.write(f'- {f}, {score} - {classes[idx]}',
end='; ')
else:
tqdm.write(f'{score} - {classes[idx]}', end='; ')
tqdm.write('***')
print(f'same_smallest_score: {same_smallest_score}')
print(f'diff_biggest_score: {diff_biggest_score}')
sys.exit(1)