-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathTDNNs.py
234 lines (187 loc) · 7.61 KB
/
TDNNs.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
#!/usr/bin/env python
# encoding: utf-8
"""
@Author: yangwenhao
@Contact: [email protected]
@Software: PyCharm
@File: TDNNs.py
@Time: 2021/10/15 09:17
@Overview:
"""
import torch
import numpy as np
import torch.nn as nn
from models.FilterLayers import Mean_Norm, AttentionweightLayer
from models.poolings import StatisticPooling
def get_activation(activation):
if activation == 'relu':
nonlinearity = nn.ReLU
elif activation in ['leakyrelu', 'leaky_relu']:
nonlinearity = nn.LeakyReLU
elif activation == 'prelu':
nonlinearity = nn.PReLU
return nonlinearity
class TimeDelayLayer(nn.Module):
def __init__(self, input_dim=23, output_dim=512, context_size=5, stride=1, dilation=1,
dropout_p=0.0, padding=0, groups=1, activation='relu'):
super(TimeDelayLayer, self).__init__()
self.context_size = context_size
self.stride = stride
self.input_dim = input_dim
self.output_dim = output_dim
self.dilation = dilation
self.dropout_p = dropout_p
self.padding = padding
self.groups = groups
self.kernel = nn.Conv1d(self.input_dim, self.output_dim, self.context_size, stride=self.stride,
padding=self.padding, dilation=self.dilation, groups=self.groups)
if activation == 'relu':
self.nonlinearity = nn.ReLU(inplace=True)
elif activation in ['leakyrelu', 'leaky_relu']:
self.nonlinearity = nn.LeakyReLU()
elif activation == 'prelu':
self.nonlinearity = nn.PReLU()
self.bn = nn.BatchNorm1d(output_dim)
def forward(self, x):
'''
input: size (batch, seq_len, input_features)
outpu: size (batch, new_seq_len, output_features)
'''
x = self.kernel(x.transpose(1, 2))
x = self.nonlinearity(x)
x = self.bn(x)
return x.transpose(1, 2)
class TDNN(nn.Module):
def __init__(self, num_classes, embedding_size, input_dim, alpha=0., input_norm='Mean',
filter=None, feat_dim=64, dropout_p=0.0, dropout_layer=False, encoder_type='STAP', activation='relu',
num_classes_b=0, block_type='basic', stride=[1],
mask='None', channels=[512, 512, 512, 512, 1500], **kwargs):
super(TDNN, self).__init__()
self.num_classes = num_classes
self.num_classes_b = num_classes_b
self.dropout_p = dropout_p
self.dropout_layer = dropout_layer
self.input_dim = input_dim
self.channels = channels
self.alpha = alpha
self.mask = mask
self.filter = filter
self.feat_dim = feat_dim
self.block_type = block_type.lower()
self.stride = stride
self.activation = activation
if len(self.stride) == 1:
while len(self.stride) < 4:
self.stride.append(self.stride[0])
if np.sum((self.stride)) > 4:
print('The stride for tdnn layers are: ', str(self.stride))
if activation != 'relu':
print('The activation function is : ', activation)
nonlinearity = get_activation(activation)
self.filter_layer = None
if input_norm == 'Instance':
self.inst_layer = nn.InstanceNorm1d(input_dim)
elif input_norm == 'Mean':
self.inst_layer = Mean_Norm()
else:
self.inst_layer = None
if self.mask == 'attention':
self.mask_layer = AttentionweightLayer(input_dim=input_dim, weight=init_weight)
else:
self.mask_layer = None
if self.filter_layer != None:
self.input_dim = feat_dim
if self.block_type == 'basic':
TDlayer = TimeDelayLayer
else:
raise ValueError(self.block_type)
self.frame1 = TimeDelayLayer(input_dim=self.input_dim, output_dim=self.channels[0],
context_size=5, stride=self.stride[0], dilation=1,
activation=self.activation, groups=first_groups)
self.frame2 = TDlayer(input_dim=self.channels[0], output_dim=self.channels[1],
context_size=3, stride=self.stride[1], dilation=2, activation=self.activation)
self.frame3 = TDlayer(input_dim=self.channels[1], output_dim=self.channels[2],
context_size=3, stride=self.stride[2], dilation=3, activation=self.activation)
self.frame4 = TDlayer(input_dim=self.channels[2], output_dim=self.channels[3],
context_size=1, stride=self.stride[0], dilation=1, activation=self.activation)
self.frame5 = TimeDelayLayer(input_dim=self.channels[3], output_dim=self.channels[4],
context_size=1, stride=self.stride[3], dilation=1,
activation=self.activation)
self.drop = nn.Dropout(p=self.dropout_p)
if encoder_type == 'STAP':
self.encoder = StatisticPooling(input_dim=self.channels[4])
self.encoder_output = self.channels[4] * 2
else:
raise ValueError(encoder_type)
self.segment6 = nn.Sequential(
nn.Linear(self.encoder_output, 512),
nonlinearity(),
nn.BatchNorm1d(512)
)
self.segment7 = nn.Sequential(
nn.Linear(512, embedding_size),
nonlinearity(),
nn.BatchNorm1d(embedding_size)
)
if num_classes > 0:
self.classifier = nn.Linear(embedding_size, num_classes)
else:
print("Set not classifier in xvectors model!")
self.classifier = None
# self.bn = nn.BatchNorm1d(num_classes)
for m in self.modules(): # 对于各层参数的初始化
if isinstance(m, nn.BatchNorm1d): # weight设置为1，bias为0
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, TimeDelayLayer):
# nn.init.normal(m.kernel.weight, mean=0., std=1.)
nn.init.kaiming_normal_(m.kernel.weight, mode='fan_out', nonlinearity='relu')
def set_global_dropout(self, dropout_p):
self.dropout_p = dropout_p
self.drop.p = dropout_p
def forward(self, x):
# pdb.set_trace()
# print(x.shape)
if len(x.shape) == 4:
x = x.squeeze(1).float()
if self.inst_layer != None:
x = self.inst_layer(x)
if self.mask_layer != None:
x = self.mask_layer(x)
x = self.frame1(x)
x = self.frame2(x)
x = self.frame3(x)
x = self.frame4(x)
x = self.frame5(x)
if self.dropout_layer:
x = self.drop(x)
# print(x.shape)
x = self.encoder(x)
embedding_a = self.segment6(x)
embedding_b = self.segment7(embedding_a)
if self.classifier == None:
logits = ""
else:
logits = self.classifier(embedding_b)
return logits, embedding_b
def xvector(self, x):
# pdb.set_trace()
if self.filter_layer != None:
x = self.filter_layer(x)
if len(x.shape) == 4:
x = x.squeeze(1).float()
if self.inst_layer != None:
x = self.inst_layer(x)
if self.mask_layer != None:
x = self.mask_layer(x)
x = self.frame1(x)
x = self.frame2(x)
x = self.frame3(x)
x = self.frame4(x)
x = self.frame5(x)
if self.dropout_layer:
x = self.drop(x)
# print(x.shape)
x = self.encoder(x)
embedding_a = self.segment6[0](x)
return "", embedding_a