Spaces:
Sleeping
Sleeping
Upload 8 files
Browse files- lib/infer_pack/attentions.py +417 -0
- lib/infer_pack/commons.py +166 -0
- lib/infer_pack/models.py +1142 -0
- lib/infer_pack/models_dml.py +1124 -0
- lib/infer_pack/models_onnx.py +819 -0
- lib/infer_pack/modules.py +522 -0
- lib/infer_pack/onnx_inference.py +145 -0
- lib/infer_pack/transforms.py +209 -0
lib/infer_pack/attentions.py
ADDED
@@ -0,0 +1,417 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import copy
|
2 |
+
import math
|
3 |
+
import numpy as np
|
4 |
+
import torch
|
5 |
+
from torch import nn
|
6 |
+
from torch.nn import functional as F
|
7 |
+
|
8 |
+
from lib.infer_pack import commons
|
9 |
+
from lib.infer_pack import modules
|
10 |
+
from lib.infer_pack.modules import LayerNorm
|
11 |
+
|
12 |
+
|
13 |
+
class Encoder(nn.Module):
|
14 |
+
def __init__(
|
15 |
+
self,
|
16 |
+
hidden_channels,
|
17 |
+
filter_channels,
|
18 |
+
n_heads,
|
19 |
+
n_layers,
|
20 |
+
kernel_size=1,
|
21 |
+
p_dropout=0.0,
|
22 |
+
window_size=10,
|
23 |
+
**kwargs
|
24 |
+
):
|
25 |
+
super().__init__()
|
26 |
+
self.hidden_channels = hidden_channels
|
27 |
+
self.filter_channels = filter_channels
|
28 |
+
self.n_heads = n_heads
|
29 |
+
self.n_layers = n_layers
|
30 |
+
self.kernel_size = kernel_size
|
31 |
+
self.p_dropout = p_dropout
|
32 |
+
self.window_size = window_size
|
33 |
+
|
34 |
+
self.drop = nn.Dropout(p_dropout)
|
35 |
+
self.attn_layers = nn.ModuleList()
|
36 |
+
self.norm_layers_1 = nn.ModuleList()
|
37 |
+
self.ffn_layers = nn.ModuleList()
|
38 |
+
self.norm_layers_2 = nn.ModuleList()
|
39 |
+
for i in range(self.n_layers):
|
40 |
+
self.attn_layers.append(
|
41 |
+
MultiHeadAttention(
|
42 |
+
hidden_channels,
|
43 |
+
hidden_channels,
|
44 |
+
n_heads,
|
45 |
+
p_dropout=p_dropout,
|
46 |
+
window_size=window_size,
|
47 |
+
)
|
48 |
+
)
|
49 |
+
self.norm_layers_1.append(LayerNorm(hidden_channels))
|
50 |
+
self.ffn_layers.append(
|
51 |
+
FFN(
|
52 |
+
hidden_channels,
|
53 |
+
hidden_channels,
|
54 |
+
filter_channels,
|
55 |
+
kernel_size,
|
56 |
+
p_dropout=p_dropout,
|
57 |
+
)
|
58 |
+
)
|
59 |
+
self.norm_layers_2.append(LayerNorm(hidden_channels))
|
60 |
+
|
61 |
+
def forward(self, x, x_mask):
|
62 |
+
attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
|
63 |
+
x = x * x_mask
|
64 |
+
for i in range(self.n_layers):
|
65 |
+
y = self.attn_layers[i](x, x, attn_mask)
|
66 |
+
y = self.drop(y)
|
67 |
+
x = self.norm_layers_1[i](x + y)
|
68 |
+
|
69 |
+
y = self.ffn_layers[i](x, x_mask)
|
70 |
+
y = self.drop(y)
|
71 |
+
x = self.norm_layers_2[i](x + y)
|
72 |
+
x = x * x_mask
|
73 |
+
return x
|
74 |
+
|
75 |
+
|
76 |
+
class Decoder(nn.Module):
|
77 |
+
def __init__(
|
78 |
+
self,
|
79 |
+
hidden_channels,
|
80 |
+
filter_channels,
|
81 |
+
n_heads,
|
82 |
+
n_layers,
|
83 |
+
kernel_size=1,
|
84 |
+
p_dropout=0.0,
|
85 |
+
proximal_bias=False,
|
86 |
+
proximal_init=True,
|
87 |
+
**kwargs
|
88 |
+
):
|
89 |
+
super().__init__()
|
90 |
+
self.hidden_channels = hidden_channels
|
91 |
+
self.filter_channels = filter_channels
|
92 |
+
self.n_heads = n_heads
|
93 |
+
self.n_layers = n_layers
|
94 |
+
self.kernel_size = kernel_size
|
95 |
+
self.p_dropout = p_dropout
|
96 |
+
self.proximal_bias = proximal_bias
|
97 |
+
self.proximal_init = proximal_init
|
98 |
+
|
99 |
+
self.drop = nn.Dropout(p_dropout)
|
100 |
+
self.self_attn_layers = nn.ModuleList()
|
101 |
+
self.norm_layers_0 = nn.ModuleList()
|
102 |
+
self.encdec_attn_layers = nn.ModuleList()
|
103 |
+
self.norm_layers_1 = nn.ModuleList()
|
104 |
+
self.ffn_layers = nn.ModuleList()
|
105 |
+
self.norm_layers_2 = nn.ModuleList()
|
106 |
+
for i in range(self.n_layers):
|
107 |
+
self.self_attn_layers.append(
|
108 |
+
MultiHeadAttention(
|
109 |
+
hidden_channels,
|
110 |
+
hidden_channels,
|
111 |
+
n_heads,
|
112 |
+
p_dropout=p_dropout,
|
113 |
+
proximal_bias=proximal_bias,
|
114 |
+
proximal_init=proximal_init,
|
115 |
+
)
|
116 |
+
)
|
117 |
+
self.norm_layers_0.append(LayerNorm(hidden_channels))
|
118 |
+
self.encdec_attn_layers.append(
|
119 |
+
MultiHeadAttention(
|
120 |
+
hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
|
121 |
+
)
|
122 |
+
)
|
123 |
+
self.norm_layers_1.append(LayerNorm(hidden_channels))
|
124 |
+
self.ffn_layers.append(
|
125 |
+
FFN(
|
126 |
+
hidden_channels,
|
127 |
+
hidden_channels,
|
128 |
+
filter_channels,
|
129 |
+
kernel_size,
|
130 |
+
p_dropout=p_dropout,
|
131 |
+
causal=True,
|
132 |
+
)
|
133 |
+
)
|
134 |
+
self.norm_layers_2.append(LayerNorm(hidden_channels))
|
135 |
+
|
136 |
+
def forward(self, x, x_mask, h, h_mask):
|
137 |
+
"""
|
138 |
+
x: decoder input
|
139 |
+
h: encoder output
|
140 |
+
"""
|
141 |
+
self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
|
142 |
+
device=x.device, dtype=x.dtype
|
143 |
+
)
|
144 |
+
encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
|
145 |
+
x = x * x_mask
|
146 |
+
for i in range(self.n_layers):
|
147 |
+
y = self.self_attn_layers[i](x, x, self_attn_mask)
|
148 |
+
y = self.drop(y)
|
149 |
+
x = self.norm_layers_0[i](x + y)
|
150 |
+
|
151 |
+
y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
|
152 |
+
y = self.drop(y)
|
153 |
+
x = self.norm_layers_1[i](x + y)
|
154 |
+
|
155 |
+
y = self.ffn_layers[i](x, x_mask)
|
156 |
+
y = self.drop(y)
|
157 |
+
x = self.norm_layers_2[i](x + y)
|
158 |
+
x = x * x_mask
|
159 |
+
return x
|
160 |
+
|
161 |
+
|
162 |
+
class MultiHeadAttention(nn.Module):
|
163 |
+
def __init__(
|
164 |
+
self,
|
165 |
+
channels,
|
166 |
+
out_channels,
|
167 |
+
n_heads,
|
168 |
+
p_dropout=0.0,
|
169 |
+
window_size=None,
|
170 |
+
heads_share=True,
|
171 |
+
block_length=None,
|
172 |
+
proximal_bias=False,
|
173 |
+
proximal_init=False,
|
174 |
+
):
|
175 |
+
super().__init__()
|
176 |
+
assert channels % n_heads == 0
|
177 |
+
|
178 |
+
self.channels = channels
|
179 |
+
self.out_channels = out_channels
|
180 |
+
self.n_heads = n_heads
|
181 |
+
self.p_dropout = p_dropout
|
182 |
+
self.window_size = window_size
|
183 |
+
self.heads_share = heads_share
|
184 |
+
self.block_length = block_length
|
185 |
+
self.proximal_bias = proximal_bias
|
186 |
+
self.proximal_init = proximal_init
|
187 |
+
self.attn = None
|
188 |
+
|
189 |
+
self.k_channels = channels // n_heads
|
190 |
+
self.conv_q = nn.Conv1d(channels, channels, 1)
|
191 |
+
self.conv_k = nn.Conv1d(channels, channels, 1)
|
192 |
+
self.conv_v = nn.Conv1d(channels, channels, 1)
|
193 |
+
self.conv_o = nn.Conv1d(channels, out_channels, 1)
|
194 |
+
self.drop = nn.Dropout(p_dropout)
|
195 |
+
|
196 |
+
if window_size is not None:
|
197 |
+
n_heads_rel = 1 if heads_share else n_heads
|
198 |
+
rel_stddev = self.k_channels**-0.5
|
199 |
+
self.emb_rel_k = nn.Parameter(
|
200 |
+
torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
|
201 |
+
* rel_stddev
|
202 |
+
)
|
203 |
+
self.emb_rel_v = nn.Parameter(
|
204 |
+
torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
|
205 |
+
* rel_stddev
|
206 |
+
)
|
207 |
+
|
208 |
+
nn.init.xavier_uniform_(self.conv_q.weight)
|
209 |
+
nn.init.xavier_uniform_(self.conv_k.weight)
|
210 |
+
nn.init.xavier_uniform_(self.conv_v.weight)
|
211 |
+
if proximal_init:
|
212 |
+
with torch.no_grad():
|
213 |
+
self.conv_k.weight.copy_(self.conv_q.weight)
|
214 |
+
self.conv_k.bias.copy_(self.conv_q.bias)
|
215 |
+
|
216 |
+
def forward(self, x, c, attn_mask=None):
|
217 |
+
q = self.conv_q(x)
|
218 |
+
k = self.conv_k(c)
|
219 |
+
v = self.conv_v(c)
|
220 |
+
|
221 |
+
x, self.attn = self.attention(q, k, v, mask=attn_mask)
|
222 |
+
|
223 |
+
x = self.conv_o(x)
|
224 |
+
return x
|
225 |
+
|
226 |
+
def attention(self, query, key, value, mask=None):
|
227 |
+
# reshape [b, d, t] -> [b, n_h, t, d_k]
|
228 |
+
b, d, t_s, t_t = (*key.size(), query.size(2))
|
229 |
+
query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
|
230 |
+
key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
|
231 |
+
value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
|
232 |
+
|
233 |
+
scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
|
234 |
+
if self.window_size is not None:
|
235 |
+
assert (
|
236 |
+
t_s == t_t
|
237 |
+
), "Relative attention is only available for self-attention."
|
238 |
+
key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
|
239 |
+
rel_logits = self._matmul_with_relative_keys(
|
240 |
+
query / math.sqrt(self.k_channels), key_relative_embeddings
|
241 |
+
)
|
242 |
+
scores_local = self._relative_position_to_absolute_position(rel_logits)
|
243 |
+
scores = scores + scores_local
|
244 |
+
if self.proximal_bias:
|
245 |
+
assert t_s == t_t, "Proximal bias is only available for self-attention."
|
246 |
+
scores = scores + self._attention_bias_proximal(t_s).to(
|
247 |
+
device=scores.device, dtype=scores.dtype
|
248 |
+
)
|
249 |
+
if mask is not None:
|
250 |
+
scores = scores.masked_fill(mask == 0, -1e4)
|
251 |
+
if self.block_length is not None:
|
252 |
+
assert (
|
253 |
+
t_s == t_t
|
254 |
+
), "Local attention is only available for self-attention."
|
255 |
+
block_mask = (
|
256 |
+
torch.ones_like(scores)
|
257 |
+
.triu(-self.block_length)
|
258 |
+
.tril(self.block_length)
|
259 |
+
)
|
260 |
+
scores = scores.masked_fill(block_mask == 0, -1e4)
|
261 |
+
p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
|
262 |
+
p_attn = self.drop(p_attn)
|
263 |
+
output = torch.matmul(p_attn, value)
|
264 |
+
if self.window_size is not None:
|
265 |
+
relative_weights = self._absolute_position_to_relative_position(p_attn)
|
266 |
+
value_relative_embeddings = self._get_relative_embeddings(
|
267 |
+
self.emb_rel_v, t_s
|
268 |
+
)
|
269 |
+
output = output + self._matmul_with_relative_values(
|
270 |
+
relative_weights, value_relative_embeddings
|
271 |
+
)
|
272 |
+
output = (
|
273 |
+
output.transpose(2, 3).contiguous().view(b, d, t_t)
|
274 |
+
) # [b, n_h, t_t, d_k] -> [b, d, t_t]
|
275 |
+
return output, p_attn
|
276 |
+
|
277 |
+
def _matmul_with_relative_values(self, x, y):
|
278 |
+
"""
|
279 |
+
x: [b, h, l, m]
|
280 |
+
y: [h or 1, m, d]
|
281 |
+
ret: [b, h, l, d]
|
282 |
+
"""
|
283 |
+
ret = torch.matmul(x, y.unsqueeze(0))
|
284 |
+
return ret
|
285 |
+
|
286 |
+
def _matmul_with_relative_keys(self, x, y):
|
287 |
+
"""
|
288 |
+
x: [b, h, l, d]
|
289 |
+
y: [h or 1, m, d]
|
290 |
+
ret: [b, h, l, m]
|
291 |
+
"""
|
292 |
+
ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
|
293 |
+
return ret
|
294 |
+
|
295 |
+
def _get_relative_embeddings(self, relative_embeddings, length):
|
296 |
+
max_relative_position = 2 * self.window_size + 1
|
297 |
+
# Pad first before slice to avoid using cond ops.
|
298 |
+
pad_length = max(length - (self.window_size + 1), 0)
|
299 |
+
slice_start_position = max((self.window_size + 1) - length, 0)
|
300 |
+
slice_end_position = slice_start_position + 2 * length - 1
|
301 |
+
if pad_length > 0:
|
302 |
+
padded_relative_embeddings = F.pad(
|
303 |
+
relative_embeddings,
|
304 |
+
commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
|
305 |
+
)
|
306 |
+
else:
|
307 |
+
padded_relative_embeddings = relative_embeddings
|
308 |
+
used_relative_embeddings = padded_relative_embeddings[
|
309 |
+
:, slice_start_position:slice_end_position
|
310 |
+
]
|
311 |
+
return used_relative_embeddings
|
312 |
+
|
313 |
+
def _relative_position_to_absolute_position(self, x):
|
314 |
+
"""
|
315 |
+
x: [b, h, l, 2*l-1]
|
316 |
+
ret: [b, h, l, l]
|
317 |
+
"""
|
318 |
+
batch, heads, length, _ = x.size()
|
319 |
+
# Concat columns of pad to shift from relative to absolute indexing.
|
320 |
+
x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
|
321 |
+
|
322 |
+
# Concat extra elements so to add up to shape (len+1, 2*len-1).
|
323 |
+
x_flat = x.view([batch, heads, length * 2 * length])
|
324 |
+
x_flat = F.pad(
|
325 |
+
x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
|
326 |
+
)
|
327 |
+
|
328 |
+
# Reshape and slice out the padded elements.
|
329 |
+
x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
|
330 |
+
:, :, :length, length - 1 :
|
331 |
+
]
|
332 |
+
return x_final
|
333 |
+
|
334 |
+
def _absolute_position_to_relative_position(self, x):
|
335 |
+
"""
|
336 |
+
x: [b, h, l, l]
|
337 |
+
ret: [b, h, l, 2*l-1]
|
338 |
+
"""
|
339 |
+
batch, heads, length, _ = x.size()
|
340 |
+
# padd along column
|
341 |
+
x = F.pad(
|
342 |
+
x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
|
343 |
+
)
|
344 |
+
x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
|
345 |
+
# add 0's in the beginning that will skew the elements after reshape
|
346 |
+
x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
|
347 |
+
x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
|
348 |
+
return x_final
|
349 |
+
|
350 |
+
def _attention_bias_proximal(self, length):
|
351 |
+
"""Bias for self-attention to encourage attention to close positions.
|
352 |
+
Args:
|
353 |
+
length: an integer scalar.
|
354 |
+
Returns:
|
355 |
+
a Tensor with shape [1, 1, length, length]
|
356 |
+
"""
|
357 |
+
r = torch.arange(length, dtype=torch.float32)
|
358 |
+
diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
|
359 |
+
return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
|
360 |
+
|
361 |
+
|
362 |
+
class FFN(nn.Module):
|
363 |
+
def __init__(
|
364 |
+
self,
|
365 |
+
in_channels,
|
366 |
+
out_channels,
|
367 |
+
filter_channels,
|
368 |
+
kernel_size,
|
369 |
+
p_dropout=0.0,
|
370 |
+
activation=None,
|
371 |
+
causal=False,
|
372 |
+
):
|
373 |
+
super().__init__()
|
374 |
+
self.in_channels = in_channels
|
375 |
+
self.out_channels = out_channels
|
376 |
+
self.filter_channels = filter_channels
|
377 |
+
self.kernel_size = kernel_size
|
378 |
+
self.p_dropout = p_dropout
|
379 |
+
self.activation = activation
|
380 |
+
self.causal = causal
|
381 |
+
|
382 |
+
if causal:
|
383 |
+
self.padding = self._causal_padding
|
384 |
+
else:
|
385 |
+
self.padding = self._same_padding
|
386 |
+
|
387 |
+
self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
|
388 |
+
self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
|
389 |
+
self.drop = nn.Dropout(p_dropout)
|
390 |
+
|
391 |
+
def forward(self, x, x_mask):
|
392 |
+
x = self.conv_1(self.padding(x * x_mask))
|
393 |
+
if self.activation == "gelu":
|
394 |
+
x = x * torch.sigmoid(1.702 * x)
|
395 |
+
else:
|
396 |
+
x = torch.relu(x)
|
397 |
+
x = self.drop(x)
|
398 |
+
x = self.conv_2(self.padding(x * x_mask))
|
399 |
+
return x * x_mask
|
400 |
+
|
401 |
+
def _causal_padding(self, x):
|
402 |
+
if self.kernel_size == 1:
|
403 |
+
return x
|
404 |
+
pad_l = self.kernel_size - 1
|
405 |
+
pad_r = 0
|
406 |
+
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
|
407 |
+
x = F.pad(x, commons.convert_pad_shape(padding))
|
408 |
+
return x
|
409 |
+
|
410 |
+
def _same_padding(self, x):
|
411 |
+
if self.kernel_size == 1:
|
412 |
+
return x
|
413 |
+
pad_l = (self.kernel_size - 1) // 2
|
414 |
+
pad_r = self.kernel_size // 2
|
415 |
+
padding = [[0, 0], [0, 0], [pad_l, pad_r]]
|
416 |
+
x = F.pad(x, commons.convert_pad_shape(padding))
|
417 |
+
return x
|
lib/infer_pack/commons.py
ADDED
@@ -0,0 +1,166 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math
|
2 |
+
import numpy as np
|
3 |
+
import torch
|
4 |
+
from torch import nn
|
5 |
+
from torch.nn import functional as F
|
6 |
+
|
7 |
+
|
8 |
+
def init_weights(m, mean=0.0, std=0.01):
|
9 |
+
classname = m.__class__.__name__
|
10 |
+
if classname.find("Conv") != -1:
|
11 |
+
m.weight.data.normal_(mean, std)
|
12 |
+
|
13 |
+
|
14 |
+
def get_padding(kernel_size, dilation=1):
|
15 |
+
return int((kernel_size * dilation - dilation) / 2)
|
16 |
+
|
17 |
+
|
18 |
+
def convert_pad_shape(pad_shape):
|
19 |
+
l = pad_shape[::-1]
|
20 |
+
pad_shape = [item for sublist in l for item in sublist]
|
21 |
+
return pad_shape
|
22 |
+
|
23 |
+
|
24 |
+
def kl_divergence(m_p, logs_p, m_q, logs_q):
|
25 |
+
"""KL(P||Q)"""
|
26 |
+
kl = (logs_q - logs_p) - 0.5
|
27 |
+
kl += (
|
28 |
+
0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
|
29 |
+
)
|
30 |
+
return kl
|
31 |
+
|
32 |
+
|
33 |
+
def rand_gumbel(shape):
|
34 |
+
"""Sample from the Gumbel distribution, protect from overflows."""
