/
model_vc.py
executable file
·246 lines (204 loc) · 9.82 KB
/
model_vc.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
import torch, pdb
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
class LinearNorm(torch.nn.Module):
def __init__(self, in_dim, out_dim, bias=True, w_init_gain='linear'):
super(LinearNorm, self).__init__()
self.linear_layer = torch.nn.Linear(in_dim, out_dim, bias=bias)
torch.nn.init.xavier_uniform_(
self.linear_layer.weight,
gain=torch.nn.init.calculate_gain(w_init_gain))
def forward(self, x):
return self.linear_layer(x)
class ConvNorm1d(torch.nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=1, stride=1,
padding=None, dilation=1, bias=True, w_init_gain='linear'):
super(ConvNorm1d, self).__init__()
if padding is None:
assert(kernel_size % 2 == 1)
padding = int(dilation * (kernel_size - 1) / 2)
self.conv = torch.nn.Conv1d(in_channels, out_channels,
kernel_size=kernel_size, stride=stride,
padding=padding, dilation=dilation,
bias=bias)
torch.nn.init.xavier_uniform_(
self.conv.weight, gain=torch.nn.init.calculate_gain(w_init_gain))
def forward(self, signal):
conv_signal = self.conv(signal)
return conv_signal
class ConvT2d(torch.nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=1, stride=1,
padding=None, dilation=1, bias=True, w_init_gain='relu'):
super(ConvT2d, self).__init__()
if padding is None:
assert(kernel_size % 2 == 1)
padding = int(dilation * (kernel_size - 1) / 2)
self.conv = torch.nn.ConvTranspose2d(in_channels, out_channels,
kernel_size=kernel_size, stride=stride,
padding=padding, dilation=dilation,
bias=bias)
torch.nn.init.xavier_uniform_(
self.conv.weight, gain=torch.nn.init.calculate_gain(w_init_gain)),
def forward(self, signal):
conv_signal = self.conv(signal)
return conv_signal
# "4.2. The Content Encoder"
class Encoder(nn.Module):
"""Encoder module:
"""
def __init__(self, dim_neck, dim_emb, freq):
super(Encoder, self).__init__()
self.dim_neck = dim_neck
self.freq = freq
convolutions = []
for i in range(3):
# "the input to the content encoder is the 80-dimensional mel-spectrogram of X1 concatenated with the speaker embedding" - I think the embeddings are copy pasted from a dataset, as the Speaker Decoder is pretrained and may not actually appear in this implementation?
conv_layer = nn.Sequential(
# "the input to the content encoder is the 80-dimensional mel-spectrogram of X1 concatenated with the speaker embedding. The concatenated features are fed into three 5 × 1 convolutional layers, each followed by batch normalization and ReLU activation. The number of channels i
ConvNorm1d(80+dim_emb if i==0 else 512,
512,
kernel_size=5, stride=1,
padding=2,
dilation=1, w_init_gain='relu'),
nn.BatchNorm1d(512))
convolutions.append(conv_layer)
self.convolutions = nn.ModuleList(convolutions)
# "Both the forward and backward cell dimensions are 32, so their (LSTMs) combined dimension is 64."
self.lstm = nn.LSTM(512, dim_neck, 2, batch_first=True, bidirectional=True)
# c_org is speaker embedding
def forward(self, x, c_org):
x = x.squeeze(1).transpose(2,1)
# broadcasts c_org to a compatible shape to merge with x
c_org = c_org.unsqueeze(-1).expand(-1, -1, x.size(-1))
x = torch.cat((x, c_org), dim=1)
saved_enc_outs = [x] ###
for conv in self.convolutions:
x = F.relu(conv(x))
saved_enc_outs.append(x) ###
x = x.transpose(1, 2)
self.lstm.flatten_parameters()
# lstms output 64 dim
outputs, _ = self.lstm(x)
saved_enc_outs.append(outputs.transpose(2,1)) ###
# backward is the first half of dimensions, forward is the second half
out_forward = outputs[:, :, :self.dim_neck]
out_backward = outputs[:, :, self.dim_neck:]
pdb.set_trace()
codes = []
# for each timestep, skipping self.freq frames
for i in range(0, outputs.size(1), self.freq):
# remeber that i is self.freq, not increments of 1)
codes.append(torch.cat((out_forward[:,i+self.freq-1,:],out_backward[:,i,:]), dim=-1))
#saved_enc_outs.append(codes_cat) ###
