Spaces:
Runtime error
Runtime error
File size: 14,237 Bytes
11120b4 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 |
""" from https://github.com/jaywalnut310/glow-tts """
import math
import torch
from model.base import BaseModule
from model.utils import sequence_mask, convert_pad_shape
class LayerNorm(BaseModule):
def __init__(self, channels, eps=1e-4):
super(LayerNorm, self).__init__()
self.channels = channels
self.eps = eps
self.gamma = torch.nn.Parameter(torch.ones(channels))
self.beta = torch.nn.Parameter(torch.zeros(channels))
def forward(self, x):
n_dims = len(x.shape)
mean = torch.mean(x, 1, keepdim=True)
variance = torch.mean((x - mean)**2, 1, keepdim=True)
x = (x - mean) * torch.rsqrt(variance + self.eps)
shape = [1, -1] + [1] * (n_dims - 2)
x = x * self.gamma.view(*shape) + self.beta.view(*shape)
return x
class ConvReluNorm(BaseModule):
def __init__(self, in_channels, hidden_channels, out_channels, kernel_size,
n_layers, p_dropout):
super(ConvReluNorm, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.kernel_size = kernel_size
self.n_layers = n_layers
self.p_dropout = p_dropout
self.conv_layers = torch.nn.ModuleList()
self.norm_layers = torch.nn.ModuleList()
self.conv_layers.append(torch.nn.Conv1d(in_channels, hidden_channels,
kernel_size, padding=kernel_size//2))
self.norm_layers.append(LayerNorm(hidden_channels))
self.relu_drop = torch.nn.Sequential(torch.nn.ReLU(), torch.nn.Dropout(p_dropout))
for _ in range(n_layers - 1):
self.conv_layers.append(torch.nn.Conv1d(hidden_channels, hidden_channels,
kernel_size, padding=kernel_size//2))
self.norm_layers.append(LayerNorm(hidden_channels))
self.proj = torch.nn.Conv1d(hidden_channels, out_channels, 1)
self.proj.weight.data.zero_()
self.proj.bias.data.zero_()
def forward(self, x, x_mask):
x_org = x
for i in range(self.n_layers):
x = self.conv_layers[i](x * x_mask)
x = self.norm_layers[i](x)
x = self.relu_drop(x)
x = x_org + self.proj(x)
return x * x_mask
class DurationPredictor(BaseModule):
def __init__(self, in_channels, filter_channels, kernel_size, p_dropout):
super(DurationPredictor, self).__init__()
self.in_channels = in_channels
self.filter_channels = filter_channels
self.p_dropout = p_dropout
self.drop = torch.nn.Dropout(p_dropout)
self.conv_1 = torch.nn.Conv1d(in_channels, filter_channels,
kernel_size, padding=kernel_size//2)
self.norm_1 = LayerNorm(filter_channels)
self.conv_2 = torch.nn.Conv1d(filter_channels, filter_channels,
kernel_size, padding=kernel_size//2)
self.norm_2 = LayerNorm(filter_channels)
self.proj = torch.nn.Conv1d(filter_channels, 1, 1)
def forward(self, x, x_mask):
x = self.conv_1(x * x_mask)
x = torch.relu(x)
x = self.norm_1(x)
x = self.drop(x)
x = self.conv_2(x * x_mask)
x = torch.relu(x)
x = self.norm_2(x)
x = self.drop(x)
x = self.proj(x * x_mask)
return x * x_mask
class MultiHeadAttention(BaseModule):
def __init__(self, channels, out_channels, n_heads, window_size=None,
heads_share=True, p_dropout=0.0, proximal_bias=False,
proximal_init=False):
super(MultiHeadAttention, self).__init__()
assert channels % n_heads == 0
self.channels = channels
self.out_channels = out_channels
self.n_heads = n_heads
self.window_size = window_size
self.heads_share = heads_share
self.proximal_bias = proximal_bias
self.p_dropout = p_dropout
self.attn = None
self.k_channels = channels // n_heads
self.conv_q = torch.nn.Conv1d(channels, channels, 1)
self.conv_k = torch.nn.Conv1d(channels, channels, 1)
self.conv_v = torch.nn.Conv1d(channels, channels, 1)
if window_size is not None:
n_heads_rel = 1 if heads_share else n_heads
rel_stddev = self.k_channels**-0.5
self.emb_rel_k = torch.nn.Parameter(torch.randn(n_heads_rel,