|
35 |
+
uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
|
36 |
+
return -torch.log(-torch.log(uniform_samples))
|
37 |
+
|
38 |
+
|
39 |
+
def rand_gumbel_like(x):
|
40 |
+
g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
|
41 |
+
return g
|
42 |
+
|
43 |
+
|
44 |
+
def slice_segments(x, ids_str, segment_size=4):
|
45 |
+
ret = torch.zeros_like(x[:, :, :segment_size])
|
46 |
+
for i in range(x.size(0)):
|
47 |
+
idx_str = ids_str[i]
|
48 |
+
idx_end = idx_str + segment_size
|
49 |
+
ret[i] = x[i, :, idx_str:idx_end]
|
50 |
+
return ret
|
51 |
+
|
52 |
+
|
53 |
+
def slice_segments2(x, ids_str, segment_size=4):
|
54 |
+
ret = torch.zeros_like(x[:, :segment_size])
|
55 |
+
for i in range(x.size(0)):
|
56 |
+
idx_str = ids_str[i]
|
57 |
+
idx_end = idx_str + segment_size
|
58 |
+
ret[i] = x[i, idx_str:idx_end]
|
59 |
+
return ret
|
60 |
+
|
61 |
+
|
62 |
+
def rand_slice_segments(x, x_lengths=None, segment_size=4):
|
63 |
+
b, d, t = x.size()
|
64 |
+
if x_lengths is None:
|
65 |
+
x_lengths = t
|
66 |
+
ids_str_max = x_lengths - segment_size + 1
|
67 |
+
ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
|
68 |
+
ret = slice_segments(x, ids_str, segment_size)
|
69 |
+
return ret, ids_str
|
70 |
+
|
71 |
+
|
72 |
+
def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
|
73 |
+
position = torch.arange(length, dtype=torch.float)
|
74 |
+
num_timescales = channels // 2
|
75 |
+
log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
|
76 |
+
num_timescales - 1
|
77 |
+
)
|
78 |
+
inv_timescales = min_timescale * torch.exp(
|
79 |
+
torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
|
80 |
+
)
|
81 |
+
scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
|
82 |
+
signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
|
83 |
+
signal = F.pad(signal, [0, 0, 0, channels % 2])
|
84 |
+
signal = signal.view(1, channels, length)
|
85 |
+
return signal
|
86 |
+
|
87 |
+
|
88 |
+
def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
|
89 |
+
b, channels, length = x.size()
|
90 |
+
signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
|
91 |
+
return x + signal.to(dtype=x.dtype, device=x.device)
|
92 |
+
|
93 |
+
|
94 |
+
def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
|
95 |
+
b, channels, length = x.size()
|
96 |
+
signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
|
97 |
+
return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
|
98 |
+
|
99 |
+
|
100 |
+
def subsequent_mask(length):
|
101 |
+
mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
|
102 |
+
return mask
|
103 |
+
|
104 |
+
|
105 |
+
@torch.jit.script
|
106 |
+
def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
|
107 |
+
n_channels_int = n_channels[0]
|
108 |
+
in_act = input_a + input_b
|
109 |
+
t_act = torch.tanh(in_act[:, :n_channels_int, :])
|
110 |
+
s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
|
111 |
+
acts = t_act * s_act
|
112 |
+
return acts
|
113 |
+
|
114 |
+
|
115 |
+
def convert_pad_shape(pad_shape):
|
116 |
+
l = pad_shape[::-1]
|
117 |
+
pad_shape = [item for sublist in l for item in sublist]
|
118 |
+
return pad_shape
|
119 |
+
|
120 |
+
|
121 |
+
def shift_1d(x):
|
122 |
+
x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
|
123 |
+
return x
|
124 |
+
|
125 |
+
|
126 |
+
def sequence_mask(length, max_length=None):
|
127 |
+
if max_length is None:
|
128 |
+
max_length = length.max()
|
129 |
+
x = torch.arange(max_length, dtype=length.dtype, device=length.device)
|
130 |
+
return x.unsqueeze(0) < length.unsqueeze(1)
|
131 |
+
|
132 |
+
|
133 |
+
def generate_path(duration, mask):
|
134 |
+
"""
|
135 |
+
duration: [b, 1, t_x]
|
136 |
+
mask: [b, 1, t_y, t_x]
|
137 |
+
"""
|
138 |
+
device = duration.device
|
139 |
+
|
140 |
+
b, _, t_y, t_x = mask.shape
|
141 |
+
cum_duration = torch.cumsum(duration, -1)
|
142 |
+
|
143 |
+
cum_duration_flat = cum_duration.view(b * t_x)
|
144 |
+
path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
|
145 |
+
path = path.view(b, t_x, t_y)
|
146 |
+
path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
|
147 |
+
path = path.unsqueeze(1).transpose(2, 3) * mask
|
148 |
+
return path
|
149 |
+
|
150 |
+
|
151 |
+
def clip_grad_value_(parameters, clip_value, norm_type=2):
|
152 |
+
if isinstance(parameters, torch.Tensor):
|
153 |
+
parameters = [parameters]
|
154 |
+
parameters = list(filter(lambda p: p.grad is not None, parameters))
|
155 |
+
norm_type = float(norm_type)
|
156 |
+
if clip_value is not None:
|
157 |
+
clip_value = float(clip_value)
|
158 |
+
|
159 |
+
total_norm = 0
|
160 |
+
for p in parameters:
|
161 |
+
param_norm = p.grad.data.norm(norm_type)
|
162 |
+
total_norm += param_norm.item() ** norm_type
|
163 |
+
if clip_value is not None:
|
164 |
+
p.grad.data.clamp_(min=-clip_value, max=clip_value)
|
165 |
+
total_norm = total_norm ** (1.0 / norm_type)
|
166 |
+
return total_norm
|
lib/infer_pack/models.py
ADDED
@@ -0,0 +1,1142 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math, pdb, os
|
2 |
+
from time import time as ttime
|
3 |
+
import torch
|
4 |
+
from torch import nn
|
5 |
+
from torch.nn import functional as F
|
6 |
+
from lib.infer_pack import modules
|
7 |
+
from lib.infer_pack import attentions
|
8 |
+
from lib.infer_pack import commons
|
9 |
+
from lib.infer_pack.commons import init_weights, get_padding
|
10 |
+
from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
|
11 |
+
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
|
12 |
+
from lib.infer_pack.commons import init_weights
|
13 |
+
import numpy as np
|
14 |
+
from lib.infer_pack import commons
|
15 |
+
|
16 |
+
|
17 |
+
class TextEncoder256(nn.Module):
|
18 |
+
def __init__(
|
19 |
+
self,
|
20 |
+
out_channels,
|
21 |
+
hidden_channels,
|
22 |
+
filter_channels,
|
23 |
+
n_heads,
|
24 |
+
n_layers,
|
25 |
+
kernel_size,
|
26 |
+
p_dropout,
|
27 |
+
f0=True,
|
28 |
+
):
|
29 |
+
super().__init__()
|
30 |
+
self.out_channels = out_channels
|
31 |
+
self.hidden_channels = hidden_channels
|
32 |
+
self.filter_channels = filter_channels
|
33 |
+
self.n_heads = n_heads
|
34 |
+
self.n_layers = n_layers
|
35 |
+
self.kernel_size = kernel_size
|
36 |
+
self.p_dropout = p_dropout
|
37 |
+
self.emb_phone = nn.Linear(256, hidden_channels)
|
38 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
39 |
+
if f0 == True:
|
40 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
41 |
+
self.encoder = attentions.Encoder(
|
42 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
43 |
+
)
|
44 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
45 |
+
|
46 |
+
def forward(self, phone, pitch, lengths):
|
47 |
+
if pitch == None:
|
48 |
+
x = self.emb_phone(phone)
|
49 |
+
else:
|
50 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
51 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
52 |
+
x = self.lrelu(x)
|
53 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
54 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
55 |
+
x.dtype
|
56 |
+
)
|
57 |
+
x = self.encoder(x * x_mask, x_mask)
|
58 |
+
stats = self.proj(x) * x_mask
|
59 |
+
|
60 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
61 |
+
return m, logs, x_mask
|
62 |
+
|
63 |
+
|
64 |
+
class TextEncoder768(nn.Module):
|
65 |
+
def __init__(
|
66 |
+
self,
|
67 |
+
out_channels,
|
68 |
+
hidden_channels,
|
69 |
+
filter_channels,
|
70 |
+
n_heads,
|
71 |
+
n_layers,
|
72 |
+
kernel_size,
|
73 |
+
p_dropout,
|
74 |
+
f0=True,
|
75 |
+
):
|
76 |
+
super().__init__()
|
77 |
+
self.out_channels = out_channels
|
78 |
+
self.hidden_channels = hidden_channels
|
79 |
+
self.filter_channels = filter_channels
|
80 |
+
self.n_heads = n_heads
|
81 |
+
self.n_layers = n_layers
|
82 |
+
self.kernel_size = kernel_size
|
83 |
+
self.p_dropout = p_dropout
|
84 |
+
self.emb_phone = nn.Linear(768, hidden_channels)
|
85 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
86 |
+
if f0 == True:
|
87 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
88 |
+
self.encoder = attentions.Encoder(
|
89 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
90 |
+
)
|
91 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
92 |
+
|
93 |
+
def forward(self, phone, pitch, lengths):
|
94 |
+
if pitch == None:
|
95 |
+
x = self.emb_phone(phone)
|
96 |
+
else:
|
97 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
98 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
99 |
+
x = self.lrelu(x)
|
100 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
101 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
102 |
+
x.dtype
|
103 |
+
)
|
104 |
+
x = self.encoder(x * x_mask, x_mask)
|
105 |
+
stats = self.proj(x) * x_mask
|
106 |
+
|
107 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
108 |
+
return m, logs, x_mask
|
109 |
+
|
110 |
+
|
111 |
+
class ResidualCouplingBlock(nn.Module):
|
112 |
+
def __init__(
|
113 |
+
self,
|
114 |
+
channels,
|
115 |
+
hidden_channels,
|
116 |
+
kernel_size,
|
117 |
+
dilation_rate,
|
118 |
+
n_layers,
|
119 |
+
n_flows=4,
|
120 |
+
gin_channels=0,
|
121 |
+
):
|
122 |
+
super().__init__()
|
123 |
+
self.channels = channels
|
124 |
+
self.hidden_channels = hidden_channels
|
125 |
+
self.kernel_size = kernel_size
|
126 |
+
self.dilation_rate = dilation_rate
|
127 |
+
self.n_layers = n_layers
|
128 |
+
self.n_flows = n_flows
|
129 |
+
self.gin_channels = gin_channels
|
130 |
+
|
131 |
+
self.flows = nn.ModuleList()
|
132 |
+
for i in range(n_flows):
|
133 |
+
self.flows.append(
|
134 |
+
modules.ResidualCouplingLayer(
|
135 |
+
channels,
|
136 |
+
hidden_channels,
|
137 |
+
kernel_size,
|
138 |
+
dilation_rate,
|
139 |
+
n_layers,
|
140 |
+
gin_channels=gin_channels,
|
141 |
+
mean_only=True,
|
142 |
+
)
|
143 |
+
)
|
144 |
+
self.flows.append(modules.Flip())
|
145 |
+
|
146 |
+
def forward(self, x, x_mask, g=None, reverse=False):
|
147 |
+
if not reverse:
|
148 |
+
for flow in self.flows:
|
149 |
+
x, _ = flow(x, x_mask, g=g, reverse=reverse)
|
150 |
+
else:
|
151 |
+
for flow in reversed(self.flows):
|
152 |
+
x = flow(x, x_mask, g=g, reverse=reverse)
|
153 |
+
return x
|
154 |
+
|
155 |
+
def remove_weight_norm(self):
|
156 |
+
for i in range(self.n_flows):
|
157 |
+
self.flows[i * 2].remove_weight_norm()
|
158 |
+
|
159 |
+
|
160 |
+
class PosteriorEncoder(nn.Module):
|
161 |
+
def __init__(
|
162 |
+
self,
|
163 |
+
in_channels,
|
164 |
+
out_channels,
|
165 |
+
hidden_channels,
|
166 |
+
kernel_size,
|
167 |
+
dilation_rate,
|
168 |
+
n_layers,
|
169 |
+
gin_channels=0,
|
170 |
+
):
|
171 |
+
super().__init__()
|
172 |
+
self.in_channels = in_channels
|
173 |
+
self.out_channels = out_channels
|
174 |
+
self.hidden_channels = hidden_channels
|
175 |
+
self.kernel_size = kernel_size
|
176 |
+
self.dilation_rate = dilation_rate
|
177 |
+
self.n_layers = n_layers
|
178 |
+
self.gin_channels = gin_channels
|
179 |
+
|
180 |
+
self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
|
181 |
+
self.enc = modules.WN(
|
182 |
+
hidden_channels,
|
183 |
+
kernel_size,
|
184 |
+
dilation_rate,
|
185 |
+
n_layers,
|
186 |
+
gin_channels=gin_channels,
|
187 |
+
)
|
188 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
189 |
+
|
190 |
+
def forward(self, x, x_lengths, g=None):
|
191 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
|
192 |
+
x.dtype
|
193 |
+
)
|
194 |
+
x = self.pre(x) * x_mask
|
195 |
+
x = self.enc(x, x_mask, g=g)
|
196 |
+
stats = self.proj(x) * x_mask
|
197 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
198 |
+
z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
|
199 |
+
return z, m, logs, x_mask
|
200 |
+
|
201 |
+
def remove_weight_norm(self):
|
202 |
+
self.enc.remove_weight_norm()
|
203 |
+
|
204 |
+
|
205 |
+
class Generator(torch.nn.Module):
|
206 |
+
def __init__(
|
207 |
+
self,
|
208 |
+
initial_channel,
|
209 |
+
resblock,
|
210 |
+
resblock_kernel_sizes,
|
211 |
+
resblock_dilation_sizes,
|
212 |
+
upsample_rates,
|
213 |
+
upsample_initial_channel,
|
214 |
+
upsample_kernel_sizes,
|
215 |
+
gin_channels=0,
|
216 |
+
):
|
217 |
+
super(Generator, self).__init__()
|
218 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
219 |
+
self.num_upsamples = len(upsample_rates)
|
220 |
+
self.conv_pre = Conv1d(
|
221 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
222 |
+
)
|
223 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
224 |
+
|
225 |
+
self.ups = nn.ModuleList()
|
226 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
227 |
+
self.ups.append(
|
228 |
+
weight_norm(
|
229 |
+
ConvTranspose1d(
|
230 |
+
upsample_initial_channel // (2**i),
|
231 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
232 |
+
k,
|
233 |
+
u,
|
234 |
+
padding=(k - u) // 2,
|
235 |
+
)
|
236 |
+
)
|
237 |
+
)
|
238 |
+
|
239 |
+
self.resblocks = nn.ModuleList()
|
240 |
+
for i in range(len(self.ups)):
|
241 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
242 |
+
for j, (k, d) in enumerate(
|
243 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
244 |
+
):
|
245 |
+
self.resblocks.append(resblock(ch, k, d))
|
246 |
+
|
247 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
248 |
+
self.ups.apply(init_weights)
|
249 |
+
|
250 |
+
if gin_channels != 0:
|
251 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
252 |
+
|
253 |
+
def forward(self, x, g=None):
|
254 |
+
x = self.conv_pre(x)
|
255 |
+
if g is not None:
|
256 |
+
x = x + self.cond(g)
|
257 |
+
|
258 |
+
for i in range(self.num_upsamples):
|
259 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
260 |
+
x = self.ups[i](x)
|
261 |
+
xs = None
|
262 |
+
for j in range(self.num_kernels):
|
263 |
+
if xs is None:
|
264 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
265 |
+
else:
|
266 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
267 |
+
x = xs / self.num_kernels
|
268 |
+
x = F.leaky_relu(x)
|
269 |
+
x = self.conv_post(x)
|
270 |
+
x = torch.tanh(x)
|
271 |
+
|
272 |
+
return x
|
273 |
+
|
274 |
+
def remove_weight_norm(self):
|
275 |
+
for l in self.ups:
|
276 |
+
remove_weight_norm(l)
|
277 |
+
for l in self.resblocks:
|
278 |
+
l.remove_weight_norm()
|
279 |
+
|
280 |
+
|
281 |
+
class SineGen(torch.nn.Module):
|
282 |
+
"""Definition of sine generator
|
283 |
+
SineGen(samp_rate, harmonic_num = 0,
|
284 |
+
sine_amp = 0.1, noise_std = 0.003,
|
285 |
+
voiced_threshold = 0,
|
286 |
+
flag_for_pulse=False)
|
287 |
+
samp_rate: sampling rate in Hz
|
288 |
+
harmonic_num: number of harmonic overtones (default 0)
|
289 |
+
sine_amp: amplitude of sine-wavefrom (default 0.1)
|
290 |
+
noise_std: std of Gaussian noise (default 0.003)
|
291 |
+
voiced_thoreshold: F0 threshold for U/V classification (default 0)
|
292 |
+
flag_for_pulse: this SinGen is used inside PulseGen (default False)
|
293 |
+
Note: when flag_for_pulse is True, the first time step of a voiced
|
294 |
+
segment is always sin(np.pi) or cos(0)
|
295 |
+
"""
|
296 |
+
|
297 |
+
def __init__(
|
298 |
+
self,
|
299 |
+
samp_rate,
|
300 |
+
harmonic_num=0,
|
301 |
+
sine_amp=0.1,
|
302 |
+
noise_std=0.003,
|
303 |
+
voiced_threshold=0,
|
304 |
+
flag_for_pulse=False,
|
305 |
+
):
|
306 |
+
super(SineGen, self).__init__()
|
307 |
+
self.sine_amp = sine_amp
|
308 |
+
self.noise_std = noise_std
|
309 |
+
self.harmonic_num = harmonic_num
|
310 |
+
self.dim = self.harmonic_num + 1
|
311 |
+
self.sampling_rate = samp_rate
|
312 |
+
self.voiced_threshold = voiced_threshold
|
313 |
+
|
314 |
+
def _f02uv(self, f0):
|
315 |
+
# generate uv signal
|
316 |
+
uv = torch.ones_like(f0)
|
317 |
+
uv = uv * (f0 > self.voiced_threshold)
|
318 |
+
return uv
|
319 |
+
|
320 |
+
def forward(self, f0, upp):
|
321 |
+
"""sine_tensor, uv = forward(f0)
|
322 |
+
input F0: tensor(batchsize=1, length, dim=1)
|
323 |
+
f0 for unvoiced steps should be 0
|
324 |
+
output sine_tensor: tensor(batchsize=1, length, dim)
|
325 |
+
output uv: tensor(batchsize=1, length, 1)
|
326 |
+
"""
|
327 |
+
with torch.no_grad():
|
328 |
+
f0 = f0[:, None].transpose(1, 2)
|
329 |
+
f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
|
330 |
+
# fundamental component
|
331 |
+
f0_buf[:, :, 0] = f0[:, :, 0]
|
332 |
+
for idx in np.arange(self.harmonic_num):
|
333 |
+
f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
|
334 |
+
idx + 2
|
335 |
+
) # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
|
336 |
+
rad_values = (f0_buf / self.sampling_rate) % 1 ###%1意味着n_har的乘积无法后处理优化
|
337 |
+
rand_ini = torch.rand(
|
338 |
+
f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
|
339 |
+
)
|
340 |
+
rand_ini[:, 0] = 0
|
341 |
+
rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
|
342 |
+
tmp_over_one = torch.cumsum(rad_values, 1) # % 1 #####%1意味着后面的cumsum无法再优化
|
343 |
+
tmp_over_one *= upp
|
344 |
+
tmp_over_one = F.interpolate(
|
345 |
+
tmp_over_one.transpose(2, 1),
|
346 |
+
scale_factor=upp,
|
347 |
+
mode="linear",
|
348 |
+
align_corners=True,
|
349 |
+
).transpose(2, 1)
|
350 |
+
rad_values = F.interpolate(
|
351 |
+
rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
|
352 |
+
).transpose(
|
353 |
+
2, 1
|
354 |
+
) #######
|
355 |
+
tmp_over_one %= 1
|
356 |
+
tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
|
357 |
+
cumsum_shift = torch.zeros_like(rad_values)
|
358 |
+
cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
|
359 |
+
sine_waves = torch.sin(
|
360 |
+
torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
|
361 |
+
)
|
362 |
+
sine_waves = sine_waves * self.sine_amp
|
363 |
+
uv = self._f02uv(f0)
|
364 |
+
uv = F.interpolate(
|
365 |
+
uv.transpose(2, 1), scale_factor=upp, mode="nearest"
|
366 |
+
).transpose(2, 1)
|
367 |
+
noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
|
368 |
+
noise = noise_amp * torch.randn_like(sine_waves)
|
369 |
+
sine_waves = sine_waves * uv + noise
|
370 |
+
return sine_waves, uv, noise
|
371 |
+
|
372 |
+
|
373 |
+
class SourceModuleHnNSF(torch.nn.Module):
|
374 |
+
"""SourceModule for hn-nsf
|
375 |
+
SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
|
376 |
+
add_noise_std=0.003, voiced_threshod=0)
|
377 |
+
sampling_rate: sampling_rate in Hz
|
378 |
+
harmonic_num: number of harmonic above F0 (default: 0)
|
379 |
+
sine_amp: amplitude of sine source signal (default: 0.1)
|
380 |
+
add_noise_std: std of additive Gaussian noise (default: 0.003)
|
381 |
+
note that amplitude of noise in unvoiced is decided
|
382 |
+
by sine_amp
|
383 |
+
voiced_threshold: threhold to set U/V given F0 (default: 0)
|
384 |
+
Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
|
385 |
+
F0_sampled (batchsize, length, 1)
|
386 |
+
Sine_source (batchsize, length, 1)
|
387 |
+
noise_source (batchsize, length 1)
|
388 |
+
uv (batchsize, length, 1)
|
389 |
+
"""
|
390 |
+
|
391 |
+
def __init__(
|
392 |
+
self,
|
393 |
+
sampling_rate,
|
394 |
+
harmonic_num=0,
|
395 |
+