# if self.freq is 32, then codes is a list of 4 tensors of size 64
return codes, saved_enc_outs
class Decoder(nn.Module):
"""Decoder module:
"""
def __init__(self, dim_neck, dim_emb, dim_pre):
super(Decoder, self).__init__()
self.lstm1 = nn.LSTM(dim_neck*2+dim_emb, dim_pre, 1, batch_first=True)
convolutions = []
for i in range(3):
conv_layer = nn.Sequential(
ConvNorm1d(dim_pre,
dim_pre,
kernel_size=5, stride=1,
padding=2,
dilation=1, w_init_gain='relu'),
nn.BatchNorm1d(dim_pre))
convolutions.append(conv_layer)
self.convolutions = nn.ModuleList(convolutions)
self.lstm2 = nn.LSTM(dim_pre, 1024, 2, batch_first=True)
self.linear_projection = LinearNorm(1024, 80)
def forward(self, x):
#self.lstm1.flatten_parameters()
# pdb.set_trace()
saved_dec_outs = [x.transpose(1,2)] ###
x, _ = self.lstm1(x)
saved_dec_outs.append(x.transpose(1,2)) ###
x = x.transpose(1, 2)
for conv in self.convolutions:
x = F.relu(conv(x))
saved_dec_outs.append(x) ###
x = x.transpose(1, 2)
outputs, _ = self.lstm2(x)
saved_dec_outs.append(outputs.transpose(1,2)) ###
decoder_output = self.linear_projection(outputs)
saved_dec_outs.append(decoder_output.transpose(1,2)) ###
return decoder_output, saved_dec_outs
# Still part of Decoder as indicated in paper Fig. 3 (c) - last two blocks
class Postnet(nn.Module):
"""Postnet
- Five 1-d convolution with 512 channels and kernel size 5
"""
def __init__(self):
super(Postnet, self).__init__()
self.convolutions = nn.ModuleList()
self.convolutions.append(
nn.Sequential(
ConvNorm1d(80, 512,
kernel_size=5, stride=1,
padding=2,
dilation=1, w_init_gain='tanh'),
nn.BatchNorm1d(512))
)
for i in range(1, 5 - 1):
self.convolutions.append(
nn.Sequential(
ConvNorm1d(512,
512,
kernel_size=5, stride=1,
padding=2,
dilation=1, w_init_gain='tanh'),
nn.BatchNorm1d(512))
)
self.convolutions.append(
nn.Sequential(
ConvNorm1d(512, 80,
kernel_size=5, stride=1,
padding=2,
dilation=1, w_init_gain='linear'),
nn.BatchNorm1d(80))
)
def forward(self, x):
# pdb.set_trace()
for i in range(len(self.convolutions) - 1):
x = torch.tanh(self.convolutions[i](x))
x = self.convolutions[-1](x)
return x
class Generator(nn.Module):
"""Generator network."""
def __init__(self, dim_neck, dim_emb, dim_pre, freq):
super(Generator, self).__init__()
self.encoder = Encoder(dim_neck, dim_emb, freq)
self.decoder = Decoder(dim_neck, dim_emb, dim_pre)
self.postnet = Postnet()
def forward(self, x, c_org, c_trg):
# codes is a LIST of tensors
codes, saved_enc_outs = self.encoder(x, c_org)
# if no c_trg given, then just return the formatted encoder codes
if c_trg is None:
# concatenates the by stacking over the last (in 2D this would be vertical) dimensio by stacking over the last (in 2D this would be vertical) dimension. For lists it means the same
return torch.cat(codes, dim=-1)
# list of reformatted codes
tmp = []
for code in codes:
# reformatting tmp from list to tensor, and resample it at the specified freq
tmp.append(code.unsqueeze(1).expand(-1,int(x.size(1)/len(codes)),-1))
code_exp = torch.cat(tmp, dim=1)
pdb.set_trace()
# concat reformated encoder output with target speaker embedding
encoder_outputs = torch.cat((code_exp, c_trg.unsqueeze(1).expand(-1,x.size(1),-1)), dim=-1)
mel_outputs, saved_dec_outs = self.decoder(encoder_outputs)
# then put mel_ouputs through remaining postnet section of NN
# the postnet process produces the RESIDUAL information that gets added to the mel output
mel_outputs_postnet = self.postnet(mel_outputs.transpose(2,1))
#pdb.set_trace()
# add together, as done in Fig. 3 (c) ensuring the mel_out_psnt is same shape (2,128,80). new mel_out_psnt will be the same
mel_outputs_postnet = mel_outputs + mel_outputs_postnet.transpose(2,1)
#insert channel dimension into tensors to become (2,1,128,80)
mel_outputs = mel_outputs.unsqueeze(1)
mel_outputs_postnet = mel_outputs_postnet.unsqueeze(1)
return mel_outputs, mel_outputs_postnet, torch.cat(codes, dim=-1), saved_enc_outs, saved_dec_outs