File size: 5,624 Bytes
8a06b33 | 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 | import math
import torch
from typing import Optional
from packaging import version
is_pytorch2_1 = version.parse(torch.__version__) >= version.parse("2.1.0")
from .commons import sequence_mask
from .modules import WaveNet
from .normalization import LayerNorm
from .attentions import FFN, MultiHeadAttention
class Encoder(torch.nn.Module):
def __init__(
self,
hidden_channels: int,
filter_channels: int,
n_heads: int,
n_layers: int,
kernel_size: int = 1,
p_dropout: float = 0.0,
window_size: int = 10,
):
super().__init__()
self.hidden_channels = hidden_channels
self.n_layers = n_layers
self.drop = torch.nn.Dropout(p_dropout)
self.attn_layers = torch.nn.ModuleList(
[
MultiHeadAttention(
hidden_channels,
hidden_channels,
n_heads,
p_dropout=p_dropout,
window_size=window_size,
)
for _ in range(n_layers)
]
)
self.norm_layers_1 = torch.nn.ModuleList(
[LayerNorm(hidden_channels) for _ in range(n_layers)]
)
self.ffn_layers = torch.nn.ModuleList(
[
FFN(
hidden_channels,
hidden_channels,
filter_channels,
kernel_size,
p_dropout=p_dropout,
)
for _ in range(n_layers)
]
)
self.norm_layers_2 = torch.nn.ModuleList(
[LayerNorm(hidden_channels) for _ in range(n_layers)]
)
def forward(self, x, x_mask):
attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
x = x * x_mask
for i in range(self.n_layers):
y = self.attn_layers[i](x, x, attn_mask)
y = self.drop(y)
x = self.norm_layers_1[i](x + y)
y = self.ffn_layers[i](x, x_mask)
y = self.drop(y)
x = self.norm_layers_2[i](x + y)
return x * x_mask
class TextEncoder(torch.nn.Module):
def __init__(
self,
out_channels: int,
hidden_channels: int,
filter_channels: int,
n_heads: int,
n_layers: int,
kernel_size: int,
p_dropout: float,
embedding_dim: int,
f0: bool = True,
):
super().__init__()
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.emb_phone = torch.nn.Linear(embedding_dim, hidden_channels)
self.lrelu = torch.nn.LeakyReLU(0.1, inplace=True)
self.emb_pitch = torch.nn.Embedding(256, hidden_channels) if f0 else None
self.encoder = Encoder(
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
)
self.proj = torch.nn.Conv1d(hidden_channels, out_channels * 2, 1)
def forward(
self, phone: torch.Tensor, pitch: Optional[torch.Tensor], lengths: torch.Tensor
):
x = self.emb_phone(phone)
if pitch is not None and self.emb_pitch:
x += self.emb_pitch(pitch)
x *= math.sqrt(self.hidden_channels)
x = self.lrelu(x)
x = x.transpose(1, -1)
x_mask = sequence_mask(lengths, x.size(2)).unsqueeze(1).to(x.dtype)
x = self.encoder(x, x_mask)
stats = self.proj(x) * x_mask
m, logs = torch.split(stats, self.out_channels, dim=1)
return m, logs, x_mask
class PosteriorEncoder(torch.nn.Module):
def __init__(
self,
in_channels: int,
out_channels: int,
hidden_channels: int,
kernel_size: int,
dilation_rate: int,
n_layers: int,
gin_channels: int = 0,
):
super().__init__()
self.out_channels = out_channels
self.pre = torch.nn.Conv1d(in_channels, hidden_channels, 1)
self.enc = WaveNet(
hidden_channels,
kernel_size,
dilation_rate,
n_layers,
gin_channels=gin_channels,
)
self.proj = torch.nn.Conv1d(hidden_channels, out_channels * 2, 1)
def forward(
self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None
):
x_mask = sequence_mask(x_lengths, x.size(2)).unsqueeze(1).to(x.dtype)
x = self.pre(x) * x_mask
x = self.enc(x, x_mask, g=g)
stats = self.proj(x) * x_mask
m, logs = torch.split(stats, self.out_channels, dim=1)
z = m + torch.randn_like(m) * torch.exp(logs)
z *= x_mask
return z, m, logs, x_mask
def remove_weight_norm(self):
self.enc.remove_weight_norm()
def __prepare_scriptable__(self):
for hook in self.enc._forward_pre_hooks.values():
if is_pytorch2_1:
if (
hook.__module__ == "torch.nn.utils.parametrizations.weight_norm"
and hook.__class__.__name__ == "WeightNorm"
):
torch.nn.utils.remove_weight_norm(self.enc)
else:
if (
hook.__module__ == "torch.nn.utils.weight_norm"
and hook.__class__.__name__ == "WeightNorm"
):
torch.nn.utils.remove_weight_norm(self.enc)
return self
|