sine_amp=0.1,
|
396 |
+
add_noise_std=0.003,
|
397 |
+
voiced_threshod=0,
|
398 |
+
is_half=True,
|
399 |
+
):
|
400 |
+
super(SourceModuleHnNSF, self).__init__()
|
401 |
+
|
402 |
+
self.sine_amp = sine_amp
|
403 |
+
self.noise_std = add_noise_std
|
404 |
+
self.is_half = is_half
|
405 |
+
# to produce sine waveforms
|
406 |
+
self.l_sin_gen = SineGen(
|
407 |
+
sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
|
408 |
+
)
|
409 |
+
|
410 |
+
# to merge source harmonics into a single excitation
|
411 |
+
self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
|
412 |
+
self.l_tanh = torch.nn.Tanh()
|
413 |
+
|
414 |
+
def forward(self, x, upp=None):
|
415 |
+
sine_wavs, uv, _ = self.l_sin_gen(x, upp)
|
416 |
+
if self.is_half:
|
417 |
+
sine_wavs = sine_wavs.half()
|
418 |
+
sine_merge = self.l_tanh(self.l_linear(sine_wavs))
|
419 |
+
return sine_merge, None, None # noise, uv
|
420 |
+
|
421 |
+
|
422 |
+
class GeneratorNSF(torch.nn.Module):
|
423 |
+
def __init__(
|
424 |
+
self,
|
425 |
+
initial_channel,
|
426 |
+
resblock,
|
427 |
+
resblock_kernel_sizes,
|
428 |
+
resblock_dilation_sizes,
|
429 |
+
upsample_rates,
|
430 |
+
upsample_initial_channel,
|
431 |
+
upsample_kernel_sizes,
|
432 |
+
gin_channels,
|
433 |
+
sr,
|
434 |
+
is_half=False,
|
435 |
+
):
|
436 |
+
super(GeneratorNSF, self).__init__()
|
437 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
438 |
+
self.num_upsamples = len(upsample_rates)
|
439 |
+
|
440 |
+
self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
|
441 |
+
self.m_source = SourceModuleHnNSF(
|
442 |
+
sampling_rate=sr, harmonic_num=0, is_half=is_half
|
443 |
+
)
|
444 |
+
self.noise_convs = nn.ModuleList()
|
445 |
+
self.conv_pre = Conv1d(
|
446 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
447 |
+
)
|
448 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
449 |
+
|
450 |
+
self.ups = nn.ModuleList()
|
451 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
452 |
+
c_cur = upsample_initial_channel // (2 ** (i + 1))
|
453 |
+
self.ups.append(
|
454 |
+
weight_norm(
|
455 |
+
ConvTranspose1d(
|
456 |
+
upsample_initial_channel // (2**i),
|
457 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
458 |
+
k,
|
459 |
+
u,
|
460 |
+
padding=(k - u) // 2,
|
461 |
+
)
|
462 |
+
)
|
463 |
+
)
|
464 |
+
if i + 1 < len(upsample_rates):
|
465 |
+
stride_f0 = np.prod(upsample_rates[i + 1 :])
|
466 |
+
self.noise_convs.append(
|
467 |
+
Conv1d(
|
468 |
+
1,
|
469 |
+
c_cur,
|
470 |
+
kernel_size=stride_f0 * 2,
|
471 |
+
stride=stride_f0,
|
472 |
+
padding=stride_f0 // 2,
|
473 |
+
)
|
474 |
+
)
|
475 |
+
else:
|
476 |
+
self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
|
477 |
+
|
478 |
+
self.resblocks = nn.ModuleList()
|
479 |
+
for i in range(len(self.ups)):
|
480 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
481 |
+
for j, (k, d) in enumerate(
|
482 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
483 |
+
):
|
484 |
+
self.resblocks.append(resblock(ch, k, d))
|
485 |
+
|
486 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
487 |
+
self.ups.apply(init_weights)
|
488 |
+
|
489 |
+
if gin_channels != 0:
|
490 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
491 |
+
|
492 |
+
self.upp = np.prod(upsample_rates)
|
493 |
+
|
494 |
+
def forward(self, x, f0, g=None):
|
495 |
+
har_source, noi_source, uv = self.m_source(f0, self.upp)
|
496 |
+
har_source = har_source.transpose(1, 2)
|
497 |
+
x = self.conv_pre(x)
|
498 |
+
if g is not None:
|
499 |
+
x = x + self.cond(g)
|
500 |
+
|
501 |
+
for i in range(self.num_upsamples):
|
502 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
503 |
+
x = self.ups[i](x)
|
504 |
+
x_source = self.noise_convs[i](har_source)
|
505 |
+
x = x + x_source
|
506 |
+
xs = None
|
507 |
+
for j in range(self.num_kernels):
|
508 |
+
if xs is None:
|
509 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
510 |
+
else:
|
511 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
512 |
+
x = xs / self.num_kernels
|
513 |
+
x = F.leaky_relu(x)
|
514 |
+
x = self.conv_post(x)
|
515 |
+
x = torch.tanh(x)
|
516 |
+
return x
|
517 |
+
|
518 |
+
def remove_weight_norm(self):
|
519 |
+
for l in self.ups:
|
520 |
+
remove_weight_norm(l)
|
521 |
+
for l in self.resblocks:
|
522 |
+
l.remove_weight_norm()
|
523 |
+
|
524 |
+
|
525 |
+
sr2sr = {
|
526 |
+
"32k": 32000,
|
527 |
+
"40k": 40000,
|
528 |
+
"48k": 48000,
|
529 |
+
}
|
530 |
+
|
531 |
+
|
532 |
+
class SynthesizerTrnMs256NSFsid(nn.Module):
|
533 |
+
def __init__(
|
534 |
+
self,
|
535 |
+
spec_channels,
|
536 |
+
segment_size,
|
537 |
+
inter_channels,
|
538 |
+
hidden_channels,
|
539 |
+
filter_channels,
|
540 |
+
n_heads,
|
541 |
+
n_layers,
|
542 |
+
kernel_size,
|
543 |
+
p_dropout,
|
544 |
+
resblock,
|
545 |
+
resblock_kernel_sizes,
|
546 |
+
resblock_dilation_sizes,
|
547 |
+
upsample_rates,
|
548 |
+
upsample_initial_channel,
|
549 |
+
upsample_kernel_sizes,
|
550 |
+
spk_embed_dim,
|
551 |
+
gin_channels,
|
552 |
+
sr,
|
553 |
+
**kwargs
|
554 |
+
):
|
555 |
+
super().__init__()
|
556 |
+
if type(sr) == type("strr"):
|
557 |
+
sr = sr2sr[sr]
|
558 |
+
self.spec_channels = spec_channels
|
559 |
+
self.inter_channels = inter_channels
|
560 |
+
self.hidden_channels = hidden_channels
|
561 |
+
self.filter_channels = filter_channels
|
562 |
+
self.n_heads = n_heads
|
563 |
+
self.n_layers = n_layers
|
564 |
+
self.kernel_size = kernel_size
|
565 |
+
self.p_dropout = p_dropout
|
566 |
+
self.resblock = resblock
|
567 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
568 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
569 |
+
self.upsample_rates = upsample_rates
|
570 |
+
self.upsample_initial_channel = upsample_initial_channel
|
571 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
572 |
+
self.segment_size = segment_size
|
573 |
+
self.gin_channels = gin_channels
|
574 |
+
# self.hop_length = hop_length#
|
575 |
+
self.spk_embed_dim = spk_embed_dim
|
576 |
+
self.enc_p = TextEncoder256(
|
577 |
+
inter_channels,
|
578 |
+
hidden_channels,
|
579 |
+
filter_channels,
|
580 |
+
n_heads,
|
581 |
+
n_layers,
|
582 |
+
kernel_size,
|
583 |
+
p_dropout,
|
584 |
+
)
|
585 |
+
self.dec = GeneratorNSF(
|
586 |
+
inter_channels,
|
587 |
+
resblock,
|
588 |
+
resblock_kernel_sizes,
|
589 |
+
resblock_dilation_sizes,
|
590 |
+
upsample_rates,
|
591 |
+
upsample_initial_channel,
|
592 |
+
upsample_kernel_sizes,
|
593 |
+
gin_channels=gin_channels,
|
594 |
+
sr=sr,
|
595 |
+
is_half=kwargs["is_half"],
|
596 |
+
)
|
597 |
+
self.enc_q = PosteriorEncoder(
|
598 |
+
spec_channels,
|
599 |
+
inter_channels,
|
600 |
+
hidden_channels,
|
601 |
+
5,
|
602 |
+
1,
|
603 |
+
16,
|
604 |
+
gin_channels=gin_channels,
|
605 |
+
)
|
606 |
+
self.flow = ResidualCouplingBlock(
|
607 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
608 |
+
)
|
609 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
610 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
611 |
+
|
612 |
+
def remove_weight_norm(self):
|
613 |
+
self.dec.remove_weight_norm()
|
614 |
+
self.flow.remove_weight_norm()
|
615 |
+
self.enc_q.remove_weight_norm()
|
616 |
+
|
617 |
+
def forward(
|
618 |
+
self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
|
619 |
+
): # 这里ds是id,[bs,1]
|
620 |
+
# print(1,pitch.shape)#[bs,t]
|
621 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
622 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
623 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
624 |
+
z_p = self.flow(z, y_mask, g=g)
|
625 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
626 |
+
z, y_lengths, self.segment_size
|
627 |
+
)
|
628 |
+
# print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
|
629 |
+
pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
|
630 |
+
# print(-2,pitchf.shape,z_slice.shape)
|
631 |
+
o = self.dec(z_slice, pitchf, g=g)
|
632 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
633 |
+
|
634 |
+
def infer(self, phone, phone_lengths, pitch, nsff0, sid, rate=None):
|
635 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
636 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
637 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
638 |
+
if rate:
|
639 |
+
head = int(z_p.shape[2] * rate)
|
640 |
+
z_p = z_p[:, :, -head:]
|
641 |
+
x_mask = x_mask[:, :, -head:]
|
642 |
+
nsff0 = nsff0[:, -head:]
|
643 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
644 |
+
o = self.dec(z * x_mask, nsff0, g=g)
|
645 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
646 |
+
|
647 |
+
|
648 |
+
class SynthesizerTrnMs768NSFsid(nn.Module):
|
649 |
+
def __init__(
|
650 |
+
self,
|
651 |
+
spec_channels,
|
652 |
+
segment_size,
|
653 |
+
inter_channels,
|
654 |
+
hidden_channels,
|
655 |
+
filter_channels,
|
656 |
+
n_heads,
|
657 |
+
n_layers,
|
658 |
+
kernel_size,
|
659 |
+
p_dropout,
|
660 |
+
resblock,
|
661 |
+
resblock_kernel_sizes,
|
662 |
+
resblock_dilation_sizes,
|
663 |
+
upsample_rates,
|
664 |
+
upsample_initial_channel,
|
665 |
+
upsample_kernel_sizes,
|
666 |
+
spk_embed_dim,
|
667 |
+
gin_channels,
|
668 |
+
sr,
|
669 |
+
**kwargs
|
670 |
+
):
|
671 |
+
super().__init__()
|
672 |
+
if type(sr) == type("strr"):
|
673 |
+
sr = sr2sr[sr]
|
674 |
+
self.spec_channels = spec_channels
|
675 |
+
self.inter_channels = inter_channels
|
676 |
+
self.hidden_channels = hidden_channels
|
677 |
+
self.filter_channels = filter_channels
|
678 |
+
self.n_heads = n_heads
|
679 |
+
self.n_layers = n_layers
|
680 |
+
self.kernel_size = kernel_size
|
681 |
+
self.p_dropout = p_dropout
|
682 |
+
self.resblock = resblock
|
683 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
684 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
685 |
+
self.upsample_rates = upsample_rates
|
686 |
+
self.upsample_initial_channel = upsample_initial_channel
|
687 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
688 |
+
self.segment_size = segment_size
|
689 |
+
self.gin_channels = gin_channels
|
690 |
+
# self.hop_length = hop_length#
|
691 |
+
self.spk_embed_dim = spk_embed_dim
|
692 |
+
self.enc_p = TextEncoder768(
|
693 |
+
inter_channels,
|
694 |
+
hidden_channels,
|
695 |
+
filter_channels,
|
696 |
+
n_heads,
|
697 |
+
n_layers,
|
698 |
+
kernel_size,
|
699 |
+
p_dropout,
|
700 |
+
)
|
701 |
+
self.dec = GeneratorNSF(
|
702 |
+
inter_channels,
|
703 |
+
resblock,
|
704 |
+
resblock_kernel_sizes,
|
705 |
+
resblock_dilation_sizes,
|
706 |
+
upsample_rates,
|
707 |
+
upsample_initial_channel,
|
708 |
+
upsample_kernel_sizes,
|
709 |
+
gin_channels=gin_channels,
|
710 |
+
sr=sr,
|
711 |
+
is_half=kwargs["is_half"],
|
712 |
+
)
|
713 |
+
self.enc_q = PosteriorEncoder(
|
714 |
+
spec_channels,
|
715 |
+
inter_channels,
|
716 |
+
hidden_channels,
|
717 |
+
5,
|
718 |
+
1,
|
719 |
+
16,
|
720 |
+
gin_channels=gin_channels,
|
721 |
+
)
|
722 |
+
self.flow = ResidualCouplingBlock(
|
723 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
724 |
+
)
|
725 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
726 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
727 |
+
|
728 |
+
def remove_weight_norm(self):
|
729 |
+
self.dec.remove_weight_norm()
|
730 |
+
self.flow.remove_weight_norm()
|
731 |
+
self.enc_q.remove_weight_norm()
|
732 |
+
|
733 |
+
def forward(
|
734 |
+
self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
|
735 |
+
): # 这里ds是id,[bs,1]
|
736 |
+
# print(1,pitch.shape)#[bs,t]
|
737 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
738 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
739 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
740 |
+
z_p = self.flow(z, y_mask, g=g)
|
741 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
742 |
+
z, y_lengths, self.segment_size
|
743 |
+
)
|
744 |
+
# print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
|
745 |
+
pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
|
746 |
+
# print(-2,pitchf.shape,z_slice.shape)
|
747 |
+
o = self.dec(z_slice, pitchf, g=g)
|
748 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
749 |
+
|
750 |
+
def infer(self, phone, phone_lengths, pitch, nsff0, sid, rate=None):
|
751 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
752 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
753 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
754 |
+
if rate:
|
755 |
+
head = int(z_p.shape[2] * rate)
|
756 |
+
z_p = z_p[:, :, -head:]
|
757 |
+
x_mask = x_mask[:, :, -head:]
|
758 |
+
nsff0 = nsff0[:, -head:]
|
759 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
760 |
+
o = self.dec(z * x_mask, nsff0, g=g)
|
761 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
762 |
+
|
763 |
+
|
764 |
+
class SynthesizerTrnMs256NSFsid_nono(nn.Module):
|
765 |
+
def __init__(
|
766 |
+
self,
|
767 |
+
spec_channels,
|
768 |
+
segment_size,
|
769 |
+
inter_channels,
|
770 |
+
hidden_channels,
|
771 |
+
filter_channels,
|
772 |
+
n_heads,
|
773 |
+
n_layers,
|
774 |
+
kernel_size,
|
775 |
+
p_dropout,
|
776 |
+
resblock,
|
777 |
+
resblock_kernel_sizes,
|
778 |
+
resblock_dilation_sizes,
|
779 |
+
upsample_rates,
|
780 |
+
upsample_initial_channel,
|
781 |
+
upsample_kernel_sizes,
|
782 |
+
spk_embed_dim,
|
783 |
+
gin_channels,
|
784 |
+
sr=None,
|
785 |
+
**kwargs
|
786 |
+
):
|
787 |
+
super().__init__()
|
788 |
+
self.spec_channels = spec_channels
|
789 |
+
self.inter_channels = inter_channels
|
790 |
+
self.hidden_channels = hidden_channels
|
791 |
+
self.filter_channels = filter_channels
|
792 |
+
self.n_heads = n_heads
|
793 |
+
self.n_layers = n_layers
|
794 |
+
self.kernel_size = kernel_size
|
795 |
+
self.p_dropout = p_dropout
|
796 |
+
self.resblock = resblock
|
797 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
798 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
799 |
+
self.upsample_rates = upsample_rates
|
800 |
+
self.upsample_initial_channel = upsample_initial_channel
|
801 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
802 |
+
self.segment_size = segment_size
|
803 |
+
self.gin_channels = gin_channels
|
804 |
+
# self.hop_length = hop_length#
|
805 |
+
self.spk_embed_dim = spk_embed_dim
|
806 |
+
self.enc_p = TextEncoder256(
|
807 |
+
inter_channels,
|
808 |
+
hidden_channels,
|
809 |
+
filter_channels,
|
810 |
+
n_heads,
|
811 |
+
n_layers,
|
812 |
+
kernel_size,
|
813 |
+
p_dropout,
|
814 |
+
f0=False,
|
815 |
+
)
|
816 |
+
self.dec = Generator(
|
817 |
+
inter_channels,
|
818 |
+
resblock,
|
819 |
+
resblock_kernel_sizes,
|
820 |
+
resblock_dilation_sizes,
|
821 |
+
upsample_rates,
|
822 |
+
upsample_initial_channel,
|
823 |
+
upsample_kernel_sizes,
|
824 |
+
gin_channels=gin_channels,
|
825 |
+
)
|
826 |
+
self.enc_q = PosteriorEncoder(
|
827 |
+
spec_channels,
|
828 |
+
inter_channels,
|
829 |
+
hidden_channels,
|
830 |
+
5,
|
831 |
+
1,
|
832 |
+
16,
|
833 |
+
gin_channels=gin_channels,
|
834 |
+
)
|
835 |
+
self.flow = ResidualCouplingBlock(
|
836 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
837 |
+
)
|
838 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
839 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
840 |
+
|
841 |
+
def remove_weight_norm(self):
|
842 |
+
self.dec.remove_weight_norm()
|
843 |
+
self.flow.remove_weight_norm()
|
844 |
+
self.enc_q.remove_weight_norm()
|
845 |
+
|
846 |
+
def forward(self, phone, phone_lengths, y, y_lengths, ds): # 这里ds是id,[bs,1]
|
847 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
848 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
849 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
850 |
+
z_p = self.flow(z, y_mask, g=g)
|
851 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
852 |
+
z, y_lengths, self.segment_size
|
853 |
+
)
|
854 |
+
o = self.dec(z_slice, g=g)
|
855 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
856 |
+
|
857 |
+
def infer(self, phone, phone_lengths, sid, rate=None):
|
858 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
859 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
860 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
861 |
+
if rate:
|
862 |
+
head = int(z_p.shape[2] * rate)
|
863 |
+
z_p = z_p[:, :, -head:]
|
864 |
+
x_mask = x_mask[:, :, -head:]
|
865 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
866 |
+
o = self.dec(z * x_mask, g=g)
|
867 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
868 |
+
|
869 |
+
|
870 |
+
class SynthesizerTrnMs768NSFsid_nono(nn.Module):
|
871 |
+
def __init__(
|
872 |
+
self,
|
873 |
+
spec_channels,
|
874 |
+
segment_size,
|
875 |
+
inter_channels,
|
876 |
+
hidden_channels,
|
877 |
+
filter_channels,
|
878 |
+
n_heads,
|
879 |
+
n_layers,
|
880 |
+
kernel_size,
|
881 |
+
p_dropout,
|
882 |
+
resblock,
|
883 |
+
resblock_kernel_sizes,
|
884 |
+
resblock_dilation_sizes,
|
885 |
+
upsample_rates,
|
886 |
+
upsample_initial_channel,
|
887 |
+
upsample_kernel_sizes,
|
888 |
+
spk_embed_dim,
|
889 |
+
gin_channels,
|
890 |
+
sr=None,
|
891 |
+
**kwargs
|
892 |
+
):
|
893 |
+
super().__init__()
|
894 |
+
self.spec_channels = spec_channels
|
895 |
+
self.inter_channels = inter_channels
|
896 |
+
self.hidden_channels = hidden_channels
|
897 |
+
self.filter_channels = filter_channels
|
898 |
+
self.n_heads = n_heads
|
899 |
+
self.n_layers = n_layers
|
900 |
+
self.kernel_size = kernel_size
|
901 |
+
self.p_dropout = p_dropout
|
902 |
+
self.resblock = resblock
|
903 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
904 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
905 |
+
self.upsample_rates = upsample_rates
|
906 |
+
self.upsample_initial_channel = upsample_initial_channel
|
907 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
908 |
+
self.segment_size = segment_size
|
909 |
+
self.gin_channels = gin_channels
|
910 |
+
# self.hop_length = hop_length#
|
911 |
+
self.spk_embed_dim = spk_embed_dim
|
912 |
+
self.enc_p = TextEncoder768(
|
913 |
+
inter_channels,
|
914 |
+
hidden_channels,
|
915 |
+
filter_channels,
|
916 |
+
n_heads,
|
917 |
+
n_layers,
|
918 |
+
kernel_size,
|
919 |
+
p_dropout,
|
920 |
+
f0=False,
|
921 |
+
)
|
922 |
+
self.dec = Generator(
|
923 |
+
inter_channels,
|
924 |
+
resblock,
|
925 |
+
resblock_kernel_sizes,
|
926 |
+
resblock_dilation_sizes,
|
927 |
+
upsample_rates,
|
928 |
+
upsample_initial_channel,
|
929 |
+
upsample_kernel_sizes,
|
930 |
+
gin_channels=gin_channels,
|
931 |
+
)
|
932 |
+
self.enc_q = PosteriorEncoder(
|
933 |
+
spec_channels,
|
934 |
+
inter_channels,
|
935 |
+
hidden_channels,
|
936 |
+
5,
|
937 |
+
1,
|
938 |
+
16,
|
939 |
+
gin_channels=gin_channels,
|
940 |
+
)
|
941 |
+
self.flow = ResidualCouplingBlock(
|
942 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
943 |
+
)
|
944 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
945 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
946 |
+
