forked from ghua-ac/end-to-end-synthetic-speech-detection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
133 lines (105 loc) · 4.81 KB
/
test.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
import torch
from torch.utils.data.dataloader import DataLoader
from data import PrepASV15Dataset, PrepASV19Dataset
import models
import torch.nn.functional as F
import matplotlib.pyplot as plt
def asv_cal_accuracies(protocol, path_data, net, device, data_type='time_frame', dataset=19):
net = net.to(device)
net.eval()
with torch.no_grad():
softmax_acc = 0
num_files = 0
probs = torch.empty(0, 3).to(device)
if dataset == 15:
test_set = PrepASV15Dataset(protocol, path_data, data_type=data_type)
else:
test_set = PrepASV19Dataset(protocol, path_data, data_type=data_type)
test_loader = DataLoader(test_set, batch_size=64, shuffle=False, num_workers=4)
for test_batch in test_loader:
# load batch and infer
test_sample, test_label, sub_class = test_batch
# # sub_class level test, comment if unwanted
# # train & dev 0~6; eval 7~19
# # selected_index = torch.nonzero(torch.logical_xor(sub_class == 10, sub_class == 0))[:, 0]
# selected_index = torch.nonzero(sub_class.ne(10))[:, 0]
# if len(selected_index) == 0:
# continue
# test_sample = test_sample[selected_index, :, :]
# test_label = test_label[selected_index]
num_files += len(test_label)
test_sample = test_sample.to(device)
test_label = test_label.to(device)
infer = net(test_sample)
# obtain output probabilities
t1 = F.softmax(infer, dim=1)
t2 = test_label.unsqueeze(-1)
row = torch.cat((t1, t2), dim=1)
probs = torch.cat((probs, row), dim=0)
# calculate example level accuracy
infer = infer.argmax(dim=1)
batch_acc = infer.eq(test_label).sum().item()
softmax_acc += batch_acc
softmax_acc = softmax_acc / num_files
return softmax_acc, probs.to('cpu')
def cal_roc_eer(probs, show_plot=True):
"""
probs: tensor, number of samples * 3, containing softmax probabilities
row wise: [genuine prob, fake prob, label]
TP: True Fake
FP: False Fake
"""
all_labels = probs[:, 2]
zero_index = torch.nonzero((all_labels == 0)).squeeze(-1)
one_index = torch.nonzero(all_labels).squeeze(-1)
zero_probs = probs[zero_index, 0]
one_probs = probs[one_index, 0]
threshold_index = torch.linspace(-0.1, 1.01, 10000)
tpr = torch.zeros(len(threshold_index),)
fpr = torch.zeros(len(threshold_index),)
cnt = 0
for i in threshold_index:
tpr[cnt] = one_probs.le(i).sum().item()/len(one_probs)
fpr[cnt] = zero_probs.le(i).sum().item()/len(zero_probs)
cnt += 1
sum_rate = tpr + fpr
distance_to_one = torch.abs(sum_rate - 1)
eer_index = distance_to_one.argmin(dim=0).item()
out_eer = 0.5*(fpr[eer_index] + 1 - tpr[eer_index]).numpy()
if show_plot:
print('EER: {:.4f}%.'.format(out_eer * 100))
plt.figure(1)
plt.plot(torch.linspace(-0.2, 1.2, 1000), torch.histc(zero_probs, bins=1000, min=-0.2, max=1.2) / len(zero_probs))
plt.plot(torch.linspace(-0.2, 1.2, 1000), torch.histc(one_probs, bins=1000, min=-0.2, max=1.2) / len(one_probs))
plt.xlabel("Probability of 'Genuine'")
plt.ylabel('Per Class Ratio')
plt.legend(['Real', 'Fake'])
plt.grid()
plt.figure(3)
plt.scatter(fpr, tpr)
plt.xlabel('False Positive (Fake) Rate')
plt.ylabel('True Positive (Fake) Rate')
plt.grid()
plt.show()
return out_eer
if __name__ == '__main__':
test_device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
protocol_file_path = 'F:/ASVSpoof2019/LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.dev.trl.txt'
data_path = 'F:/ASVSpoof2019/LA/data/dev_6/'
# 'F:/ASVSpoof2019/LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.train.trn.txt'
# 'F:/ASVSpoof2019/LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.dev.trl.txt'
# 'F:/ASVSpoof2019/LA/ASVspoof2019_LA_cm_protocols/ASVspoof2019.LA.cm.eval.trl.txt'
# protocol_file_path = 'F:/ASVspoof2015/CM_protocol/cm_develop.ndx.txt'
# # cm_train.trn
# # cm_develop.ndx
# # cm_evaluation.ndx
# data_path = 'F:/ASVspoof2015/data/dev_6/'
Net = models.SSDNet1D()
num_total_learnable_params = sum(i.numel() for i in Net.parameters() if i.requires_grad)
print('Number of learnable params: {}.'.format(num_total_learnable_params))
check_point = torch.load('./trained_models/***.pth')
Net.load_state_dict(check_point['model_state_dict'])
accuracy, probabilities = asv_cal_accuracies(protocol_file_path, data_path, Net, test_device, data_type='time_frame', dataset=19)
print(accuracy * 100)
eer = cal_roc_eer(probabilities)
print('End of Program.')