File size: 6,552 Bytes
9b2107c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
from torch import nn

# from TTS.utils.audio.torch_transforms import TorchSTFT
from TTS.encoder.models.base_encoder import BaseEncoder


class SELayer(nn.Module):
    def __init__(self, channel, reduction=8):
        super(SELayer, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Linear(channel, channel // reduction),
            nn.ReLU(inplace=True),
            nn.Linear(channel // reduction, channel),
            nn.Sigmoid(),
        )

    def forward(self, x):
        b, c, _, _ = x.size()
        y = self.avg_pool(x).view(b, c)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y


class SEBasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None, reduction=8):
        super(SEBasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.se = SELayer(planes, reduction)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.relu(out)
        out = self.bn1(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.se(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)
        return out


class ResNetSpeakerEncoder(BaseEncoder):
    """Implementation of the model H/ASP without batch normalization in speaker embedding. This model was proposed in: https://arxiv.org/abs/2009.14153
    Adapted from: https://github.com/clovaai/voxceleb_trainer
    """

    # pylint: disable=W0102
    def __init__(
        self,
        input_dim=64,
        proj_dim=512,
        layers=[3, 4, 6, 3],
        num_filters=[32, 64, 128, 256],
        encoder_type="ASP",
        log_input=False,
        use_torch_spec=False,
        audio_config=None,
    ):
        super(ResNetSpeakerEncoder, self).__init__()

        self.encoder_type = encoder_type
        self.input_dim = input_dim
        self.log_input = log_input
        self.use_torch_spec = use_torch_spec
        self.audio_config = audio_config
        self.proj_dim = proj_dim

        self.conv1 = nn.Conv2d(1, num_filters[0], kernel_size=3, stride=1, padding=1)
        self.relu = nn.ReLU(inplace=True)
        self.bn1 = nn.BatchNorm2d(num_filters[0])

        self.inplanes = num_filters[0]
        self.layer1 = self.create_layer(SEBasicBlock, num_filters[0], layers[0])
        self.layer2 = self.create_layer(SEBasicBlock, num_filters[1], layers[1], stride=(2, 2))
        self.layer3 = self.create_layer(SEBasicBlock, num_filters[2], layers[2], stride=(2, 2))
        self.layer4 = self.create_layer(SEBasicBlock, num_filters[3], layers[3], stride=(2, 2))

        self.instancenorm = nn.InstanceNorm1d(input_dim)

        if self.use_torch_spec:
            self.torch_spec = self.get_torch_mel_spectrogram_class(audio_config)
        else:
            self.torch_spec = None

        outmap_size = int(self.input_dim / 8)

        self.attention = nn.Sequential(
            nn.Conv1d(num_filters[3] * outmap_size, 128, kernel_size=1),
            nn.ReLU(),
            nn.BatchNorm1d(128),
            nn.Conv1d(128, num_filters[3] * outmap_size, kernel_size=1),
            nn.Softmax(dim=2),
        )

        if self.encoder_type == "SAP":
            out_dim = num_filters[3] * outmap_size
        elif self.encoder_type == "ASP":
            out_dim = num_filters[3] * outmap_size * 2
        else:
            raise ValueError("Undefined encoder")

        self.fc = nn.Linear(out_dim, proj_dim)

        self._init_layers()

    def _init_layers(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

    def create_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    # pylint: disable=R0201
    def new_parameter(self, *size):
        out = nn.Parameter(torch.FloatTensor(*size))
        nn.init.xavier_normal_(out)
        return out

    def forward(self, x, l2_norm=False):
        """Forward pass of the model.

        Args:
            x (Tensor): Raw waveform signal or spectrogram frames. If input is a waveform, `torch_spec` must be `True`
                to compute the spectrogram on-the-fly.
            l2_norm (bool): Whether to L2-normalize the outputs.

        Shapes:
            - x: :math:`(N, 1, T_{in})` or :math:`(N, D_{spec}, T_{in})`
        """
        x.squeeze_(1)
        # if you torch spec compute it otherwise use the mel spec computed by the AP
        if self.use_torch_spec:
            x = self.torch_spec(x)

        if self.log_input:
            x = (x + 1e-6).log()
        x = self.instancenorm(x).unsqueeze(1)

        x = self.conv1(x)
        x = self.relu(x)
        x = self.bn1(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = x.reshape(x.size()[0], -1, x.size()[-1])

        w = self.attention(x)

        if self.encoder_type == "SAP":
            x = torch.sum(x * w, dim=2)
        elif self.encoder_type == "ASP":
            mu = torch.sum(x * w, dim=2)
            sg = torch.sqrt((torch.sum((x**2) * w, dim=2) - mu**2).clamp(min=1e-5))
            x = torch.cat((mu, sg), 1)

        x = x.view(x.size()[0], -1)
        x = self.fc(x)

        if l2_norm:
            x = torch.nn.functional.normalize(x, p=2, dim=1)
        return x