|
947 |
+
def remove_weight_norm(self):
|
948 |
+
self.dec.remove_weight_norm()
|
949 |
+
self.flow.remove_weight_norm()
|
950 |
+
self.enc_q.remove_weight_norm()
|
951 |
+
|
952 |
+
def forward(self, phone, phone_lengths, y, y_lengths, ds): # 这里ds是id,[bs,1]
|
953 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
954 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
955 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
956 |
+
z_p = self.flow(z, y_mask, g=g)
|
957 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
958 |
+
z, y_lengths, self.segment_size
|
959 |
+
)
|
960 |
+
o = self.dec(z_slice, g=g)
|
961 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
962 |
+
|
963 |
+
def infer(self, phone, phone_lengths, sid, rate=None):
|
964 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
965 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
966 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
967 |
+
if rate:
|
968 |
+
head = int(z_p.shape[2] * rate)
|
969 |
+
z_p = z_p[:, :, -head:]
|
970 |
+
x_mask = x_mask[:, :, -head:]
|
971 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
972 |
+
o = self.dec(z * x_mask, g=g)
|
973 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
974 |
+
|
975 |
+
|
976 |
+
class MultiPeriodDiscriminator(torch.nn.Module):
|
977 |
+
def __init__(self, use_spectral_norm=False):
|
978 |
+
super(MultiPeriodDiscriminator, self).__init__()
|
979 |
+
periods = [2, 3, 5, 7, 11, 17]
|
980 |
+
# periods = [3, 5, 7, 11, 17, 23, 37]
|
981 |
+
|
982 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
983 |
+
discs = discs + [
|
984 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
985 |
+
]
|
986 |
+
self.discriminators = nn.ModuleList(discs)
|
987 |
+
|
988 |
+
def forward(self, y, y_hat):
|
989 |
+
y_d_rs = [] #
|
990 |
+
y_d_gs = []
|
991 |
+
fmap_rs = []
|
992 |
+
fmap_gs = []
|
993 |
+
for i, d in enumerate(self.discriminators):
|
994 |
+
y_d_r, fmap_r = d(y)
|
995 |
+
y_d_g, fmap_g = d(y_hat)
|
996 |
+
# for j in range(len(fmap_r)):
|
997 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
998 |
+
y_d_rs.append(y_d_r)
|
999 |
+
y_d_gs.append(y_d_g)
|
1000 |
+
fmap_rs.append(fmap_r)
|
1001 |
+
fmap_gs.append(fmap_g)
|
1002 |
+
|
1003 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
1004 |
+
|
1005 |
+
|
1006 |
+
class MultiPeriodDiscriminatorV2(torch.nn.Module):
|
1007 |
+
def __init__(self, use_spectral_norm=False):
|
1008 |
+
super(MultiPeriodDiscriminatorV2, self).__init__()
|
1009 |
+
# periods = [2, 3, 5, 7, 11, 17]
|
1010 |
+
periods = [2, 3, 5, 7, 11, 17, 23, 37]
|
1011 |
+
|
1012 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
1013 |
+
discs = discs + [
|
1014 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
1015 |
+
]
|
1016 |
+
self.discriminators = nn.ModuleList(discs)
|
1017 |
+
|
1018 |
+
def forward(self, y, y_hat):
|
1019 |
+
y_d_rs = [] #
|
1020 |
+
y_d_gs = []
|
1021 |
+
fmap_rs = []
|
1022 |
+
fmap_gs = []
|
1023 |
+
for i, d in enumerate(self.discriminators):
|
1024 |
+
y_d_r, fmap_r = d(y)
|
1025 |
+
y_d_g, fmap_g = d(y_hat)
|
1026 |
+
# for j in range(len(fmap_r)):
|
1027 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
1028 |
+
y_d_rs.append(y_d_r)
|
1029 |
+
y_d_gs.append(y_d_g)
|
1030 |
+
fmap_rs.append(fmap_r)
|
1031 |
+
fmap_gs.append(fmap_g)
|
1032 |
+
|
1033 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
1034 |
+
|
1035 |
+
|
1036 |
+
class DiscriminatorS(torch.nn.Module):
|
1037 |
+
def __init__(self, use_spectral_norm=False):
|
1038 |
+
super(DiscriminatorS, self).__init__()
|
1039 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
1040 |
+
self.convs = nn.ModuleList(
|
1041 |
+
[
|
1042 |
+
norm_f(Conv1d(1, 16, 15, 1, padding=7)),
|
1043 |
+
norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
|
1044 |
+
norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
|
1045 |
+
norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
|
1046 |
+
norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
|
1047 |
+
norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
|
1048 |
+
]
|
1049 |
+
)
|
1050 |
+
self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
|
1051 |
+
|
1052 |
+
def forward(self, x):
|
1053 |
+
fmap = []
|
1054 |
+
|
1055 |
+
for l in self.convs:
|
1056 |
+
x = l(x)
|
1057 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
1058 |
+
fmap.append(x)
|
1059 |
+
x = self.conv_post(x)
|
1060 |
+
fmap.append(x)
|
1061 |
+
x = torch.flatten(x, 1, -1)
|
1062 |
+
|
1063 |
+
return x, fmap
|
1064 |
+
|
1065 |
+
|
1066 |
+
class DiscriminatorP(torch.nn.Module):
|
1067 |
+
def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
|
1068 |
+
super(DiscriminatorP, self).__init__()
|
1069 |
+
self.period = period
|
1070 |
+
self.use_spectral_norm = use_spectral_norm
|
1071 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
1072 |
+
self.convs = nn.ModuleList(
|
1073 |
+
[
|
1074 |
+
norm_f(
|
1075 |
+
Conv2d(
|
1076 |
+
1,
|
1077 |
+
32,
|
1078 |
+
(kernel_size, 1),
|
1079 |
+
(stride, 1),
|
1080 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1081 |
+
)
|
1082 |
+
),
|
1083 |
+
norm_f(
|
1084 |
+
Conv2d(
|
1085 |
+
32,
|
1086 |
+
128,
|
1087 |
+
(kernel_size, 1),
|
1088 |
+
(stride, 1),
|
1089 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1090 |
+
)
|
1091 |
+
),
|
1092 |
+
norm_f(
|
1093 |
+
Conv2d(
|
1094 |
+
128,
|
1095 |
+
512,
|
1096 |
+
(kernel_size, 1),
|
1097 |
+
(stride, 1),
|
1098 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1099 |
+
)
|
1100 |
+
),
|
1101 |
+
norm_f(
|
1102 |
+
Conv2d(
|
1103 |
+
512,
|
1104 |
+
1024,
|
1105 |
+
(kernel_size, 1),
|
1106 |
+
(stride, 1),
|
1107 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1108 |
+
)
|
1109 |
+
),
|
1110 |
+
norm_f(
|
1111 |
+
Conv2d(
|
1112 |
+
1024,
|
1113 |
+
1024,
|
1114 |
+
(kernel_size, 1),
|
1115 |
+
1,
|
1116 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1117 |
+
)
|
1118 |
+
),
|
1119 |
+
]
|
1120 |
+
)
|
1121 |
+
self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
|
1122 |
+
|
1123 |
+
def forward(self, x):
|
1124 |
+
fmap = []
|
1125 |
+
|
1126 |
+
# 1d to 2d
|
1127 |
+
b, c, t = x.shape
|
1128 |
+
if t % self.period != 0: # pad first
|
1129 |
+
n_pad = self.period - (t % self.period)
|
1130 |
+
x = F.pad(x, (0, n_pad), "reflect")
|
1131 |
+
t = t + n_pad
|
1132 |
+
x = x.view(b, c, t // self.period, self.period)
|
1133 |
+
|
1134 |
+
for l in self.convs:
|
1135 |
+
x = l(x)
|
1136 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
1137 |
+
fmap.append(x)
|
1138 |
+
x = self.conv_post(x)
|
1139 |
+
fmap.append(x)
|
1140 |
+
x = torch.flatten(x, 1, -1)
|
1141 |
+
|
1142 |
+
return x, fmap
|
lib/infer_pack/models_dml.py
ADDED
@@ -0,0 +1,1124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math, pdb, os
|
2 |
+
from time import time as ttime
|
3 |
+
import torch
|
4 |
+
from torch import nn
|
5 |
+
from torch.nn import functional as F
|
6 |
+
from lib.infer_pack import modules
|
7 |
+
from lib.infer_pack import attentions
|
8 |
+
from lib.infer_pack import commons
|
9 |
+
from lib.infer_pack.commons import init_weights, get_padding
|
10 |
+
from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
|
11 |
+
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
|
12 |
+
from lib.infer_pack.commons import init_weights
|
13 |
+
import numpy as np
|
14 |
+
from lib.infer_pack import commons
|
15 |
+
|
16 |
+
|
17 |
+
class TextEncoder256(nn.Module):
|
18 |
+
def __init__(
|
19 |
+
self,
|
20 |
+
out_channels,
|
21 |
+
hidden_channels,
|
22 |
+
filter_channels,
|
23 |
+
n_heads,
|
24 |
+
n_layers,
|
25 |
+
kernel_size,
|
26 |
+
p_dropout,
|
27 |
+
f0=True,
|
28 |
+
):
|
29 |
+
super().__init__()
|
30 |
+
self.out_channels = out_channels
|
31 |
+
self.hidden_channels = hidden_channels
|
32 |
+
self.filter_channels = filter_channels
|
33 |
+
self.n_heads = n_heads
|
34 |
+
self.n_layers = n_layers
|
35 |
+
self.kernel_size = kernel_size
|
36 |
+
self.p_dropout = p_dropout
|
37 |
+
self.emb_phone = nn.Linear(256, hidden_channels)
|
38 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
39 |
+
if f0 == True:
|
40 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
41 |
+
self.encoder = attentions.Encoder(
|
42 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
43 |
+
)
|
44 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
45 |
+
|
46 |
+
def forward(self, phone, pitch, lengths):
|
47 |
+
if pitch == None:
|
48 |
+
x = self.emb_phone(phone)
|
49 |
+
else:
|
50 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
51 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
52 |
+
x = self.lrelu(x)
|
53 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
54 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
55 |
+
x.dtype
|
56 |
+
)
|
57 |
+
x = self.encoder(x * x_mask, x_mask)
|
58 |
+
stats = self.proj(x) * x_mask
|
59 |
+
|
60 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
61 |
+
return m, logs, x_mask
|
62 |
+
|
63 |
+
|
64 |
+
class TextEncoder768(nn.Module):
|
65 |
+
def __init__(
|
66 |
+
self,
|
67 |
+
out_channels,
|
68 |
+
hidden_channels,
|
69 |
+
filter_channels,
|
70 |
+
n_heads,
|
71 |
+
n_layers,
|
72 |
+
kernel_size,
|
73 |
+
p_dropout,
|
74 |
+
f0=True,
|
75 |
+
):
|
76 |
+
super().__init__()
|
77 |
+
self.out_channels = out_channels
|
78 |
+
self.hidden_channels = hidden_channels
|
79 |
+
self.filter_channels = filter_channels
|
80 |
+
self.n_heads = n_heads
|
81 |
+
self.n_layers = n_layers
|
82 |
+
self.kernel_size = kernel_size
|
83 |
+
self.p_dropout = p_dropout
|
84 |
+
self.emb_phone = nn.Linear(768, hidden_channels)
|
85 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
86 |
+
if f0 == True:
|
87 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
88 |
+
self.encoder = attentions.Encoder(
|
89 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
90 |
+
)
|
91 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
92 |
+
|
93 |
+
def forward(self, phone, pitch, lengths):
|
94 |
+
if pitch == None:
|
95 |
+
x = self.emb_phone(phone)
|
96 |
+
else:
|
97 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
98 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
99 |
+
x = self.lrelu(x)
|
100 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
101 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
102 |
+
x.dtype
|
103 |
+
)
|
104 |
+
x = self.encoder(x * x_mask, x_mask)
|
105 |
+
stats = self.proj(x) * x_mask
|
106 |
+
|
107 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
108 |
+
return m, logs, x_mask
|
109 |
+
|
110 |
+
|
111 |
+
class ResidualCouplingBlock(nn.Module):
|
112 |
+
def __init__(
|
113 |
+
self,
|
114 |
+
channels,
|
115 |
+
hidden_channels,
|
116 |
+
kernel_size,
|
117 |
+
dilation_rate,
|
118 |
+
n_layers,
|
119 |
+
n_flows=4,
|
120 |
+
gin_channels=0,
|
121 |
+
):
|
122 |
+
super().__init__()
|
123 |
+
self.channels = channels
|
124 |
+
self.hidden_channels = hidden_channels
|
125 |
+
self.kernel_size = kernel_size
|
126 |
+
self.dilation_rate = dilation_rate
|
127 |
+
self.n_layers = n_layers
|
128 |
+
self.n_flows = n_flows
|
129 |
+
self.gin_channels = gin_channels
|
130 |
+
|
131 |
+
self.flows = nn.ModuleList()
|
132 |
+
for i in range(n_flows):
|
133 |
+
self.flows.append(
|
134 |
+
modules.ResidualCouplingLayer(
|
135 |
+
channels,
|
136 |
+
hidden_channels,
|
137 |
+
kernel_size,
|
138 |
+
dilation_rate,
|
139 |
+
n_layers,
|
140 |
+
gin_channels=gin_channels,
|
141 |
+
mean_only=True,
|
142 |
+
)
|
143 |
+
)
|
144 |
+
self.flows.append(modules.Flip())
|
145 |
+
|
146 |
+
def forward(self, x, x_mask, g=None, reverse=False):
|
147 |
+
if not reverse:
|
148 |
+
for flow in self.flows:
|
149 |
+
x, _ = flow(x, x_mask, g=g, reverse=reverse)
|
150 |
+
else:
|
151 |
+
for flow in reversed(self.flows):
|
152 |
+
x = flow(x, x_mask, g=g, reverse=reverse)
|
153 |
+
return x
|
154 |
+
|
155 |
+
def remove_weight_norm(self):
|
156 |
+
for i in range(self.n_flows):
|
157 |
+
self.flows[i * 2].remove_weight_norm()
|
158 |
+
|
159 |
+
|
160 |
+
class PosteriorEncoder(nn.Module):
|
161 |
+
def __init__(
|
162 |
+
self,
|
163 |
+
in_channels,
|
164 |
+
out_channels,
|
165 |
+
hidden_channels,
|
166 |
+
kernel_size,
|
167 |
+
dilation_rate,
|
168 |
+
n_layers,
|
169 |
+
gin_channels=0,
|
170 |
+
):
|
171 |
+
super().__init__()
|
172 |
+
self.in_channels = in_channels
|
173 |
+
self.out_channels = out_channels
|
174 |
+
self.hidden_channels = hidden_channels
|
175 |
+
self.kernel_size = kernel_size
|
176 |
+
self.dilation_rate = dilation_rate
|
177 |
+
self.n_layers = n_layers
|
178 |
+
self.gin_channels = gin_channels
|
179 |
+
|
180 |
+
self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
|
181 |
+
self.enc = modules.WN(
|
182 |
+
hidden_channels,
|
183 |
+
kernel_size,
|
184 |
+
dilation_rate,
|
185 |
+
n_layers,
|
186 |
+
gin_channels=gin_channels,
|
187 |
+
)
|
188 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
189 |
+
|
190 |
+
def forward(self, x, x_lengths, g=None):
|
191 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
|
192 |
+
x.dtype
|
193 |
+
)
|
194 |
+
x = self.pre(x) * x_mask
|
195 |
+
x = self.enc(x, x_mask, g=g)
|
196 |
+
stats = self.proj(x) * x_mask
|
197 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
198 |
+
z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
|
199 |
+
return z, m, logs, x_mask
|
200 |
+
|
201 |
+
def remove_weight_norm(self):
|
202 |
+
self.enc.remove_weight_norm()
|
203 |
+
|
204 |
+
|
205 |
+
class Generator(torch.nn.Module):
|
206 |
+
def __init__(
|
207 |
+
self,
|
208 |
+
initial_channel,
|
209 |
+
resblock,
|
210 |
+
resblock_kernel_sizes,
|
211 |
+
resblock_dilation_sizes,
|
212 |
+
upsample_rates,
|
213 |
+
upsample_initial_channel,
|
214 |
+
upsample_kernel_sizes,
|
215 |
+
gin_channels=0,
|
216 |
+
):
|
217 |
+
super(Generator, self).__init__()
|
218 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
219 |
+
self.num_upsamples = len(upsample_rates)
|
220 |
+
self.conv_pre = Conv1d(
|
221 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
222 |
+
)
|
223 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
224 |
+
|
225 |
+
self.ups = nn.ModuleList()
|
226 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
227 |
+
self.ups.append(
|
228 |
+
weight_norm(
|
229 |
+
ConvTranspose1d(
|
230 |
+
upsample_initial_channel // (2**i),
|
231 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
232 |
+
k,
|
233 |
+
u,
|
234 |
+
padding=(k - u) // 2,
|
235 |
+
)
|
236 |
+
)
|
237 |
+
)
|
238 |
+
|
239 |
+
self.resblocks = nn.ModuleList()
|
240 |
+
for i in range(len(self.ups)):
|
241 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
242 |
+
for j, (k, d) in enumerate(
|
243 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
244 |
+
):
|
245 |
+
self.resblocks.append(resblock(ch, k, d))
|
246 |
+
|
247 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
248 |
+
self.ups.apply(init_weights)
|
249 |
+
|
250 |
+
if gin_channels != 0:
|
251 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
252 |
+
|
253 |
+
def forward(self, x, g=None):
|
254 |
+
x = self.conv_pre(x)
|
255 |
+
if g is not None:
|
256 |
+
x = x + self.cond(g)
|
257 |
+
|
258 |
+
for i in range(self.num_upsamples):
|
259 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
260 |
+
x = self.ups[i](x)
|
261 |
+
xs = None
|
262 |
+
for j in range(self.num_kernels):
|
263 |
+
if xs is None:
|
264 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
265 |
+
else:
|
266 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
267 |
+
x = xs / self.num_kernels
|
268 |
+
x = F.leaky_relu(x)
|
269 |
+
x = self.conv_post(x)
|
270 |
+
x = torch.tanh(x)
|
271 |
+
|
272 |
+
return x
|
273 |
+
|
274 |
+
def remove_weight_norm(self):
|
275 |
+
for l in self.ups:
|
276 |
+
remove_weight_norm(l)
|
277 |
+
for l in self.resblocks:
|
278 |
+
l.remove_weight_norm()
|
279 |
+
|
280 |
+
|
281 |
+
class SineGen(torch.nn.Module):
|
282 |
+
"""Definition of sine generator
|
283 |
+
SineGen(samp_rate, harmonic_num = 0,
|
284 |
+
sine_amp = 0.1, noise_std = 0.003,
|
285 |
+
voiced_threshold = 0,
|
286 |
+
flag_for_pulse=False)
|
287 |
+
samp_rate: sampling rate in Hz
|
288 |
+
harmonic_num: number of harmonic overtones (default 0)
|
289 |
+
sine_amp: amplitude of sine-wavefrom (default 0.1)
|
290 |
+
noise_std: std of Gaussian noise (default 0.003)
|
291 |
+
voiced_thoreshold: F0 threshold for U/V classification (default 0)
|
292 |
+
flag_for_pulse: this SinGen is used inside PulseGen (default False)
|
293 |
+
Note: when flag_for_pulse is True, the first time step of a voiced
|
294 |
+
segment is always sin(np.pi) or cos(0)
|
295 |
+
"""
|
296 |
+
|
297 |
+
def __init__(
|
298 |
+
self,
|
299 |
+
samp_rate,
|
300 |
+
harmonic_num=0,
|
301 |
+
sine_amp=0.1,
|
302 |
+
noise_std=0.003,
|
303 |
+
voiced_threshold=0,
|
304 |
+
flag_for_pulse=False,
|
305 |
+
):
|
306 |
+
super(SineGen, self).__init__()
|
307 |
+
self.sine_amp = sine_amp
|
308 |
+
self.noise_std = noise_std
|
309 |
+
self.harmonic_num = harmonic_num
|
310 |
+
self.dim = self.harmonic_num + 1
|
311 |
+
self.sampling_rate = samp_rate
|
312 |
+
self.voiced_threshold = voiced_threshold
|
313 |
+
|
314 |
+
def _f02uv(self, f0):
|
315 |
+
# generate uv signal
|
316 |
+
uv = torch.ones_like(f0)
|
317 |
+
uv = uv * (f0 > self.voiced_threshold)
|
318 |
+
return uv.float()
|
319 |
+
|
320 |
+
def forward(self, f0, upp):
|
321 |
+
"""sine_tensor, uv = forward(f0)
|
322 |
+
input F0: tensor(batchsize=1, length, dim=1)
|
323 |
+
f0 for unvoiced steps should be 0
|
324 |
+
output sine_tensor: tensor(batchsize=1, length, dim)
|
325 |
+
output uv: tensor(batchsize=1, length, 1)
|
326 |
+
"""
|
327 |
+
with torch.no_grad():
|
328 |
+
f0 = f0[:, None].transpose(1, 2)
|
329 |
+
f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
|
330 |
+
# fundamental component
|
331 |
+
f0_buf[:, :, 0] = f0[:, :, 0]
|
332 |
+
for idx in np.arange(self.harmonic_num):
|
333 |
+
f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
|
334 |
+
idx + 2
|
335 |
+
) # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
|
336 |
+
rad_values = (f0_buf / self.sampling_rate) % 1 ###%1意味着n_har的乘积无法后处理优化
|
337 |
+
rand_ini = torch.rand(
|
338 |
+
f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
|
339 |
+
)
|
340 |
+
rand_ini[:, 0] = 0
|
341 |
+
rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
|
342 |
+
tmp_over_one = torch.cumsum(rad_values, 1) # % 1 #####%1意味着后面的cumsum无法再优化
|
343 |
+
tmp_over_one *= upp
|
344 |
+
tmp_over_one = F.interpolate(
|
345 |
+
tmp_over_one.transpose(2, 1),
|
346 |
+
scale_factor=upp,
|
347 |
+
mode="linear",
|
348 |
+
align_corners=True,
|
349 |
+
).transpose(2, 1)
|
350 |
+
rad_values = F.interpolate(
|
351 |
+
rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
