Spaces:
Build error
Build error
File size: 4,408 Bytes
22871e7 |
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 |
import torch
import torch.nn as nn
class BasicDiscriminatorBlock(nn.Module):
def __init__(self, in_channel, out_channel):
super(BasicDiscriminatorBlock, self).__init__()
self.block = nn.Sequential(
nn.utils.weight_norm(nn.Conv1d(
in_channel,
out_channel,
kernel_size=3,
stride=2,
padding=1,
)),
nn.LeakyReLU(0.2, True),
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
nn.LeakyReLU(0.2, True),
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
nn.LeakyReLU(0.2, True),
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
)
def forward(self, x):
return self.block(x)
class ResDiscriminatorBlock(nn.Module):
def __init__(self, in_channel, out_channel):
super(ResDiscriminatorBlock, self).__init__()
self.block1 = nn.Sequential(
nn.utils.weight_norm(nn.Conv1d(
in_channel,
out_channel,
kernel_size=3,
stride=2,
padding=1,
)),
nn.LeakyReLU(0.2, True),
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
)
self.shortcut1 = nn.utils.weight_norm(nn.Conv1d(
in_channel,
out_channel,
kernel_size=1,
stride=2,
))
self.block2 = nn.Sequential(
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
nn.LeakyReLU(0.2, True),
nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=3,
stride=1,
padding=1,
)),
)
self.shortcut2 = nn.utils.weight_norm(nn.Conv1d(
out_channel,
out_channel,
kernel_size=1,
stride=1,
))
def forward(self, x):
x1 = self.block1(x)
x1 = x1 + self.shortcut1(x)
return self.block2(x1) + self.shortcut2(x1)
class ResNet18Discriminator(nn.Module):
def __init__(self, stft_channel, in_channel=64):
super(ResNet18Discriminator, self).__init__()
self.input = nn.Sequential(
nn.utils.weight_norm(nn.Conv1d(stft_channel, in_channel, kernel_size=7, stride=2, padding=1, )),
nn.LeakyReLU(0.2, True),
)
self.df1 = BasicDiscriminatorBlock(in_channel, in_channel)
self.df2 = ResDiscriminatorBlock(in_channel, in_channel * 2)
self.df3 = ResDiscriminatorBlock(in_channel * 2, in_channel * 4)
self.df4 = ResDiscriminatorBlock(in_channel * 4, in_channel * 8)
def forward(self, x):
x = self.input(x)
x = self.df1(x)
x = self.df2(x)
x = self.df3(x)
return self.df4(x)
class FrequencyDiscriminator(nn.Module):
def __init__(self, in_channel=64, fft_size=1024, hop_length=256, win_length=1024, window="hann_window"):
super(FrequencyDiscriminator, self).__init__()
self.fft_size = fft_size
self.hop_length = hop_length
self.win_length = win_length
self.window = nn.Parameter(getattr(torch, window)(win_length), requires_grad=False)
self.stft_channel = fft_size // 2 + 1
self.resnet_disc = ResNet18Discriminator(self.stft_channel, in_channel)
def forward(self, x):
x_stft = torch.stft(x, self.fft_size, self.hop_length, self.win_length, self.window)
real = x_stft[..., 0]
imag = x_stft[..., 1]
x_real = self.resnet_disc(real)
x_imag = self.resnet_disc(imag)
return x_real, x_imag
|