File size: 2,968 Bytes
ed1cdd1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# -*- coding: utf-8 -*-

# Copyright 2020 Tomoki Hayashi
#  MIT License (https://opensource.org/licenses/MIT)

"""Residual stack module in MelGAN."""

import torch

from . import CausalConv1d


class ResidualStack(torch.nn.Module):
    """Residual stack module introduced in MelGAN."""

    def __init__(self,
                 kernel_size=3,
                 channels=32,
                 dilation=1,
                 bias=True,
                 nonlinear_activation="LeakyReLU",
                 nonlinear_activation_params={"negative_slope": 0.2},
                 pad="ReflectionPad1d",
                 pad_params={},
                 use_causal_conv=False,
                 ):
        """Initialize ResidualStack module.

        Args:
            kernel_size (int): Kernel size of dilation convolution layer.
            channels (int): Number of channels of convolution layers.
            dilation (int): Dilation factor.
            bias (bool): Whether to add bias parameter in convolution layers.
            nonlinear_activation (str): Activation function module name.
            nonlinear_activation_params (dict): Hyperparameters for activation function.
            pad (str): Padding function module name before dilated convolution layer.
            pad_params (dict): Hyperparameters for padding function.
            use_causal_conv (bool): Whether to use causal convolution.

        """
        super(ResidualStack, self).__init__()

        # defile residual stack part
        if not use_causal_conv:
            assert (kernel_size - 1) % 2 == 0, "Not support even number kernel size."
            self.stack = torch.nn.Sequential(
                getattr(torch.nn, nonlinear_activation)(**nonlinear_activation_params),
                getattr(torch.nn, pad)((kernel_size - 1) // 2 * dilation, **pad_params),
                torch.nn.Conv1d(channels, channels, kernel_size, dilation=dilation, bias=bias),
                getattr(torch.nn, nonlinear_activation)(**nonlinear_activation_params),
                torch.nn.Conv1d(channels, channels, 1, bias=bias),
            )
        else:
            self.stack = torch.nn.Sequential(
                getattr(torch.nn, nonlinear_activation)(**nonlinear_activation_params),
                CausalConv1d(channels, channels, kernel_size, dilation=dilation,
                             bias=bias, pad=pad, pad_params=pad_params),
                getattr(torch.nn, nonlinear_activation)(**nonlinear_activation_params),
                torch.nn.Conv1d(channels, channels, 1, bias=bias),
            )

        # defile extra layer for skip connection
        self.skip_layer = torch.nn.Conv1d(channels, channels, 1, bias=bias)

    def forward(self, c):
        """Calculate forward propagation.

        Args:
            c (Tensor): Input tensor (B, channels, T).

        Returns:
            Tensor: Output tensor (B, chennels, T).

        """
        return self.stack(c) + self.skip_layer(c)