|
352 |
+
).transpose(
|
353 |
+
2, 1
|
354 |
+
) #######
|
355 |
+
tmp_over_one %= 1
|
356 |
+
tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
|
357 |
+
cumsum_shift = torch.zeros_like(rad_values)
|
358 |
+
cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
|
359 |
+
sine_waves = torch.sin(
|
360 |
+
torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
|
361 |
+
)
|
362 |
+
sine_waves = sine_waves * self.sine_amp
|
363 |
+
uv = self._f02uv(f0)
|
364 |
+
uv = F.interpolate(
|
365 |
+
uv.transpose(2, 1), scale_factor=upp, mode="nearest"
|
366 |
+
).transpose(2, 1)
|
367 |
+
noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
|
368 |
+
noise = noise_amp * torch.randn_like(sine_waves)
|
369 |
+
sine_waves = sine_waves * uv + noise
|
370 |
+
return sine_waves, uv, noise
|
371 |
+
|
372 |
+
|
373 |
+
class SourceModuleHnNSF(torch.nn.Module):
|
374 |
+
"""SourceModule for hn-nsf
|
375 |
+
SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
|
376 |
+
add_noise_std=0.003, voiced_threshod=0)
|
377 |
+
sampling_rate: sampling_rate in Hz
|
378 |
+
harmonic_num: number of harmonic above F0 (default: 0)
|
379 |
+
sine_amp: amplitude of sine source signal (default: 0.1)
|
380 |
+
add_noise_std: std of additive Gaussian noise (default: 0.003)
|
381 |
+
note that amplitude of noise in unvoiced is decided
|
382 |
+
by sine_amp
|
383 |
+
voiced_threshold: threhold to set U/V given F0 (default: 0)
|
384 |
+
Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
|
385 |
+
F0_sampled (batchsize, length, 1)
|
386 |
+
Sine_source (batchsize, length, 1)
|
387 |
+
noise_source (batchsize, length 1)
|
388 |
+
uv (batchsize, length, 1)
|
389 |
+
"""
|
390 |
+
|
391 |
+
def __init__(
|
392 |
+
self,
|
393 |
+
sampling_rate,
|
394 |
+
harmonic_num=0,
|
395 |
+
sine_amp=0.1,
|
396 |
+
add_noise_std=0.003,
|
397 |
+
voiced_threshod=0,
|
398 |
+
is_half=True,
|
399 |
+
):
|
400 |
+
super(SourceModuleHnNSF, self).__init__()
|
401 |
+
|
402 |
+
self.sine_amp = sine_amp
|
403 |
+
self.noise_std = add_noise_std
|
404 |
+
self.is_half = is_half
|
405 |
+
# to produce sine waveforms
|
406 |
+
self.l_sin_gen = SineGen(
|
407 |
+
sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
|
408 |
+
)
|
409 |
+
|
410 |
+
# to merge source harmonics into a single excitation
|
411 |
+
self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
|
412 |
+
self.l_tanh = torch.nn.Tanh()
|
413 |
+
|
414 |
+
def forward(self, x, upp=None):
|
415 |
+
sine_wavs, uv, _ = self.l_sin_gen(x, upp)
|
416 |
+
if self.is_half:
|
417 |
+
sine_wavs = sine_wavs.half()
|
418 |
+
sine_merge = self.l_tanh(self.l_linear(sine_wavs))
|
419 |
+
return sine_merge, None, None # noise, uv
|
420 |
+
|
421 |
+
|
422 |
+
class GeneratorNSF(torch.nn.Module):
|
423 |
+
def __init__(
|
424 |
+
self,
|
425 |
+
initial_channel,
|
426 |
+
resblock,
|
427 |
+
resblock_kernel_sizes,
|
428 |
+
resblock_dilation_sizes,
|
429 |
+
upsample_rates,
|
430 |
+
upsample_initial_channel,
|
431 |
+
upsample_kernel_sizes,
|
432 |
+
gin_channels,
|
433 |
+
sr,
|
434 |
+
is_half=False,
|
435 |
+
):
|
436 |
+
super(GeneratorNSF, self).__init__()
|
437 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
438 |
+
self.num_upsamples = len(upsample_rates)
|
439 |
+
|
440 |
+
self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
|
441 |
+
self.m_source = SourceModuleHnNSF(
|
442 |
+
sampling_rate=sr, harmonic_num=0, is_half=is_half
|
443 |
+
)
|
444 |
+
self.noise_convs = nn.ModuleList()
|
445 |
+
self.conv_pre = Conv1d(
|
446 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
447 |
+
)
|
448 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
449 |
+
|
450 |
+
self.ups = nn.ModuleList()
|
451 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
452 |
+
c_cur = upsample_initial_channel // (2 ** (i + 1))
|
453 |
+
self.ups.append(
|
454 |
+
weight_norm(
|
455 |
+
ConvTranspose1d(
|
456 |
+
upsample_initial_channel // (2**i),
|
457 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
458 |
+
k,
|
459 |
+
u,
|
460 |
+
padding=(k - u) // 2,
|
461 |
+
)
|
462 |
+
)
|
463 |
+
)
|
464 |
+
if i + 1 < len(upsample_rates):
|
465 |
+
stride_f0 = np.prod(upsample_rates[i + 1 :])
|
466 |
+
self.noise_convs.append(
|
467 |
+
Conv1d(
|
468 |
+
1,
|
469 |
+
c_cur,
|
470 |
+
kernel_size=stride_f0 * 2,
|
471 |
+
stride=stride_f0,
|
472 |
+
padding=stride_f0 // 2,
|
473 |
+
)
|
474 |
+
)
|
475 |
+
else:
|
476 |
+
self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
|
477 |
+
|
478 |
+
self.resblocks = nn.ModuleList()
|
479 |
+
for i in range(len(self.ups)):
|
480 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
481 |
+
for j, (k, d) in enumerate(
|
482 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
483 |
+
):
|
484 |
+
self.resblocks.append(resblock(ch, k, d))
|
485 |
+
|
486 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
487 |
+
self.ups.apply(init_weights)
|
488 |
+
|
489 |
+
if gin_channels != 0:
|
490 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
491 |
+
|
492 |
+
self.upp = np.prod(upsample_rates)
|
493 |
+
|
494 |
+
def forward(self, x, f0, g=None):
|
495 |
+
har_source, noi_source, uv = self.m_source(f0, self.upp)
|
496 |
+
har_source = har_source.transpose(1, 2)
|
497 |
+
x = self.conv_pre(x)
|
498 |
+
if g is not None:
|
499 |
+
x = x + self.cond(g)
|
500 |
+
|
501 |
+
for i in range(self.num_upsamples):
|
502 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
503 |
+
x = self.ups[i](x)
|
504 |
+
x_source = self.noise_convs[i](har_source)
|
505 |
+
x = x + x_source
|
506 |
+
xs = None
|
507 |
+
for j in range(self.num_kernels):
|
508 |
+
if xs is None:
|
509 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
510 |
+
else:
|
511 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
512 |
+
x = xs / self.num_kernels
|
513 |
+
x = F.leaky_relu(x)
|
514 |
+
x = self.conv_post(x)
|
515 |
+
x = torch.tanh(x)
|
516 |
+
return x
|
517 |
+
|
518 |
+
def remove_weight_norm(self):
|
519 |
+
for l in self.ups:
|
520 |
+
remove_weight_norm(l)
|
521 |
+
for l in self.resblocks:
|
522 |
+
l.remove_weight_norm()
|
523 |
+
|
524 |
+
|
525 |
+
sr2sr = {
|
526 |
+
"32k": 32000,
|
527 |
+
"40k": 40000,
|
528 |
+
"48k": 48000,
|
529 |
+
}
|
530 |
+
|
531 |
+
|
532 |
+
class SynthesizerTrnMs256NSFsid(nn.Module):
|
533 |
+
def __init__(
|
534 |
+
self,
|
535 |
+
spec_channels,
|
536 |
+
segment_size,
|
537 |
+
inter_channels,
|
538 |
+
hidden_channels,
|
539 |
+
filter_channels,
|
540 |
+
n_heads,
|
541 |
+
n_layers,
|
542 |
+
kernel_size,
|
543 |
+
p_dropout,
|
544 |
+
resblock,
|
545 |
+
resblock_kernel_sizes,
|
546 |
+
resblock_dilation_sizes,
|
547 |
+
upsample_rates,
|
548 |
+
upsample_initial_channel,
|
549 |
+
upsample_kernel_sizes,
|
550 |
+
spk_embed_dim,
|
551 |
+
gin_channels,
|
552 |
+
sr,
|
553 |
+
**kwargs
|
554 |
+
):
|
555 |
+
super().__init__()
|
556 |
+
if type(sr) == type("strr"):
|
557 |
+
sr = sr2sr[sr]
|
558 |
+
self.spec_channels = spec_channels
|
559 |
+
self.inter_channels = inter_channels
|
560 |
+
self.hidden_channels = hidden_channels
|
561 |
+
self.filter_channels = filter_channels
|
562 |
+
self.n_heads = n_heads
|
563 |
+
self.n_layers = n_layers
|
564 |
+
self.kernel_size = kernel_size
|
565 |
+
self.p_dropout = p_dropout
|
566 |
+
self.resblock = resblock
|
567 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
568 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
569 |
+
self.upsample_rates = upsample_rates
|
570 |
+
self.upsample_initial_channel = upsample_initial_channel
|
571 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
572 |
+
self.segment_size = segment_size
|
573 |
+
self.gin_channels = gin_channels
|
574 |
+
# self.hop_length = hop_length#
|
575 |
+
self.spk_embed_dim = spk_embed_dim
|
576 |
+
self.enc_p = TextEncoder256(
|
577 |
+
inter_channels,
|
578 |
+
hidden_channels,
|
579 |
+
filter_channels,
|
580 |
+
n_heads,
|
581 |
+
n_layers,
|
582 |
+
kernel_size,
|
583 |
+
p_dropout,
|
584 |
+
)
|
585 |
+
self.dec = GeneratorNSF(
|
586 |
+
inter_channels,
|
587 |
+
resblock,
|
588 |
+
resblock_kernel_sizes,
|
589 |
+
resblock_dilation_sizes,
|
590 |
+
upsample_rates,
|
591 |
+
upsample_initial_channel,
|
592 |
+
upsample_kernel_sizes,
|
593 |
+
gin_channels=gin_channels,
|
594 |
+
sr=sr,
|
595 |
+
is_half=kwargs["is_half"],
|
596 |
+
)
|
597 |
+
self.enc_q = PosteriorEncoder(
|
598 |
+
spec_channels,
|
599 |
+
inter_channels,
|
600 |
+
hidden_channels,
|
601 |
+
5,
|
602 |
+
1,
|
603 |
+
16,
|
604 |
+
gin_channels=gin_channels,
|
605 |
+
)
|
606 |
+
self.flow = ResidualCouplingBlock(
|
607 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
608 |
+
)
|
609 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
610 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
611 |
+
|
612 |
+
def remove_weight_norm(self):
|
613 |
+
self.dec.remove_weight_norm()
|
614 |
+
self.flow.remove_weight_norm()
|
615 |
+
self.enc_q.remove_weight_norm()
|
616 |
+
|
617 |
+
def forward(
|
618 |
+
self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
|
619 |
+
): # 这里ds是id,[bs,1]
|
620 |
+
# print(1,pitch.shape)#[bs,t]
|
621 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
622 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
623 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
624 |
+
z_p = self.flow(z, y_mask, g=g)
|
625 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
626 |
+
z, y_lengths, self.segment_size
|
627 |
+
)
|
628 |
+
# print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
|
629 |
+
pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
|
630 |
+
# print(-2,pitchf.shape,z_slice.shape)
|
631 |
+
o = self.dec(z_slice, pitchf, g=g)
|
632 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
633 |
+
|
634 |
+
def infer(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
|
635 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
636 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
637 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
638 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
639 |
+
o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
|
640 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
641 |
+
|
642 |
+
|
643 |
+
class SynthesizerTrnMs768NSFsid(nn.Module):
|
644 |
+
def __init__(
|
645 |
+
self,
|
646 |
+
spec_channels,
|
647 |
+
segment_size,
|
648 |
+
inter_channels,
|
649 |
+
hidden_channels,
|
650 |
+
filter_channels,
|
651 |
+
n_heads,
|
652 |
+
n_layers,
|
653 |
+
kernel_size,
|
654 |
+
p_dropout,
|
655 |
+
resblock,
|
656 |
+
resblock_kernel_sizes,
|
657 |
+
resblock_dilation_sizes,
|
658 |
+
upsample_rates,
|
659 |
+
upsample_initial_channel,
|
660 |
+
upsample_kernel_sizes,
|
661 |
+
spk_embed_dim,
|
662 |
+
gin_channels,
|
663 |
+
sr,
|
664 |
+
**kwargs
|
665 |
+
):
|
666 |
+
super().__init__()
|
667 |
+
if type(sr) == type("strr"):
|
668 |
+
sr = sr2sr[sr]
|
669 |
+
self.spec_channels = spec_channels
|
670 |
+
self.inter_channels = inter_channels
|
671 |
+
self.hidden_channels = hidden_channels
|
672 |
+
self.filter_channels = filter_channels
|
673 |
+
self.n_heads = n_heads
|
674 |
+
self.n_layers = n_layers
|
675 |
+
self.kernel_size = kernel_size
|
676 |
+
self.p_dropout = p_dropout
|
677 |
+
self.resblock = resblock
|
678 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
679 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
680 |
+
self.upsample_rates = upsample_rates
|
681 |
+
self.upsample_initial_channel = upsample_initial_channel
|
682 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
683 |
+
self.segment_size = segment_size
|
684 |
+
self.gin_channels = gin_channels
|
685 |
+
# self.hop_length = hop_length#
|
686 |
+
self.spk_embed_dim = spk_embed_dim
|
687 |
+
self.enc_p = TextEncoder768(
|
688 |
+
inter_channels,
|
689 |
+
hidden_channels,
|
690 |
+
filter_channels,
|
691 |
+
n_heads,
|
692 |
+
n_layers,
|
693 |
+
kernel_size,
|
694 |
+
p_dropout,
|
695 |
+
)
|
696 |
+
self.dec = GeneratorNSF(
|
697 |
+
inter_channels,
|
698 |
+
resblock,
|
699 |
+
resblock_kernel_sizes,
|
700 |
+
resblock_dilation_sizes,
|
701 |
+
upsample_rates,
|
702 |
+
upsample_initial_channel,
|
703 |
+
upsample_kernel_sizes,
|
704 |
+
gin_channels=gin_channels,
|
705 |
+
sr=sr,
|
706 |
+
is_half=kwargs["is_half"],
|
707 |
+
)
|
708 |
+
self.enc_q = PosteriorEncoder(
|
709 |
+
spec_channels,
|
710 |
+
inter_channels,
|
711 |
+
hidden_channels,
|
712 |
+
5,
|
713 |
+
1,
|
714 |
+
16,
|
715 |
+
gin_channels=gin_channels,
|
716 |
+
)
|
717 |
+
self.flow = ResidualCouplingBlock(
|
718 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
719 |
+
)
|
720 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
721 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
722 |
+
|
723 |
+
def remove_weight_norm(self):
|
724 |
+
self.dec.remove_weight_norm()
|
725 |
+
self.flow.remove_weight_norm()
|
726 |
+
self.enc_q.remove_weight_norm()
|
727 |
+
|
728 |
+
def forward(
|
729 |
+
self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
|
730 |
+
): # 这里ds是id,[bs,1]
|
731 |
+
# print(1,pitch.shape)#[bs,t]
|
732 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
733 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
734 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
735 |
+
z_p = self.flow(z, y_mask, g=g)
|
736 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
737 |
+
z, y_lengths, self.segment_size
|
738 |
+
)
|
739 |
+
# print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
|
740 |
+
pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
|
741 |
+
# print(-2,pitchf.shape,z_slice.shape)
|
742 |
+
o = self.dec(z_slice, pitchf, g=g)
|
743 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
744 |
+
|
745 |
+
def infer(self, phone, phone_lengths, pitch, nsff0, sid, max_len=None):
|
746 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
747 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
748 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
749 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
750 |
+
o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
|
751 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
752 |
+
|
753 |
+
|
754 |
+
class SynthesizerTrnMs256NSFsid_nono(nn.Module):
|
755 |
+
def __init__(
|
756 |
+
self,
|
757 |
+
spec_channels,
|
758 |
+
segment_size,
|
759 |
+
inter_channels,
|
760 |
+
hidden_channels,
|
761 |
+
filter_channels,
|
762 |
+
n_heads,
|
763 |
+
n_layers,
|
764 |
+
kernel_size,
|
765 |
+
p_dropout,
|
766 |
+
resblock,
|
767 |
+
resblock_kernel_sizes,
|
768 |
+
resblock_dilation_sizes,
|
769 |
+
upsample_rates,
|
770 |
+
upsample_initial_channel,
|
771 |
+
upsample_kernel_sizes,
|
772 |
+
spk_embed_dim,
|
773 |
+
gin_channels,
|
774 |
+
sr=None,
|
775 |
+
**kwargs
|
776 |
+
):
|
777 |
+
super().__init__()
|
778 |
+
self.spec_channels = spec_channels
|
779 |
+
self.inter_channels = inter_channels
|
780 |
+
self.hidden_channels = hidden_channels
|
781 |
+
self.filter_channels = filter_channels
|
782 |
+
self.n_heads = n_heads
|
783 |
+
self.n_layers = n_layers
|
784 |
+
self.kernel_size = kernel_size
|
785 |
+
self.p_dropout = p_dropout
|
786 |
+
self.resblock = resblock
|
787 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
788 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
789 |
+
self.upsample_rates = upsample_rates
|
790 |
+
self.upsample_initial_channel = upsample_initial_channel
|
791 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
792 |
+
self.segment_size = segment_size
|
793 |
+
self.gin_channels = gin_channels
|
794 |
+
# self.hop_length = hop_length#
|
795 |
+
self.spk_embed_dim = spk_embed_dim
|
796 |
+
self.enc_p = TextEncoder256(
|
797 |
+
inter_channels,
|
798 |
+
hidden_channels,
|
799 |
+
filter_channels,
|
800 |
+
n_heads,
|
801 |
+
n_layers,
|
802 |
+
kernel_size,
|
803 |
+
p_dropout,
|
804 |
+
f0=False,
|
805 |
+
)
|
806 |
+
self.dec = Generator(
|
807 |
+
inter_channels,
|
808 |
+
resblock,
|
809 |
+
resblock_kernel_sizes,
|
810 |
+
resblock_dilation_sizes,
|
811 |
+
upsample_rates,
|
812 |
+
upsample_initial_channel,
|
813 |
+
upsample_kernel_sizes,
|
814 |
+
gin_channels=gin_channels,
|
815 |
+
)
|
816 |
+
self.enc_q = PosteriorEncoder(
|
817 |
+
spec_channels,
|
818 |
+
inter_channels,
|
819 |
+
hidden_channels,
|
820 |
+
5,
|
821 |
+
1,
|
822 |
+
16,
|
823 |
+
gin_channels=gin_channels,
|
824 |
+
)
|
825 |
+
self.flow = ResidualCouplingBlock(
|
826 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
827 |
+
)
|
828 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
829 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
830 |
+
|
831 |
+
def remove_weight_norm(self):
|
832 |
+
self.dec.remove_weight_norm()
|
833 |
+
self.flow.remove_weight_norm()
|
834 |
+
self.enc_q.remove_weight_norm()
|
835 |
+
|
836 |
+
def forward(self, phone, phone_lengths, y, y_lengths, ds): # 这里ds是id,[bs,1]
|
837 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
838 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
839 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
840 |
+
z_p = self.flow(z, y_mask, g=g)
|
841 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
842 |
+
z, y_lengths, self.segment_size
|
843 |
+
)
|
844 |
+
o = self.dec(z_slice, g=g)
|
845 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
846 |
+
|
847 |
+
def infer(self, phone, phone_lengths, sid, max_len=None):
|
848 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
849 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
850 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
851 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
852 |
+
o = self.dec((z * x_mask)[:, :, :max_len], g=g)
|
853 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
854 |
+
|
855 |
+
|
856 |
+
class SynthesizerTrnMs768NSFsid_nono(nn.Module):
|
857 |
+
def __init__(
|
858 |
+
self,
|
859 |
+
spec_channels,
|
860 |
+
segment_size,
|
861 |
+
inter_channels,
|
862 |
+
hidden_channels,
|
863 |
+
filter_channels,
|
864 |
+
n_heads,
|
865 |
+
n_layers,
|
866 |
+
kernel_size,
|
867 |
+
p_dropout,
|
868 |
+
resblock,
|
869 |
+
resblock_kernel_sizes,
|
870 |
+
resblock_dilation_sizes,
|
871 |
+
upsample_rates,
|
872 |
+
upsample_initial_channel,
|
873 |
+
upsample_kernel_sizes,
|
874 |
+
spk_embed_dim,
|
875 |
+
gin_channels,
|
876 |
+
sr=None,
|
877 |
+
**kwargs
|
878 |
+
):
|
879 |
+
super().__init__()
|
880 |
+
self.spec_channels = spec_channels
|
881 |
+
self.inter_channels = inter_channels
|
882 |
+
self.hidden_channels = hidden_channels
|
883 |
+
self.filter_channels = filter_channels
|
884 |
+
self.n_heads = n_heads
|
885 |
+
self.n_layers = n_layers
|
886 |
+
self.kernel_size = kernel_size
|
887 |
+
self.p_dropout = p_dropout
|
888 |
+
self.resblock = resblock
|
889 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