window_size * 2 + 1, self.k_channels) * rel_stddev)
self.emb_rel_v = torch.nn.Parameter(torch.randn(n_heads_rel,
window_size * 2 + 1, self.k_channels) * rel_stddev)
self.conv_o = torch.nn.Conv1d(channels, out_channels, 1)
self.drop = torch.nn.Dropout(p_dropout)
torch.nn.init.xavier_uniform_(self.conv_q.weight)
torch.nn.init.xavier_uniform_(self.conv_k.weight)
if proximal_init:
self.conv_k.weight.data.copy_(self.conv_q.weight.data)
self.conv_k.bias.data.copy_(self.conv_q.bias.data)
torch.nn.init.xavier_uniform_(self.conv_v.weight)
def forward(self, x, c, attn_mask=None):
q = self.conv_q(x)
k = self.conv_k(c)
v = self.conv_v(c)
x, self.attn = self.attention(q, k, v, mask=attn_mask)
x = self.conv_o(x)
return x
def attention(self, query, key, value, mask=None):
b, d, t_s, t_t = (*key.size(), query.size(2))
query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(self.k_channels)
if self.window_size is not None:
assert t_s == t_t, "Relative attention is only available for self-attention."
key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
rel_logits = self._matmul_with_relative_keys(query, key_relative_embeddings)
rel_logits = self._relative_position_to_absolute_position(rel_logits)
scores_local = rel_logits / math.sqrt(self.k_channels)
scores = scores + scores_local
if self.proximal_bias:
assert t_s == t_t, "Proximal bias is only available for self-attention."
scores = scores + self._attention_bias_proximal(t_s).to(device=scores.device,
dtype=scores.dtype)
if mask is not None:
scores = scores.masked_fill(mask == 0, -1e4)
p_attn = torch.nn.functional.softmax(scores, dim=-1)
p_attn = self.drop(p_attn)
output = torch.matmul(p_attn, value)
if self.window_size is not None:
relative_weights = self._absolute_position_to_relative_position(p_attn)
value_relative_embeddings = self._get_relative_embeddings(self.emb_rel_v, t_s)
output = output + self._matmul_with_relative_values(relative_weights,
value_relative_embeddings)
output = output.transpose(2, 3).contiguous().view(b, d, t_t)
return output, p_attn
def _matmul_with_relative_values(self, x, y):
ret = torch.matmul(x, y.unsqueeze(0))
return ret
def _matmul_with_relative_keys(self, x, y):
ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
return ret
def _get_relative_embeddings(self, relative_embeddings, length):
pad_length = max(length - (self.window_size + 1), 0)
slice_start_position = max((self.window_size + 1) - length, 0)
slice_end_position = slice_start_position + 2 * length - 1
if pad_length > 0:
padded_relative_embeddings = torch.nn.functional.pad(
relative_embeddings, convert_pad_shape([[0, 0],
[pad_length, pad_length], [0, 0]]))
else:
padded_relative_embeddings = relative_embeddings
used_relative_embeddings = padded_relative_embeddings[:,
slice_start_position:slice_end_position]
return used_relative_embeddings
def _relative_position_to_absolute_position(self, x):
batch, heads, length, _ = x.size()
x = torch.nn.functional.pad(x, convert_pad_shape([[0,0],[0,0],[0,0],[0,1]]))
x_flat = x.view([batch, heads, length * 2 * length])
x_flat = torch.nn.functional.pad(x_flat, convert_pad_shape([[0,0],[0,0],[0,length-1]]))
x_final = x_flat.view([batch, heads, length+1, 2*length-1])[:, :, :length, length-1:]
return x_final
def _absolute_position_to_relative_position(self, x):
batch, heads, length, _ = x.size()
x = torch.nn.functional.pad(x, convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length-1]]))
x_flat = x.view([batch, heads, length**2 + length*(length - 1)])
x_flat = torch.nn.functional.pad(x_flat, convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
x_final = x_flat.view([batch, heads, length, 2*length])[:,:,:,1:]