890 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
891 |
+
self.upsample_rates = upsample_rates
|
892 |
+
self.upsample_initial_channel = upsample_initial_channel
|
893 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
894 |
+
self.segment_size = segment_size
|
895 |
+
self.gin_channels = gin_channels
|
896 |
+
# self.hop_length = hop_length#
|
897 |
+
self.spk_embed_dim = spk_embed_dim
|
898 |
+
self.enc_p = TextEncoder768(
|
899 |
+
inter_channels,
|
900 |
+
hidden_channels,
|
901 |
+
filter_channels,
|
902 |
+
n_heads,
|
903 |
+
n_layers,
|
904 |
+
kernel_size,
|
905 |
+
p_dropout,
|
906 |
+
f0=False,
|
907 |
+
)
|
908 |
+
self.dec = Generator(
|
909 |
+
inter_channels,
|
910 |
+
resblock,
|
911 |
+
resblock_kernel_sizes,
|
912 |
+
resblock_dilation_sizes,
|
913 |
+
upsample_rates,
|
914 |
+
upsample_initial_channel,
|
915 |
+
upsample_kernel_sizes,
|
916 |
+
gin_channels=gin_channels,
|
917 |
+
)
|
918 |
+
self.enc_q = PosteriorEncoder(
|
919 |
+
spec_channels,
|
920 |
+
inter_channels,
|
921 |
+
hidden_channels,
|
922 |
+
5,
|
923 |
+
1,
|
924 |
+
16,
|
925 |
+
gin_channels=gin_channels,
|
926 |
+
)
|
927 |
+
self.flow = ResidualCouplingBlock(
|
928 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
929 |
+
)
|
930 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
931 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
932 |
+
|
933 |
+
def remove_weight_norm(self):
|
934 |
+
self.dec.remove_weight_norm()
|
935 |
+
self.flow.remove_weight_norm()
|
936 |
+
self.enc_q.remove_weight_norm()
|
937 |
+
|
938 |
+
def forward(self, phone, phone_lengths, y, y_lengths, ds): # 这里ds是id,[bs,1]
|
939 |
+
g = self.emb_g(ds).unsqueeze(-1) # [b, 256, 1]##1是t,广播的
|
940 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
941 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
|
942 |
+
z_p = self.flow(z, y_mask, g=g)
|
943 |
+
z_slice, ids_slice = commons.rand_slice_segments(
|
944 |
+
z, y_lengths, self.segment_size
|
945 |
+
)
|
946 |
+
o = self.dec(z_slice, g=g)
|
947 |
+
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
|
948 |
+
|
949 |
+
def infer(self, phone, phone_lengths, sid, max_len=None):
|
950 |
+
g = self.emb_g(sid).unsqueeze(-1)
|
951 |
+
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
|
952 |
+
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
|
953 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
954 |
+
o = self.dec((z * x_mask)[:, :, :max_len], g=g)
|
955 |
+
return o, x_mask, (z, z_p, m_p, logs_p)
|
956 |
+
|
957 |
+
|
958 |
+
class MultiPeriodDiscriminator(torch.nn.Module):
|
959 |
+
def __init__(self, use_spectral_norm=False):
|
960 |
+
super(MultiPeriodDiscriminator, self).__init__()
|
961 |
+
periods = [2, 3, 5, 7, 11, 17]
|
962 |
+
# periods = [3, 5, 7, 11, 17, 23, 37]
|
963 |
+
|
964 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
965 |
+
discs = discs + [
|
966 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
967 |
+
]
|
968 |
+
self.discriminators = nn.ModuleList(discs)
|
969 |
+
|
970 |
+
def forward(self, y, y_hat):
|
971 |
+
y_d_rs = [] #
|
972 |
+
y_d_gs = []
|
973 |
+
fmap_rs = []
|
974 |
+
fmap_gs = []
|
975 |
+
for i, d in enumerate(self.discriminators):
|
976 |
+
y_d_r, fmap_r = d(y)
|
977 |
+
y_d_g, fmap_g = d(y_hat)
|
978 |
+
# for j in range(len(fmap_r)):
|
979 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
980 |
+
y_d_rs.append(y_d_r)
|
981 |
+
y_d_gs.append(y_d_g)
|
982 |
+
fmap_rs.append(fmap_r)
|
983 |
+
fmap_gs.append(fmap_g)
|
984 |
+
|
985 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
986 |
+
|
987 |
+
|
988 |
+
class MultiPeriodDiscriminatorV2(torch.nn.Module):
|
989 |
+
def __init__(self, use_spectral_norm=False):
|
990 |
+
super(MultiPeriodDiscriminatorV2, self).__init__()
|
991 |
+
# periods = [2, 3, 5, 7, 11, 17]
|
992 |
+
periods = [2, 3, 5, 7, 11, 17, 23, 37]
|
993 |
+
|
994 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
995 |
+
discs = discs + [
|
996 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
997 |
+
]
|
998 |
+
self.discriminators = nn.ModuleList(discs)
|
999 |
+
|
1000 |
+
def forward(self, y, y_hat):
|
1001 |
+
y_d_rs = [] #
|
1002 |
+
y_d_gs = []
|
1003 |
+
fmap_rs = []
|
1004 |
+
fmap_gs = []
|
1005 |
+
for i, d in enumerate(self.discriminators):
|
1006 |
+
y_d_r, fmap_r = d(y)
|
1007 |
+
y_d_g, fmap_g = d(y_hat)
|
1008 |
+
# for j in range(len(fmap_r)):
|
1009 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
1010 |
+
y_d_rs.append(y_d_r)
|
1011 |
+
y_d_gs.append(y_d_g)
|
1012 |
+
fmap_rs.append(fmap_r)
|
1013 |
+
fmap_gs.append(fmap_g)
|
1014 |
+
|
1015 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
1016 |
+
|
1017 |
+
|
1018 |
+
class DiscriminatorS(torch.nn.Module):
|
1019 |
+
def __init__(self, use_spectral_norm=False):
|
1020 |
+
super(DiscriminatorS, self).__init__()
|
1021 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
1022 |
+
self.convs = nn.ModuleList(
|
1023 |
+
[
|
1024 |
+
norm_f(Conv1d(1, 16, 15, 1, padding=7)),
|
1025 |
+
norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
|
1026 |
+
norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
|
1027 |
+
norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
|
1028 |
+
norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
|
1029 |
+
norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
|
1030 |
+
]
|
1031 |
+
)
|
1032 |
+
self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
|
1033 |
+
|
1034 |
+
def forward(self, x):
|
1035 |
+
fmap = []
|
1036 |
+
|
1037 |
+
for l in self.convs:
|
1038 |
+
x = l(x)
|
1039 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
1040 |
+
fmap.append(x)
|
1041 |
+
x = self.conv_post(x)
|
1042 |
+
fmap.append(x)
|
1043 |
+
x = torch.flatten(x, 1, -1)
|
1044 |
+
|
1045 |
+
return x, fmap
|
1046 |
+
|
1047 |
+
|
1048 |
+
class DiscriminatorP(torch.nn.Module):
|
1049 |
+
def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
|
1050 |
+
super(DiscriminatorP, self).__init__()
|
1051 |
+
self.period = period
|
1052 |
+
self.use_spectral_norm = use_spectral_norm
|
1053 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
1054 |
+
self.convs = nn.ModuleList(
|
1055 |
+
[
|
1056 |
+
norm_f(
|
1057 |
+
Conv2d(
|
1058 |
+
1,
|
1059 |
+
32,
|
1060 |
+
(kernel_size, 1),
|
1061 |
+
(stride, 1),
|
1062 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1063 |
+
)
|
1064 |
+
),
|
1065 |
+
norm_f(
|
1066 |
+
Conv2d(
|
1067 |
+
32,
|
1068 |
+
128,
|
1069 |
+
(kernel_size, 1),
|
1070 |
+
(stride, 1),
|
1071 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1072 |
+
)
|
1073 |
+
),
|
1074 |
+
norm_f(
|
1075 |
+
Conv2d(
|
1076 |
+
128,
|
1077 |
+
512,
|
1078 |
+
(kernel_size, 1),
|
1079 |
+
(stride, 1),
|
1080 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1081 |
+
)
|
1082 |
+
),
|
1083 |
+
norm_f(
|
1084 |
+
Conv2d(
|
1085 |
+
512,
|
1086 |
+
1024,
|
1087 |
+
(kernel_size, 1),
|
1088 |
+
(stride, 1),
|
1089 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1090 |
+
)
|
1091 |
+
),
|
1092 |
+
norm_f(
|
1093 |
+
Conv2d(
|
1094 |
+
1024,
|
1095 |
+
1024,
|
1096 |
+
(kernel_size, 1),
|
1097 |
+
1,
|
1098 |
+
padding=(get_padding(kernel_size, 1), 0),
|
1099 |
+
)
|
1100 |
+
),
|
1101 |
+
]
|
1102 |
+
)
|
1103 |
+
self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
|
1104 |
+
|
1105 |
+
def forward(self, x):
|
1106 |
+
fmap = []
|
1107 |
+
|
1108 |
+
# 1d to 2d
|
1109 |
+
b, c, t = x.shape
|
1110 |
+
if t % self.period != 0: # pad first
|
1111 |
+
n_pad = self.period - (t % self.period)
|
1112 |
+
x = F.pad(x, (0, n_pad), "reflect")
|
1113 |
+
t = t + n_pad
|
1114 |
+
x = x.view(b, c, t // self.period, self.period)
|
1115 |
+
|
1116 |
+
for l in self.convs:
|
1117 |
+
x = l(x)
|
1118 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
1119 |
+
fmap.append(x)
|
1120 |
+
x = self.conv_post(x)
|
1121 |
+
fmap.append(x)
|
1122 |
+
x = torch.flatten(x, 1, -1)
|
1123 |
+
|
1124 |
+
return x, fmap
|
lib/infer_pack/models_onnx.py
ADDED
@@ -0,0 +1,819 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math, pdb, os
|
2 |
+
from time import time as ttime
|
3 |
+
import torch
|
4 |
+
from torch import nn
|
5 |
+
from torch.nn import functional as F
|
6 |
+
from lib.infer_pack import modules
|
7 |
+
from lib.infer_pack import attentions
|
8 |
+
from lib.infer_pack import commons
|
9 |
+
from lib.infer_pack.commons import init_weights, get_padding
|
10 |
+
from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
|
11 |
+
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
|
12 |
+
from lib.infer_pack.commons import init_weights
|
13 |
+
import numpy as np
|
14 |
+
from lib.infer_pack import commons
|
15 |
+
|
16 |
+
|
17 |
+
class TextEncoder256(nn.Module):
|
18 |
+
def __init__(
|
19 |
+
self,
|
20 |
+
out_channels,
|
21 |
+
hidden_channels,
|
22 |
+
filter_channels,
|
23 |
+
n_heads,
|
24 |
+
n_layers,
|
25 |
+
kernel_size,
|
26 |
+
p_dropout,
|
27 |
+
f0=True,
|
28 |
+
):
|
29 |
+
super().__init__()
|
30 |
+
self.out_channels = out_channels
|
31 |
+
self.hidden_channels = hidden_channels
|
32 |
+
self.filter_channels = filter_channels
|
33 |
+
self.n_heads = n_heads
|
34 |
+
self.n_layers = n_layers
|
35 |
+
self.kernel_size = kernel_size
|
36 |
+
self.p_dropout = p_dropout
|
37 |
+
self.emb_phone = nn.Linear(256, hidden_channels)
|
38 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
39 |
+
if f0 == True:
|
40 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
41 |
+
self.encoder = attentions.Encoder(
|
42 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
43 |
+
)
|
44 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
45 |
+
|
46 |
+
def forward(self, phone, pitch, lengths):
|
47 |
+
if pitch == None:
|
48 |
+
x = self.emb_phone(phone)
|
49 |
+
else:
|
50 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
51 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
52 |
+
x = self.lrelu(x)
|
53 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
54 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
55 |
+
x.dtype
|
56 |
+
)
|
57 |
+
x = self.encoder(x * x_mask, x_mask)
|
58 |
+
stats = self.proj(x) * x_mask
|
59 |
+
|
60 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
61 |
+
return m, logs, x_mask
|
62 |
+
|
63 |
+
|
64 |
+
class TextEncoder768(nn.Module):
|
65 |
+
def __init__(
|
66 |
+
self,
|
67 |
+
out_channels,
|
68 |
+
hidden_channels,
|
69 |
+
filter_channels,
|
70 |
+
n_heads,
|
71 |
+
n_layers,
|
72 |
+
kernel_size,
|
73 |
+
p_dropout,
|
74 |
+
f0=True,
|
75 |
+
):
|
76 |
+
super().__init__()
|
77 |
+
self.out_channels = out_channels
|
78 |
+
self.hidden_channels = hidden_channels
|
79 |
+
self.filter_channels = filter_channels
|
80 |
+
self.n_heads = n_heads
|
81 |
+
self.n_layers = n_layers
|
82 |
+
self.kernel_size = kernel_size
|
83 |
+
self.p_dropout = p_dropout
|
84 |
+
self.emb_phone = nn.Linear(768, hidden_channels)
|
85 |
+
self.lrelu = nn.LeakyReLU(0.1, inplace=True)
|
86 |
+
if f0 == True:
|
87 |
+
self.emb_pitch = nn.Embedding(256, hidden_channels) # pitch 256
|
88 |
+
self.encoder = attentions.Encoder(
|
89 |
+
hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
|
90 |
+
)
|
91 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
92 |
+
|
93 |
+
def forward(self, phone, pitch, lengths):
|
94 |
+
if pitch == None:
|
95 |
+
x = self.emb_phone(phone)
|
96 |
+
else:
|
97 |
+
x = self.emb_phone(phone) + self.emb_pitch(pitch)
|
98 |
+
x = x * math.sqrt(self.hidden_channels) # [b, t, h]
|
99 |
+
x = self.lrelu(x)
|
100 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
101 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
|
102 |
+
x.dtype
|
103 |
+
)
|
104 |
+
x = self.encoder(x * x_mask, x_mask)
|
105 |
+
stats = self.proj(x) * x_mask
|
106 |
+
|
107 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
108 |
+
return m, logs, x_mask
|
109 |
+
|
110 |
+
|
111 |
+
class ResidualCouplingBlock(nn.Module):
|
112 |
+
def __init__(
|
113 |
+
self,
|
114 |
+
channels,
|
115 |
+
hidden_channels,
|
116 |
+
kernel_size,
|
117 |
+
dilation_rate,
|
118 |
+
n_layers,
|
119 |
+
n_flows=4,
|
120 |
+
gin_channels=0,
|
121 |
+
):
|
122 |
+
super().__init__()
|
123 |
+
self.channels = channels
|
124 |
+
self.hidden_channels = hidden_channels
|
125 |
+
self.kernel_size = kernel_size
|
126 |
+
self.dilation_rate = dilation_rate
|
127 |
+
self.n_layers = n_layers
|
128 |
+
self.n_flows = n_flows
|
129 |
+
self.gin_channels = gin_channels
|
130 |
+
|
131 |
+
self.flows = nn.ModuleList()
|
132 |
+
for i in range(n_flows):
|
133 |
+
self.flows.append(
|
134 |
+
modules.ResidualCouplingLayer(
|
135 |
+
channels,
|
136 |
+
hidden_channels,
|
137 |
+
kernel_size,
|
138 |
+
dilation_rate,
|
139 |
+
n_layers,
|
140 |
+
gin_channels=gin_channels,
|
141 |
+
mean_only=True,
|
142 |
+
)
|
143 |
+
)
|
144 |
+
self.flows.append(modules.Flip())
|
145 |
+
|
146 |
+
def forward(self, x, x_mask, g=None, reverse=False):
|
147 |
+
if not reverse:
|
148 |
+
for flow in self.flows:
|
149 |
+
x, _ = flow(x, x_mask, g=g, reverse=reverse)
|
150 |
+
else:
|
151 |
+
for flow in reversed(self.flows):
|
152 |
+
x = flow(x, x_mask, g=g, reverse=reverse)
|
153 |
+
return x
|
154 |
+
|
155 |
+
def remove_weight_norm(self):
|
156 |
+
for i in range(self.n_flows):
|
157 |
+
self.flows[i * 2].remove_weight_norm()
|
158 |
+
|
159 |
+
|
160 |
+
class PosteriorEncoder(nn.Module):
|
161 |
+
def __init__(
|
162 |
+
self,
|
163 |
+
in_channels,
|
164 |
+
out_channels,
|
165 |
+
hidden_channels,
|
166 |
+
kernel_size,
|
167 |
+
dilation_rate,
|
168 |
+
n_layers,
|
169 |
+
gin_channels=0,
|
170 |
+
):
|
171 |
+
super().__init__()
|
172 |
+
self.in_channels = in_channels
|
173 |
+
self.out_channels = out_channels
|
174 |
+
self.hidden_channels = hidden_channels
|
175 |
+
self.kernel_size = kernel_size
|
176 |
+
self.dilation_rate = dilation_rate
|
177 |
+
self.n_layers = n_layers
|
178 |
+
self.gin_channels = gin_channels
|
179 |
+
|
180 |
+
self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
|
181 |
+
self.enc = modules.WN(
|
182 |
+
hidden_channels,
|
183 |
+
kernel_size,
|
184 |
+
dilation_rate,
|
185 |
+
n_layers,
|
186 |
+
gin_channels=gin_channels,
|
187 |
+
)
|
188 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
189 |
+
|
190 |
+
def forward(self, x, x_lengths, g=None):
|
191 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
|
192 |
+
x.dtype
|
193 |
+
)
|
194 |
+
x = self.pre(x) * x_mask
|
195 |
+
x = self.enc(x, x_mask, g=g)
|
196 |
+
stats = self.proj(x) * x_mask
|
197 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
198 |
+
z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
|
199 |
+
return z, m, logs, x_mask
|
200 |
+
|
201 |
+
def remove_weight_norm(self):
|
202 |
+
self.enc.remove_weight_norm()
|
203 |
+
|
204 |
+
|
205 |
+
class Generator(torch.nn.Module):
|
206 |
+
def __init__(
|
207 |
+
self,
|
208 |
+
initial_channel,
|
209 |
+
resblock,
|
210 |
+
resblock_kernel_sizes,
|
211 |
+
resblock_dilation_sizes,
|
212 |
+
upsample_rates,
|
213 |
+
upsample_initial_channel,
|
214 |
+
upsample_kernel_sizes,
|
215 |
+
gin_channels=0,
|
216 |
+
):
|
217 |
+
super(Generator, self).__init__()
|
218 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
219 |
+
self.num_upsamples = len(upsample_rates)
|
220 |
+
self.conv_pre = Conv1d(
|
221 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
222 |
+
)
|
223 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
224 |
+
|
225 |
+
self.ups = nn.ModuleList()
|
226 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
227 |
+
self.ups.append(
|
228 |
+
weight_norm(
|
229 |
+
ConvTranspose1d(
|
230 |
+
upsample_initial_channel // (2**i),
|
231 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
232 |
+
k,
|
233 |
+
u,
|
234 |
+
padding=(k - u) // 2,
|
235 |
+
)
|
236 |
+
)
|
237 |
+
)
|
238 |
+
|
239 |
+
self.resblocks = nn.ModuleList()
|
240 |
+
for i in range(len(self.ups)):
|
241 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
242 |
+
for j, (k, d) in enumerate(
|
243 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
244 |
+
):
|
245 |
+
self.resblocks.append(resblock(ch, k, d))
|
246 |
+
|
247 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
248 |
+
self.ups.apply(init_weights)
|
249 |
+
|
250 |
+
if gin_channels != 0:
|
251 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
252 |
+
|
253 |
+
def forward(self, x, g=None):
|
254 |
+
x = self.conv_pre(x)
|
255 |
+
if g is not None:
|
256 |
+
x = x + self.cond(g)
|
257 |
+
|
258 |
+
for i in range(self.num_upsamples):
|
259 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
260 |
+
x = self.ups[i](x)
|
261 |
+
xs = None
|
262 |
+
for j in range(self.num_kernels):
|
263 |
+
if xs is None:
|
264 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
265 |
+
else:
|
266 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
267 |
+
x = xs / self.num_kernels
|
268 |
+
x = F.leaky_relu(x)
|
269 |
+
x = self.conv_post(x)
|
270 |
+
x = torch.tanh(x)
|
271 |
+
|
272 |
+
return x
|
273 |
+
|
274 |
+
def remove_weight_norm(self):
|
275 |
+
for l in self.ups:
|
276 |
+
remove_weight_norm(l)
|
277 |
+
for l in self.resblocks:
|
278 |
+
l.remove_weight_norm()
|
279 |
+
|
280 |
+
|
281 |
+
class SineGen(torch.nn.Module):
|
282 |
+
"""Definition of sine generator
|
283 |
+
SineGen(samp_rate, harmonic_num = 0,
|
284 |
+
sine_amp = 0.1, noise_std = 0.003,
|
285 |
+
voiced_threshold = 0,
|
286 |
+
flag_for_pulse=False)
|
287 |
+
samp_rate: sampling rate in Hz
|
288 |
+
harmonic_num: number of harmonic overtones (default 0)
|
289 |
+
sine_amp: amplitude of sine-wavefrom (default 0.1)
|
290 |
+
noise_std: std of Gaussian noise (default 0.003)
|
291 |
+
voiced_thoreshold: F0 threshold for U/V classification (default 0)
|
292 |
+
flag_for_pulse: this SinGen is used inside PulseGen (default False)
|
293 |
+
Note: when flag_for_pulse is True, the first time step of a voiced
|
294 |
+
segment is always sin(np.pi) or cos(0)
|
295 |
+
"""
|
296 |
+
|
297 |
+
def __init__(
|
298 |
+
self,
|
299 |
+
samp_rate,
|
300 |
+
harmonic_num=0,
|
301 |
+
sine_amp=0.1,
|
302 |
+
noise_std=0.003,
|
303 |
+
voiced_threshold=0,
|
304 |
+
flag_for_pulse=False,
|
305 |
+
):
|
306 |
+
super(SineGen, self).__init__()
|
307 |
+
self.sine_amp = sine_amp
|
308 |
+
self.noise_std = noise_std
|
309 |
+
self.harmonic_num = harmonic_num
|
310 |
+
self.dim = self.harmonic_num + 1
|
311 |
+
self.sampling_rate = samp_rate
|
312 |
+
self.voiced_threshold = voiced_threshold
|
313 |
+
|
314 |
+
def _f02uv(self, f0):
|
315 |
+
# generate uv signal
|
316 |
+
uv = torch.ones_like(f0)
|
317 |
+
uv = uv * (f0 > self.voiced_threshold)
|
318 |
+
return uv
|
319 |
+
|
320 |
+
def forward(self, f0, upp):
|
321 |
+
"""sine_tensor, uv = forward(f0)
|
322 |
+
input F0: tensor(batchsize=1, length, dim=1)
|
323 |
+
f0 for unvoiced steps should be 0
|
324 |
+
output sine_tensor: tensor(batchsize=1, length, dim)
|
325 |
+
output uv: tensor(batchsize=1, length, 1)
|
326 |
+
"""
|
327 |
+
with torch.no_grad():
|
328 |
+
f0 = f0[:, None].transpose(1, 2)
|
329 |
+
f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
|
330 |
+
# fundamental component
|
331 |
+
f0_buf[:, :, 0] = f0[:, :, 0]
|
332 |
+
for idx in np.arange(self.harmonic_num):
|
333 |
+
f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
|
334 |
+
idx + 2
|
335 |
+
) # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
|
336 |
+
rad_values = (f0_buf / self.sampling_rate) % 1 ###%1意味着n_har的乘积无法后处理优化
|
337 |
+
rand_ini = torch.rand(
|
338 |
+
f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
|
339 |
+
)
|
340 |
+
rand_ini[:, 0] = 0
|
341 |
+
rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
|
342 |
+
tmp_over_one = torch.cumsum(rad_values, 1) # % 1 #####%1意味着后面的cumsum无法再优化
|
343 |
+
tmp_over_one *= upp
|
344 |
+
tmp_over_one = F.interpolate(
|
345 |
+
tmp_over_one.transpose(2, 1),
|
346 |
+
scale_factor=upp,
|
347 |
+
mode="linear",
|
348 |
+
align_corners=True,
|
349 |
+
).transpose(2, 1)
|
350 |
+
rad_values = F.interpolate(
|
351 |
+
rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
|
352 |
+
).transpose(
|
353 |
+
2, 1
|
354 |
+
) #######
|
355 |
+
tmp_over_one %= 1
|
356 |
+
tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
|
357 |
+
cumsum_shift = torch.zeros_like(rad_values)
|
358 |
+
cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
|
359 |
+
sine_waves = torch.sin(
|
360 |
+
torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
|
361 |
+
)
|
362 |
+
sine_waves = sine_waves * self.sine_amp
|
363 |
+
uv = self._f02uv(f0)
|
364 |
+
uv = F.interpolate(
|
365 |
+
uv.transpose(2, 1), scale_factor=upp, mode="nearest"
|
366 |
+
).transpose(2, 1)
|
367 |
+
noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
|
368 |
+
noise = noise_amp * torch.randn_like(sine_waves)
|
369 |
+
sine_waves = sine_waves * uv + noise
|
370 |
+
return sine_waves, uv, noise
|
371 |
+
|
372 |
+
|
373 |
+
class SourceModuleHnNSF(torch.nn.Module):
|
374 |
+
"""SourceModule for hn-nsf
|
375 |
+
SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
|
376 |
+
add_noise_std=0.003, voiced_threshod=0)
|
377 |
+
sampling_rate: sampling_rate in Hz
|
378 |
+
harmonic_num: number of harmonic above F0 (default: 0)
|
379 |
+
sine_amp: amplitude of sine source signal (default: 0.1)
|
380 |
+
add_noise_std: std of additive Gaussian noise (default: 0.003)
|
381 |
+
note that amplitude of noise in unvoiced is decided
|
382 |
+
by sine_amp
|
383 |
+
voiced_threshold: threhold to set U/V given F0 (default: 0)
|
384 |
+
Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
|
385 |
+
F0_sampled (batchsize, length, 1)
|
386 |
+
Sine_source (batchsize, length, 1)
|
387 |
+
noise_source (batchsize, length 1)
|
388 |
+
uv (batchsize, length, 1)
|
389 |
+
"""
|
390 |
+
|
391 |
+
def __init__(
|
392 |
+
self,
|
393 |
+
sampling_rate,
|
394 |
+
harmonic_num=0,
|
395 |
+
sine_amp=0.1,
|
396 |
+
add_noise_std=0.003,
|
397 |
+
voiced_threshod=0,
|
398 |
+
is_half=True,
|
399 |
+
):
|
400 |
+
super(SourceModuleHnNSF, self).__init__()
|
401 |
+
|
402 |
+
self.sine_amp = sine_amp
|
403 |
+
self.noise_std = add_noise_std
|
404 |
+
self.is_half = is_half
|
405 |
+
# to produce sine waveforms
|
406 |
+
self.l_sin_gen = SineGen(
|
407 |
+
sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
|
408 |
+
)
|
409 |
+
|
410 |
+
# to merge source harmonics into a single excitation
|
411 |
+
self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
|
412 |
+
self.l_tanh = torch.nn.Tanh()
|
413 |
+
|
414 |
+
def forward(self, x, upp=None):
|
415 |
+
sine_wavs, uv, _ = self.l_sin_gen(x, upp)
|
416 |
+
if self.is_half:
|
417 |
+
sine_wavs = sine_wavs.half()
|
418 |
+
sine_merge = self.l_tanh(self.l_linear(sine_wavs))
|
419 |
+
return sine_merge, None, None # noise, uv
|
420 |
+
|
421 |
+
|
422 |
+
class GeneratorNSF(torch.nn.Module):
|
423 |
+
def __init__(
|
424 |
+
self,
|
425 |
+
initial_channel,
|
426 |
+
resblock,
|
427 |
+
resblock_kernel_sizes,
|
428 |
+
resblock_dilation_sizes,
|
429 |
+
upsample_rates,
|
430 |
+
upsample_initial_channel,
|
431 |
+
upsample_kernel_sizes,
|
432 |
+
gin_channels,
|
433 |
+
sr,
|
434 |
+
is_half=False,
|
435 |
+
):
|
436 |
+
super(GeneratorNSF, self).__init__()
|
437 |
+
self.num_kernels = len(resblock_kernel_sizes)
|
438 |
+
self.num_upsamples = len(upsample_rates)
|
439 |
+
|
440 |
+
self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
|
441 |
+
self.m_source = SourceModuleHnNSF(
|
442 |
+
sampling_rate=sr, harmonic_num=0, is_half=is_half
|
443 |
+
)
|
444 |
+
self.noise_convs = nn.ModuleList()
|
445 |
+
self.conv_pre = Conv1d(
|
446 |
+
initial_channel, upsample_initial_channel, 7, 1, padding=3
|
447 |
+
)
|
448 |
+
resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
|
449 |
+
|
450 |
+
self.ups = nn.ModuleList()
|
451 |
+
for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
|
452 |
+
c_cur = upsample_initial_channel // (2 ** (i + 1))
|
453 |
+
self.ups.append(
|
454 |
+
weight_norm(
|
455 |
+
ConvTranspose1d(
|
456 |
+
upsample_initial_channel // (2**i),
|
457 |
+
upsample_initial_channel // (2 ** (i + 1)),
|
458 |
+
k,
|
459 |
+
u,
|
460 |
+
padding=(k - u) // 2,
|
461 |
+
)
|
462 |
+
)
|
463 |
+
)
|
464 |
+
if i + 1 < len(upsample_rates):
|
465 |
+
stride_f0 = np.prod(upsample_rates[i + 1 :])
|
466 |
+
self.noise_convs.append(
|
467 |
+
Conv1d(
|
468 |
+
1,
|
469 |
+
c_cur,
|
470 |
+
kernel_size=stride_f0 * 2,
|
471 |
+
stride=stride_f0,
|
472 |
+
padding=stride_f0 // 2,
|
473 |
+
)
|
474 |
+
)
|
475 |
+
else:
|
476 |
+
self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
|
477 |
+
|
478 |
+
self.resblocks = nn.ModuleList()
|
479 |
+
for i in range(len(self.ups)):
|
480 |
+
ch = upsample_initial_channel // (2 ** (i + 1))
|
481 |
+
for j, (k, d) in enumerate(
|
482 |
+
zip(resblock_kernel_sizes, resblock_dilation_sizes)
|
483 |
+
):
|
484 |
+
self.resblocks.append(resblock(ch, k, d))
|
485 |
+
|
486 |
+
self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
|
487 |
+
self.ups.apply(init_weights)
|
488 |
+
|
489 |
+
if gin_channels != 0:
|
490 |
+
self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
|
491 |
+
|
492 |
+
self.upp = np.prod(upsample_rates)
|
493 |
+
|
494 |
+
def forward(self, x, f0, g=None):
|
495 |
+
har_source, noi_source, uv = self.m_source(f0, self.upp)
|
496 |
+
har_source = har_source.transpose(1, 2)
|
497 |
+
x = self.conv_pre(x)
|
498 |
+
if g is not None:
|
499 |
+
x = x + self.cond(g)
|
500 |
+
|
501 |
+
for i in range(self.num_upsamples):
|
502 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
503 |
+
x = self.ups[i](x)
|
504 |
+
x_source = self.noise_convs[i](har_source)
|
505 |
+
x = x + x_source
|
506 |
+
xs = None
|
507 |
+
for j in range(self.num_kernels):
|
508 |
+
if xs is None:
|
509 |
+
xs = self.resblocks[i * self.num_kernels + j](x)
|
510 |
+
else:
|
511 |
+
xs += self.resblocks[i * self.num_kernels + j](x)
|
512 |
+
x = xs / self.num_kernels
|
513 |
+
x = F.leaky_relu(x)
|
514 |
+
x = self.conv_post(x)
|
515 |
+
x = torch.tanh(x)
|
516 |
+
return x
|
517 |
+
|
518 |
+
def remove_weight_norm(self):
|
519 |
+
for l in self.ups:
|
520 |
+
remove_weight_norm(l)
|
521 |
+
for l in self.resblocks:
|
522 |
+
l.remove_weight_norm()
|
523 |
+
|
524 |
+
|
525 |
+
sr2sr = {
|
526 |
+
"32k": 32000,
|
527 |
+
"40k": 40000,
|
528 |
+
"48k": 48000,
|
529 |
+
}
|
530 |
+
|
531 |
+
|
532 |
+
class SynthesizerTrnMsNSFsidM(nn.Module):
|
533 |
+
def __init__(
|
534 |
+
self,
|
535 |
+
spec_channels,
|
536 |
+
segment_size,
|
537 |
+
inter_channels,
|
538 |
+
hidden_channels,
|
539 |
+
filter_channels,
|
540 |
+
n_heads,
|
541 |
+
n_layers,
|
542 |
+
kernel_size,
|
543 |
+
p_dropout,
|
544 |
+
resblock,
|
545 |
+
resblock_kernel_sizes,
|
546 |
+
resblock_dilation_sizes,
|
547 |
+
upsample_rates,
|
548 |
+
upsample_initial_channel,
|
549 |
+
upsample_kernel_sizes,
|
550 |
+
spk_embed_dim,
|
551 |
+
gin_channels,
|
552 |
+
sr,
|
553 |
+
version,
|
554 |
+
**kwargs
|
555 |
+
):
|
556 |
+
super().__init__()
|
557 |
+
if type(sr) == type("strr"):
|
558 |
+
sr = sr2sr[sr]
|
559 |
+
self.spec_channels = spec_channels
|
560 |
+
self.inter_channels = inter_channels
|
561 |
+
self.hidden_channels = hidden_channels
|
562 |
+
self.filter_channels = filter_channels
|
563 |
+
self.n_heads = n_heads
|
564 |
+
self.n_layers = n_layers
|
565 |
+
self.kernel_size = kernel_size
|
566 |
+
self.p_dropout = p_dropout
|
567 |
+
self.resblock = resblock
|
568 |
+
self.resblock_kernel_sizes = resblock_kernel_sizes
|
569 |
+
self.resblock_dilation_sizes = resblock_dilation_sizes
|
570 |
+
self.upsample_rates = upsample_rates
|
571 |
+
self.upsample_initial_channel = upsample_initial_channel
|
572 |
+
self.upsample_kernel_sizes = upsample_kernel_sizes
|
573 |
+
self.segment_size = segment_size
|
574 |
+
self.gin_channels = gin_channels
|
575 |
+
# self.hop_length = hop_length#
|
576 |
+
self.spk_embed_dim = spk_embed_dim
|
577 |
+
if version == "v1":
|
578 |
+
self.enc_p = TextEncoder256(
|
579 |
+
inter_channels,
|
580 |
+
hidden_channels,
|
581 |
+
filter_channels,
|
582 |
+
n_heads,
|
583 |
+
n_layers,
|
584 |
+
kernel_size,
|
585 |
+
p_dropout,
|
586 |
+
)
|
587 |
+
else:
|
588 |
+
self.enc_p = TextEncoder768(
|
589 |
+
inter_channels,
|
590 |
+
hidden_channels,
|
591 |
+
filter_channels,
|
592 |
+
n_heads,
|
593 |
+
n_layers,
|
594 |
+
kernel_size,
|
595 |
+
p_dropout,
|
596 |
+
)
|
597 |
+
self.dec = GeneratorNSF(
|
598 |
+
inter_channels,
|
599 |
+
resblock,
|
600 |
+
resblock_kernel_sizes,
|
601 |
+
resblock_dilation_sizes,
|
602 |
+
upsample_rates,
|
603 |
+
upsample_initial_channel,
|
604 |
+
upsample_kernel_sizes,
|
605 |
+
gin_channels=gin_channels,
|
606 |
+
sr=sr,
|
607 |
+
is_half=kwargs["is_half"],
|
608 |
+
)
|
609 |
+
self.enc_q = PosteriorEncoder(
|
610 |
+
spec_channels,
|
611 |
+
inter_channels,
|
612 |
+
hidden_channels,
|
613 |
+
5,
|
614 |
+
1,
|
615 |
+
16,
|
616 |
+
gin_channels=gin_channels,
|
617 |
+
)
|
618 |
+
self.flow = ResidualCouplingBlock(
|
619 |
+
inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
|
620 |
+
)
|
621 |
+
self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
|
622 |
+
self.speaker_map = None
|
623 |
+
print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
|
624 |
+
|
625 |
+
def remove_weight_norm(self):
|
626 |
+
self.dec.remove_weight_norm()
|
627 |
+
self.flow.remove_weight_norm()
|
628 |
+
self.enc_q.remove_weight_norm()
|
629 |
+
|
630 |
+
def construct_spkmixmap(self, n_speaker):
|
631 |
+
self.speaker_map = torch.zeros((n_speaker, 1, 1, self.gin_channels))
|
632 |
+
for i in range(n_speaker):
|
633 |
+
self.speaker_map[i] = self.emb_g(torch.LongTensor([[i]]))
|
634 |
+
self.speaker_map = self.speaker_map.unsqueeze(0)
|
635 |
+
|
636 |
+
def forward(self, phone, phone_lengths, pitch, nsff0, g, rnd, max_len=None):
|
637 |
+
if self.speaker_map is not None: # [N, S] * [S, B, 1, H]
|
638 |
+
g = g.reshape((g.shape[0], g.shape[1], 1, 1, 1)) # [N, S, B, 1, 1]
|
639 |
+
g = g * self.speaker_map # [N, S, B, 1, H]
|
640 |
+
g = torch.sum(g, dim=1) # [N, 1, B, 1, H]
|
641 |
+
g = g.transpose(0, -1).transpose(0, -2).squeeze(0) # [B, H, N]
|
642 |
+
else:
|
643 |
+
g = g.unsqueeze(0)
|
644 |
+
g = self.emb_g(g).transpose(1, 2)
|
645 |
+
|
646 |
+
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
|
647 |
+
z_p = (m_p + torch.exp(logs_p) * rnd) * x_mask
|
648 |
+
z = self.flow(z_p, x_mask, g=g, reverse=True)
|
649 |
+
o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
|
650 |
+
return o
|
651 |
+
|
652 |
+
|
653 |
+
class MultiPeriodDiscriminator(torch.nn.Module):
|
654 |
+
def __init__(self, use_spectral_norm=False):
|
655 |
+
super(MultiPeriodDiscriminator, self).__init__()
|
656 |
+
periods = [2, 3, 5, 7, 11, 17]
|
657 |
+
# periods = [3, 5, 7, 11, 17, 23, 37]
|
658 |
+
|
659 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
660 |
+
discs = discs + [
|
661 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
662 |
+
]
|
663 |
+
self.discriminators = nn.ModuleList(discs)
|
664 |
+
|
665 |
+
def forward(self, y, y_hat):
|
666 |
+
y_d_rs = [] #
|
667 |
+
y_d_gs = []
|
668 |
+
fmap_rs = []
|
669 |
+
fmap_gs = []
|
670 |
+
for i, d in enumerate(self.discriminators):
|
671 |
+
y_d_r, fmap_r = d(y)
|
672 |
+
y_d_g, fmap_g = d(y_hat)
|
673 |
+
# for j in range(len(fmap_r)):
|
674 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
675 |
+
y_d_rs.append(y_d_r)
|
676 |
+
y_d_gs.append(y_d_g)
|
677 |
+
fmap_rs.append(fmap_r)
|
678 |
+
fmap_gs.append(fmap_g)
|
679 |
+
|
680 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
681 |
+
|
682 |
+
|
683 |
+
class MultiPeriodDiscriminatorV2(torch.nn.Module):
|
684 |
+
def __init__(self, use_spectral_norm=False):
|
685 |
+
super(MultiPeriodDiscriminatorV2, self).__init__()
|
686 |
+
# periods = [2, 3, 5, 7, 11, 17]
|
687 |
+
periods = [2, 3, 5, 7, 11, 17, 23, 37]
|
688 |
+
|
689 |
+
discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
|
690 |
+
discs = discs + [
|
691 |
+
DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
|
692 |
+
]
|
693 |
+
self.discriminators = nn.ModuleList(discs)
|
694 |
+
|
695 |
+
def forward(self, y, y_hat):
|
696 |
+
y_d_rs = [] #
|
697 |
+
y_d_gs = []
|
698 |
+
fmap_rs = []
|
699 |
+
fmap_gs = []
|
700 |
+
for i, d in enumerate(self.discriminators):
|
701 |
+
y_d_r, fmap_r = d(y)
|
702 |
+
y_d_g, fmap_g = d(y_hat)
|
703 |
+
# for j in range(len(fmap_r)):
|
704 |
+
# print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
|
705 |
+
y_d_rs.append(y_d_r)
|
706 |
+
y_d_gs.append(y_d_g)
|
707 |
+
fmap_rs.append(fmap_r)
|
708 |
+
fmap_gs.append(fmap_g)
|
709 |
+
|
710 |
+
return y_d_rs, y_d_gs, fmap_rs, fmap_gs
|
711 |
+
|
712 |
+
|
713 |
+
class DiscriminatorS(torch.nn.Module):
|
714 |
+
def __init__(self, use_spectral_norm=False):
|
715 |
+
super(DiscriminatorS, self).__init__()
|
716 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
717 |
+
self.convs = nn.ModuleList(
|
718 |
+
[
|
719 |
+
norm_f(Conv1d(1, 16, 15, 1, padding=7)),
|
720 |
+
norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
|
721 |
+
norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
|
722 |
+
norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
|
723 |
+
norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
|
724 |
+
norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
|
725 |
+
]
|
726 |
+
)
|
727 |
+
self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
|
728 |
+
|
729 |
+
def forward(self, x):
|
730 |
+
fmap = []
|
731 |
+
|
732 |
+
for l in self.convs:
|
733 |
+
x = l(x)
|
734 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
735 |
+
fmap.append(x)
|
736 |
+
x = self.conv_post(x)
|
737 |
+
fmap.append(x)
|
738 |
+
x = torch.flatten(x, 1, -1)
|
739 |
+
|
740 |
+
return x, fmap
|
741 |
+
|
742 |
+
|
743 |
+
class DiscriminatorP(torch.nn.Module):
|
744 |
+
def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
|
745 |
+
super(DiscriminatorP, self).__init__()
|
746 |
+
self.period = period
|
747 |
+
self.use_spectral_norm = use_spectral_norm
|
748 |
+
norm_f = weight_norm if use_spectral_norm == False else spectral_norm
|
749 |
+
self.convs = nn.ModuleList(
|
750 |
+
[
|
751 |
+
norm_f(
|
752 |
+
Conv2d(
|
753 |
+
1,
|
754 |
+
32,
|
755 |
+
(kernel_size, 1),
|
756 |
+
(stride, 1),
|
757 |
+
padding=(get_padding(kernel_size, 1), 0),
|
758 |
+
)
|
759 |
+
),
|
760 |
+
norm_f(
|
761 |
+
Conv2d(
|
762 |
+
32,
|
763 |
+
128,
|
764 |
+
(kernel_size, 1),
|
765 |
+
(stride, 1),
|
766 |
+
padding=(get_padding(kernel_size, 1), 0),
|
767 |
+
)
|
768 |
+
),
|
769 |
+
norm_f(
|
770 |
+
Conv2d(
|
771 |
+
128,
|
772 |
+
512,
|
773 |
+
(kernel_size, 1),
|
774 |
+
(stride, 1),
|
775 |
+
padding=(get_padding(kernel_size, 1), 0),
|
776 |
+
)
|
777 |
+
),
|
778 |
+
norm_f(
|
779 |
+
Conv2d(
|
780 |
+
512,
|
781 |
+
1024,
|
782 |
+
(kernel_size, 1),
|
783 |
+
(stride, 1),
|
784 |
+
padding=(get_padding(kernel_size, 1), 0),
|
785 |
+
)
|
786 |
+
),
|
787 |
+
norm_f(
|
788 |
+
Conv2d(
|
789 |
+
1024,
|
790 |
+
1024,
|
791 |
+
(kernel_size, 1),
|
792 |
+
1,
|
793 |
+
padding=(get_padding(kernel_size, 1), 0),
|
794 |
+
)
|
795 |
+
),
|
796 |
+
]
|
797 |
+
)
|
798 |
+
self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
|
799 |
+
|
800 |
+
def forward(self, x):
|
801 |
+
fmap = []
|
802 |
+
|
803 |
+
# 1d to 2d
|
804 |
+
b, c, t = x.shape
|
805 |
+
if t % self.period != 0: # pad first
|
806 |
+
n_pad = self.period - (t % self.period)
|
807 |
+
x = F.pad(x, (0, n_pad), "reflect")
|
808 |
+
t = t + n_pad
|
809 |
+
x = x.view(b, c, t // self.period, self.period)
|
810 |
+
|
811 |
+
for l in self.convs:
|
812 |
+
x = l(x)
|
813 |
+
x = F.leaky_relu(x, modules.LRELU_SLOPE)
|
814 |
+
fmap.append(x)
|
815 |
+
x = self.conv_post(x)
|
816 |
+
fmap.append(x)
|
817 |
+
x = torch.flatten(x, 1, -1)
|
818 |
+
|
819 |
+
return x, fmap
|
lib/infer_pack/modules.py
ADDED
@@ -0,0 +1,522 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import copy
|
2 |
+
import math
|
3 |
+
import numpy as np
|
4 |
+
import scipy
|
5 |
+
import torch
|
6 |
+
from torch import nn
|
7 |
+
from torch.nn import functional as F
|
8 |
+
|
9 |
+
from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
|
10 |
+
from torch.nn.utils import weight_norm, remove_weight_norm
|
11 |
+
|
12 |
+
from lib.infer_pack import commons
|
13 |
+
from lib.infer_pack.commons import init_weights, get_padding
|
14 |
+
from lib.infer_pack.transforms import piecewise_rational_quadratic_transform
|
15 |
+
|
16 |
+
|
17 |
+
LRELU_SLOPE = 0.1
|
18 |
+
|
19 |
+
|
20 |
+
class LayerNorm(nn.Module):
|
21 |
+
def __init__(self, channels, eps=1e-5):
|
22 |
+
super().__init__()
|
23 |
+
self.channels = channels
|
24 |
+
self.eps = eps
|
25 |
+
|
26 |
+
self.gamma = nn.Parameter(torch.ones(channels))
|
27 |
+
self.beta = nn.Parameter(torch.zeros(channels))
|
28 |
+
|
29 |
+
def forward(self, x):
|
30 |
+
x = x.transpose(1, -1)
|
31 |
+
x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
|
32 |
+
return x.transpose(1, -1)
|
33 |
+
|
34 |
+
|
35 |
+
class ConvReluNorm(nn.Module):
|
36 |
+
def __init__(
|
37 |
+
self,
|
38 |
+
in_channels,
|
39 |
+
hidden_channels,
|
40 |
+
out_channels,
|
41 |
+
kernel_size,
|
42 |
+
n_layers,
|
43 |
+
p_dropout,
|
44 |
+
):
|
45 |
+
super().__init__()
|
46 |
+
self.in_channels = in_channels
|
47 |
+
self.hidden_channels = hidden_channels
|
48 |
+
self.out_channels = out_channels
|
49 |
+
self.kernel_size = kernel_size
|
50 |
+
self.n_layers = n_layers
|
51 |
+
self.p_dropout = p_dropout
|
52 |
+
assert n_layers > 1, "Number of layers should be larger than 0."