return x_final
def _attention_bias_proximal(self, length):
r = torch.arange(length, dtype=torch.float32)
diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
class FFN(BaseModule):
def __init__(self, in_channels, out_channels, filter_channels, kernel_size,
p_dropout=0.0):
super(FFN, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.filter_channels = filter_channels
self.kernel_size = kernel_size
self.p_dropout = p_dropout
self.conv_1 = torch.nn.Conv1d(in_channels, filter_channels, kernel_size,
padding=kernel_size//2)
self.conv_2 = torch.nn.Conv1d(filter_channels, out_channels, kernel_size,
padding=kernel_size//2)
self.drop = torch.nn.Dropout(p_dropout)
def forward(self, x, x_mask):
x = self.conv_1(x * x_mask)
x = torch.relu(x)
x = self.drop(x)
x = self.conv_2(x * x_mask)
return x * x_mask
class Encoder(BaseModule):
def __init__(self, hidden_channels, filter_channels, n_heads, n_layers,
kernel_size=1, p_dropout=0.0, window_size=None, **kwargs):
super(Encoder, self).__init__()
self.hidden_channels = hidden_channels
self.filter_channels = filter_channels
self.n_heads = n_heads
self.n_layers = n_layers
self.kernel_size = kernel_size
self.p_dropout = p_dropout
self.window_size = window_size
self.drop = torch.nn.Dropout(p_dropout)
self.attn_layers = torch.nn.ModuleList()
self.norm_layers_1 = torch.nn.ModuleList()
self.ffn_layers = torch.nn.ModuleList()
self.norm_layers_2 = torch.nn.ModuleList()
for _ in range(self.n_layers):
self.attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels,
n_heads, window_size=window_size, p_dropout=p_dropout))
self.norm_layers_1.append(LayerNorm(hidden_channels))
self.ffn_layers.append(FFN(hidden_channels, hidden_channels,
filter_channels, kernel_size, p_dropout=p_dropout))
self.norm_layers_2.append(LayerNorm(hidden_channels))
def forward(self, x, x_mask):
attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
for i in range(self.n_layers):
x = x * x_mask
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)
x = x * x_mask
return x
class TextEncoder(BaseModule):
def __init__(self, n_vocab, n_feats, n_channels, filter_channels,
filter_channels_dp, n_heads, n_layers, kernel_size,
p_dropout, window_size=None, spk_emb_dim=64, n_spks=1):
super(TextEncoder, self).__init__()
self.n_vocab = n_vocab
self.n_feats = n_feats
self.n_channels = n_channels
self.filter_channels = filter_channels
self.filter_channels_dp = filter_channels_dp
self.n_heads = n_heads
self.n_layers = n_layers
self.kernel_size = kernel_size
self.p_dropout = p_dropout
self.window_size = window_size
self.spk_emb_dim = spk_emb_dim
self.n_spks = n_spks
self.emb = torch.nn.Embedding(n_vocab, n_channels)
torch.nn.init.normal_(self.emb.weight, 0.0, n_channels**-0.5)
self.prenet = ConvReluNorm(n_channels, n_channels, n_channels,
kernel_size=5, n_layers=3, p_dropout=0.5)
self.encoder = Encoder(n_channels + (spk_emb_dim if n_spks > 1 else 0), filter_channels, n_heads, n_layers,
kernel_size, p_dropout, window_size=window_size)
self.proj_m = torch.nn.Conv1d(n_channels + (spk_emb_dim if n_spks > 1 else 0), n_feats, 1)
self.proj_w = DurationPredictor(n_channels + (spk_emb_dim if n_spks > 1 else 0), filter_channels_dp,
kernel_size, p_dropout)
def forward(self, x, x_lengths, spk=None):
x = self.emb(x) * math.sqrt(self.n_channels)
x = torch.transpose(x, 1, -1)
x_mask = torch.unsqueeze(sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
x = self.prenet(x, x_mask)
if self.n_spks > 1:
x = torch.cat([x, spk.unsqueeze(-1).repeat(1, 1, x.shape[-1])], dim=1)
x = self.encoder(x, x_mask)
mu = self.proj_m(x) * x_mask
x_dp = torch.detach(x)
logw = self.proj_w(x_dp, x_mask)
return mu, logw, x_mask
|