|
53 |
+
|
54 |
+
self.conv_layers = nn.ModuleList()
|
55 |
+
self.norm_layers = nn.ModuleList()
|
56 |
+
self.conv_layers.append(
|
57 |
+
nn.Conv1d(
|
58 |
+
in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
|
59 |
+
)
|
60 |
+
)
|
61 |
+
self.norm_layers.append(LayerNorm(hidden_channels))
|
62 |
+
self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
|
63 |
+
for _ in range(n_layers - 1):
|
64 |
+
self.conv_layers.append(
|
65 |
+
nn.Conv1d(
|
66 |
+
hidden_channels,
|
67 |
+
hidden_channels,
|
68 |
+
kernel_size,
|
69 |
+
padding=kernel_size // 2,
|
70 |
+
)
|
71 |
+
)
|
72 |
+
self.norm_layers.append(LayerNorm(hidden_channels))
|
73 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
|
74 |
+
self.proj.weight.data.zero_()
|
75 |
+
self.proj.bias.data.zero_()
|
76 |
+
|
77 |
+
def forward(self, x, x_mask):
|
78 |
+
x_org = x
|
79 |
+
for i in range(self.n_layers):
|
80 |
+
x = self.conv_layers[i](x * x_mask)
|
81 |
+
x = self.norm_layers[i](x)
|
82 |
+
x = self.relu_drop(x)
|
83 |
+
x = x_org + self.proj(x)
|
84 |
+
return x * x_mask
|
85 |
+
|
86 |
+
|
87 |
+
class DDSConv(nn.Module):
|
88 |
+
"""
|
89 |
+
Dialted and Depth-Separable Convolution
|
90 |
+
"""
|
91 |
+
|
92 |
+
def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
|
93 |
+
super().__init__()
|
94 |
+
self.channels = channels
|
95 |
+
self.kernel_size = kernel_size
|
96 |
+
self.n_layers = n_layers
|
97 |
+
self.p_dropout = p_dropout
|
98 |
+
|
99 |
+
self.drop = nn.Dropout(p_dropout)
|
100 |
+
self.convs_sep = nn.ModuleList()
|
101 |
+
self.convs_1x1 = nn.ModuleList()
|
102 |
+
self.norms_1 = nn.ModuleList()
|
103 |
+
self.norms_2 = nn.ModuleList()
|
104 |
+
for i in range(n_layers):
|
105 |
+
dilation = kernel_size**i
|
106 |
+
padding = (kernel_size * dilation - dilation) // 2
|
107 |
+
self.convs_sep.append(
|
108 |
+
nn.Conv1d(
|
109 |
+
channels,
|
110 |
+
channels,
|
111 |
+
kernel_size,
|
112 |
+
groups=channels,
|
113 |
+
dilation=dilation,
|
114 |
+
padding=padding,
|
115 |
+
)
|
116 |
+
)
|
117 |
+
self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
|
118 |
+
self.norms_1.append(LayerNorm(channels))
|
119 |
+
self.norms_2.append(LayerNorm(channels))
|
120 |
+
|
121 |
+
def forward(self, x, x_mask, g=None):
|
122 |
+
if g is not None:
|
123 |
+
x = x + g
|
124 |
+
for i in range(self.n_layers):
|
125 |
+
y = self.convs_sep[i](x * x_mask)
|
126 |
+
y = self.norms_1[i](y)
|
127 |
+
y = F.gelu(y)
|
128 |
+
y = self.convs_1x1[i](y)
|
129 |
+
y = self.norms_2[i](y)
|
130 |
+
y = F.gelu(y)
|
131 |
+
y = self.drop(y)
|
132 |
+
x = x + y
|
133 |
+
return x * x_mask
|
134 |
+
|
135 |
+
|
136 |
+
class WN(torch.nn.Module):
|
137 |
+
def __init__(
|
138 |
+
self,
|
139 |
+
hidden_channels,
|
140 |
+
kernel_size,
|
141 |
+
dilation_rate,
|
142 |
+
n_layers,
|
143 |
+
gin_channels=0,
|
144 |
+
p_dropout=0,
|
145 |
+
):
|
146 |
+
super(WN, self).__init__()
|
147 |
+
assert kernel_size % 2 == 1
|
148 |
+
self.hidden_channels = hidden_channels
|
149 |
+
self.kernel_size = (kernel_size,)
|
150 |
+
self.dilation_rate = dilation_rate
|
151 |
+
self.n_layers = n_layers
|
152 |
+
self.gin_channels = gin_channels
|
153 |
+
self.p_dropout = p_dropout
|
154 |
+
|
155 |
+
self.in_layers = torch.nn.ModuleList()
|
156 |
+
self.res_skip_layers = torch.nn.ModuleList()
|
157 |
+
self.drop = nn.Dropout(p_dropout)
|
158 |
+
|
159 |
+
if gin_channels != 0:
|
160 |
+
cond_layer = torch.nn.Conv1d(
|
161 |
+
gin_channels, 2 * hidden_channels * n_layers, 1
|
162 |
+
)
|
163 |
+
self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
|
164 |
+
|
165 |
+
for i in range(n_layers):
|
166 |
+
dilation = dilation_rate**i
|
167 |
+
padding = int((kernel_size * dilation - dilation) / 2)
|
168 |
+
in_layer = torch.nn.Conv1d(
|
169 |
+
hidden_channels,
|
170 |
+
2 * hidden_channels,
|
171 |
+
kernel_size,
|
172 |
+
dilation=dilation,
|
173 |
+
padding=padding,
|
174 |
+
)
|
175 |
+
in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
|
176 |
+
self.in_layers.append(in_layer)
|
177 |
+
|
178 |
+
# last one is not necessary
|
179 |
+
if i < n_layers - 1:
|
180 |
+
res_skip_channels = 2 * hidden_channels
|
181 |
+
else:
|
182 |
+
res_skip_channels = hidden_channels
|
183 |
+
|
184 |
+
res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
|
185 |
+
res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
|
186 |
+
self.res_skip_layers.append(res_skip_layer)
|
187 |
+
|
188 |
+
def forward(self, x, x_mask, g=None, **kwargs):
|
189 |
+
output = torch.zeros_like(x)
|
190 |
+
n_channels_tensor = torch.IntTensor([self.hidden_channels])
|
191 |
+
|
192 |
+
if g is not None:
|
193 |
+
g = self.cond_layer(g)
|
194 |
+
|
195 |
+
for i in range(self.n_layers):
|
196 |
+
x_in = self.in_layers[i](x)
|
197 |
+
if g is not None:
|
198 |
+
cond_offset = i * 2 * self.hidden_channels
|
199 |
+
g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
|
200 |
+
else:
|
201 |
+
g_l = torch.zeros_like(x_in)
|
202 |
+
|
203 |
+
acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
|
204 |
+
acts = self.drop(acts)
|
205 |
+
|
206 |
+
res_skip_acts = self.res_skip_layers[i](acts)
|
207 |
+
if i < self.n_layers - 1:
|
208 |
+
res_acts = res_skip_acts[:, : self.hidden_channels, :]
|
209 |
+
x = (x + res_acts) * x_mask
|
210 |
+
output = output + res_skip_acts[:, self.hidden_channels :, :]
|
211 |
+
else:
|
212 |
+
output = output + res_skip_acts
|
213 |
+
return output * x_mask
|
214 |
+
|
215 |
+
def remove_weight_norm(self):
|
216 |
+
if self.gin_channels != 0:
|
217 |
+
torch.nn.utils.remove_weight_norm(self.cond_layer)
|
218 |
+
for l in self.in_layers:
|
219 |
+
torch.nn.utils.remove_weight_norm(l)
|
220 |
+
for l in self.res_skip_layers:
|
221 |
+
torch.nn.utils.remove_weight_norm(l)
|
222 |
+
|
223 |
+
|
224 |
+
class ResBlock1(torch.nn.Module):
|
225 |
+
def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
|
226 |
+
super(ResBlock1, self).__init__()
|
227 |
+
self.convs1 = nn.ModuleList(
|
228 |
+
[
|
229 |
+
weight_norm(
|
230 |
+
Conv1d(
|
231 |
+
channels,
|
232 |
+
channels,
|
233 |
+
kernel_size,
|
234 |
+
1,
|
235 |
+
dilation=dilation[0],
|
236 |
+
padding=get_padding(kernel_size, dilation[0]),
|
237 |
+
)
|
238 |
+
),
|
239 |
+
weight_norm(
|
240 |
+
Conv1d(
|
241 |
+
channels,
|
242 |
+
channels,
|
243 |
+
kernel_size,
|
244 |
+
1,
|
245 |
+
dilation=dilation[1],
|
246 |
+
padding=get_padding(kernel_size, dilation[1]),
|
247 |
+
)
|
248 |
+
),
|
249 |
+
weight_norm(
|
250 |
+
Conv1d(
|
251 |
+
channels,
|
252 |
+
channels,
|
253 |
+
kernel_size,
|
254 |
+
1,
|
255 |
+
dilation=dilation[2],
|
256 |
+
padding=get_padding(kernel_size, dilation[2]),
|
257 |
+
)
|
258 |
+
),
|
259 |
+
]
|
260 |
+
)
|
261 |
+
self.convs1.apply(init_weights)
|
262 |
+
|
263 |
+
self.convs2 = nn.ModuleList(
|
264 |
+
[
|
265 |
+
weight_norm(
|
266 |
+
Conv1d(
|
267 |
+
channels,
|
268 |
+
channels,
|
269 |
+
kernel_size,
|
270 |
+
1,
|
271 |
+
dilation=1,
|
272 |
+
padding=get_padding(kernel_size, 1),
|
273 |
+
)
|
274 |
+
),
|
275 |
+
weight_norm(
|
276 |
+
Conv1d(
|
277 |
+
channels,
|
278 |
+
channels,
|
279 |
+
kernel_size,
|
280 |
+
1,
|
281 |
+
dilation=1,
|
282 |
+
padding=get_padding(kernel_size, 1),
|
283 |
+
)
|
284 |
+
),
|
285 |
+
weight_norm(
|
286 |
+
Conv1d(
|
287 |
+
channels,
|
288 |
+
channels,
|
289 |
+
kernel_size,
|
290 |
+
1,
|
291 |
+
dilation=1,
|
292 |
+
padding=get_padding(kernel_size, 1),
|
293 |
+
)
|
294 |
+
),
|
295 |
+
]
|
296 |
+
)
|
297 |
+
self.convs2.apply(init_weights)
|
298 |
+
|
299 |
+
def forward(self, x, x_mask=None):
|
300 |
+
for c1, c2 in zip(self.convs1, self.convs2):
|
301 |
+
xt = F.leaky_relu(x, LRELU_SLOPE)
|
302 |
+
if x_mask is not None:
|
303 |
+
xt = xt * x_mask
|
304 |
+
xt = c1(xt)
|
305 |
+
xt = F.leaky_relu(xt, LRELU_SLOPE)
|
306 |
+
if x_mask is not None:
|
307 |
+
xt = xt * x_mask
|
308 |
+
xt = c2(xt)
|
309 |
+
x = xt + x
|
310 |
+
if x_mask is not None:
|
311 |
+
x = x * x_mask
|
312 |
+
return x
|
313 |
+
|
314 |
+
def remove_weight_norm(self):
|
315 |
+
for l in self.convs1:
|
316 |
+
remove_weight_norm(l)
|
317 |
+
for l in self.convs2:
|
318 |
+
remove_weight_norm(l)
|
319 |
+
|
320 |
+
|
321 |
+
class ResBlock2(torch.nn.Module):
|
322 |
+
def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
|
323 |
+
super(ResBlock2, self).__init__()
|
324 |
+
self.convs = nn.ModuleList(
|
325 |
+
[
|
326 |
+
weight_norm(
|
327 |
+
Conv1d(
|
328 |
+
channels,
|
329 |
+
channels,
|
330 |
+
kernel_size,
|
331 |
+
1,
|
332 |
+
dilation=dilation[0],
|
333 |
+
padding=get_padding(kernel_size, dilation[0]),
|
334 |
+
)
|
335 |
+
),
|
336 |
+
weight_norm(
|
337 |
+
Conv1d(
|
338 |
+
channels,
|
339 |
+
channels,
|
340 |
+
kernel_size,
|
341 |
+
1,
|
342 |
+
dilation=dilation[1],
|
343 |
+
padding=get_padding(kernel_size, dilation[1]),
|
344 |
+
)
|
345 |
+
),
|
346 |
+
]
|
347 |
+
)
|
348 |
+
self.convs.apply(init_weights)
|
349 |
+
|
350 |
+
def forward(self, x, x_mask=None):
|
351 |
+
for c in self.convs:
|
352 |
+
xt = F.leaky_relu(x, LRELU_SLOPE)
|
353 |
+
if x_mask is not None:
|
354 |
+
xt = xt * x_mask
|
355 |
+
xt = c(xt)
|
356 |
+
x = xt + x
|
357 |
+
if x_mask is not None:
|
358 |
+
x = x * x_mask
|
359 |
+
return x
|
360 |
+
|
361 |
+
def remove_weight_norm(self):
|
362 |
+
for l in self.convs:
|
363 |
+
remove_weight_norm(l)
|
364 |
+
|
365 |
+
|
366 |
+
class Log(nn.Module):
|
367 |
+
def forward(self, x, x_mask, reverse=False, **kwargs):
|
368 |
+
if not reverse:
|
369 |
+
y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
|
370 |
+
logdet = torch.sum(-y, [1, 2])
|
371 |
+
return y, logdet
|
372 |
+
else:
|
373 |
+
x = torch.exp(x) * x_mask
|
374 |
+
return x
|
375 |
+
|
376 |
+
|
377 |
+
class Flip(nn.Module):
|
378 |
+
def forward(self, x, *args, reverse=False, **kwargs):
|
379 |
+
x = torch.flip(x, [1])
|
380 |
+
if not reverse:
|
381 |
+
logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
|
382 |
+
return x, logdet
|
383 |
+
else:
|
384 |
+
return x
|
385 |
+
|
386 |
+
|
387 |
+
class ElementwiseAffine(nn.Module):
|
388 |
+
def __init__(self, channels):
|
389 |
+
super().__init__()
|
390 |
+
self.channels = channels
|
391 |
+
self.m = nn.Parameter(torch.zeros(channels, 1))
|
392 |
+
self.logs = nn.Parameter(torch.zeros(channels, 1))
|
393 |
+
|
394 |
+
def forward(self, x, x_mask, reverse=False, **kwargs):
|
395 |
+
if not reverse:
|
396 |
+
y = self.m + torch.exp(self.logs) * x
|
397 |
+
y = y * x_mask
|
398 |
+
logdet = torch.sum(self.logs * x_mask, [1, 2])
|
399 |
+
return y, logdet
|
400 |
+
else:
|
401 |
+
x = (x - self.m) * torch.exp(-self.logs) * x_mask
|
402 |
+
return x
|
403 |
+
|
404 |
+
|
405 |
+
class ResidualCouplingLayer(nn.Module):
|
406 |
+
def __init__(
|
407 |
+
self,
|
408 |
+
channels,
|
409 |
+
hidden_channels,
|
410 |
+
kernel_size,
|
411 |
+
dilation_rate,
|
412 |
+
n_layers,
|
413 |
+
p_dropout=0,
|
414 |
+
gin_channels=0,
|
415 |
+
mean_only=False,
|
416 |
+
):
|
417 |
+
assert channels % 2 == 0, "channels should be divisible by 2"
|
418 |
+
super().__init__()
|
419 |
+
self.channels = channels
|
420 |
+
self.hidden_channels = hidden_channels
|
421 |
+
self.kernel_size = kernel_size
|
422 |
+
self.dilation_rate = dilation_rate
|
423 |
+
self.n_layers = n_layers
|
424 |
+
self.half_channels = channels // 2
|
425 |
+
self.mean_only = mean_only
|
426 |
+
|
427 |
+
self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
|
428 |
+
self.enc = WN(
|
429 |
+
hidden_channels,
|
430 |
+
kernel_size,
|
431 |
+
dilation_rate,
|
432 |
+
n_layers,
|
433 |
+
p_dropout=p_dropout,
|
434 |
+
gin_channels=gin_channels,
|
435 |
+
)
|
436 |
+
self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
|
437 |
+
self.post.weight.data.zero_()
|
438 |
+
self.post.bias.data.zero_()
|
439 |
+
|
440 |
+
def forward(self, x, x_mask, g=None, reverse=False):
|
441 |
+
x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
|
442 |
+
h = self.pre(x0) * x_mask
|
443 |
+
h = self.enc(h, x_mask, g=g)
|
444 |
+
stats = self.post(h) * x_mask
|
445 |
+
if not self.mean_only:
|
446 |
+
m, logs = torch.split(stats, [self.half_channels] * 2, 1)
|
447 |
+
else:
|
448 |
+
m = stats
|
449 |
+
logs = torch.zeros_like(m)
|
450 |
+
|
451 |
+
if not reverse:
|
452 |
+
x1 = m + x1 * torch.exp(logs) * x_mask
|
453 |
+
x = torch.cat([x0, x1], 1)
|
454 |
+
logdet = torch.sum(logs, [1, 2])
|
455 |
+
return x, logdet
|
456 |
+
else:
|
457 |
+
x1 = (x1 - m) * torch.exp(-logs) * x_mask
|
458 |
+
x = torch.cat([x0, x1], 1)
|
459 |
+
return x
|
460 |
+
|
461 |
+
def remove_weight_norm(self):
|
462 |
+
self.enc.remove_weight_norm()
|
463 |
+
|
464 |
+
|
465 |
+
class ConvFlow(nn.Module):
|
466 |
+
def __init__(
|
467 |
+
self,
|
468 |
+
in_channels,
|
469 |
+
filter_channels,
|
470 |
+
kernel_size,
|
471 |
+
n_layers,
|
472 |
+
num_bins=10,
|
473 |
+
tail_bound=5.0,
|
474 |
+
):
|
475 |
+
super().__init__()
|
476 |
+
self.in_channels = in_channels
|
477 |
+
self.filter_channels = filter_channels
|
478 |
+
self.kernel_size = kernel_size
|
479 |
+
self.n_layers = n_layers
|
480 |
+
self.num_bins = num_bins
|
481 |
+
self.tail_bound = tail_bound
|
482 |
+
self.half_channels = in_channels // 2
|
483 |
+
|
484 |
+
self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
|
485 |
+
self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
|
486 |
+
self.proj = nn.Conv1d(
|
487 |
+
filter_channels, self.half_channels * (num_bins * 3 - 1), 1
|
488 |
+
)
|
489 |
+
self.proj.weight.data.zero_()
|
490 |
+
self.proj.bias.data.zero_()
|
491 |
+
|
492 |
+
def forward(self, x, x_mask, g=None, reverse=False):
|
493 |
+
x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
|
494 |
+
h = self.pre(x0)
|
495 |
+
h = self.convs(h, x_mask, g=g)
|
496 |
+
h = self.proj(h) * x_mask
|
497 |
+
|
498 |
+
b, c, t = x0.shape
|
499 |
+
h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2) # [b, cx?, t] -> [b, c, t, ?]
|
500 |
+
|
501 |
+
unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
|
502 |
+
unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
|
503 |
+
self.filter_channels
|
504 |
+
)
|
505 |
+
unnormalized_derivatives = h[..., 2 * self.num_bins :]
|
506 |
+
|
507 |
+
x1, logabsdet = piecewise_rational_quadratic_transform(
|
508 |
+
x1,
|
509 |
+
unnormalized_widths,
|
510 |
+
unnormalized_heights,
|
511 |
+
unnormalized_derivatives,
|
512 |
+
inverse=reverse,
|
513 |
+
tails="linear",
|
514 |
+
tail_bound=self.tail_bound,
|
515 |
+
)
|
516 |
+
|
517 |
+
x = torch.cat([x0, x1], 1) * x_mask
|
518 |
+
logdet = torch.sum(logabsdet * x_mask, [1, 2])
|
519 |
+
if not reverse:
|
520 |
+
return x, logdet
|
521 |
+
else:
|
522 |
+
return x
|
lib/infer_pack/onnx_inference.py
ADDED
@@ -0,0 +1,145 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import onnxruntime
|
2 |
+
import librosa
|
3 |
+
import numpy as np
|
4 |
+
import soundfile
|
5 |
+
|
6 |
+
|
7 |
+
class ContentVec:
|
8 |
+
def __init__(self, vec_path="pretrained/vec-768-layer-12.onnx", device=None):
|
9 |
+
print("load model(s) from {}".format(vec_path))
|
10 |
+
if device == "cpu" or device is None:
|
11 |
+
providers = ["CPUExecutionProvider"]
|
12 |
+
elif device == "cuda":
|
13 |
+
providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
|
14 |
+
elif device == "dml":
|
15 |
+
providers = ["DmlExecutionProvider"]
|
16 |
+
else:
|
17 |
+
raise RuntimeError("Unsportted Device")
|
18 |
+
self.model = onnxruntime.InferenceSession(vec_path, providers=providers)
|
19 |
+
|
20 |
+
def __call__(self, wav):
|
21 |
+
return self.forward(wav)
|
22 |
+
|
23 |
+
def forward(self, wav):
|
24 |
+
feats = wav
|
25 |
+
if feats.ndim == 2: # double channels
|
26 |
+
feats = feats.mean(-1)
|
27 |
+
assert feats.ndim == 1, feats.ndim
|
28 |
+
feats = np.expand_dims(np.expand_dims(feats, 0), 0)
|
29 |
+
onnx_input = {self.model.get_inputs()[0].name: feats}
|
30 |
+
logits = self.model.run(None, onnx_input)[0]
|
31 |
+
return logits.transpose(0, 2, 1)
|
32 |
+
|
33 |
+
|
34 |
+
def get_f0_predictor(f0_predictor, hop_length, sampling_rate, **kargs):
|
35 |
+
if f0_predictor == "pm":
|
36 |
+
from lib.infer_pack.modules.F0Predictor.PMF0Predictor import PMF0Predictor
|
37 |
+
|
38 |
+
f0_predictor_object = PMF0Predictor(
|
39 |
+
hop_length=hop_length, sampling_rate=sampling_rate
|
40 |
+
)
|
41 |
+
elif f0_predictor == "harvest":
|
42 |
+
from lib.infer_pack.modules.F0Predictor.HarvestF0Predictor import (
|
43 |
+
HarvestF0Predictor,
|
44 |
+
)
|
45 |
+
|
46 |
+
f0_predictor_object = HarvestF0Predictor(
|
47 |
+
hop_length=hop_length, sampling_rate=sampling_rate
|
48 |
+
)
|
49 |
+
elif f0_predictor == "dio":
|
50 |
+
from lib.infer_pack.modules.F0Predictor.DioF0Predictor import DioF0Predictor
|
51 |
+
|
52 |
+
f0_predictor_object = DioF0Predictor(
|
53 |
+
hop_length=hop_length, sampling_rate=sampling_rate
|
54 |
+
)
|
55 |
+
else:
|
56 |
+
raise Exception("Unknown f0 predictor")
|
57 |
+
return f0_predictor_object
|
58 |
+
|
59 |
+
|
60 |
+
class OnnxRVC:
|
61 |
+
def __init__(
|
62 |
+
self,
|
63 |
+
model_path,
|
64 |
+
sr=40000,
|
65 |
+
hop_size=512,
|
66 |
+
vec_path="vec-768-layer-12",
|
67 |
+
device="cpu",
|
68 |
+
):
|
69 |
+
vec_path = f"pretrained/{vec_path}.onnx"
|
70 |
+
self.vec_model = ContentVec(vec_path, device)
|
71 |
+
if device == "cpu" or device is None:
|
72 |
+
providers = ["CPUExecutionProvider"]
|
73 |
+
elif device == "cuda":
|
74 |
+
providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
|
75 |
+
elif device == "dml":
|
76 |
+
providers = ["DmlExecutionProvider"]
|
77 |
+
else:
|
78 |
+
raise RuntimeError("Unsportted Device")
|
79 |
+
self.model = onnxruntime.InferenceSession(model_path, providers=providers)
|
80 |
+
self.sampling_rate = sr
|
81 |
+
self.hop_size = hop_size
|
82 |
+
|
83 |
+
def forward(self, hubert, hubert_length, pitch, pitchf, ds, rnd):
|
84 |
+
onnx_input = {
|
85 |
+
self.model.get_inputs()[0].name: hubert,
|
86 |
+
self.model.get_inputs()[1].name: hubert_length,
|
87 |
+
self.model.get_inputs()[2].name: pitch,
|
88 |
+
self.model.get_inputs()[3].name: pitchf,
|
89 |
+
self.model.get_inputs()[4].name: ds,
|
90 |
+
self.model.get_inputs()[5].name: rnd,
|
91 |
+
}
|
92 |
+
return (self.model.run(None, onnx_input)[0] * 32767).astype(np.int16)
|
93 |
+
|
94 |
+
def inference(
|
95 |
+
self,
|
96 |
+
raw_path,
|
97 |
+
sid,
|
98 |
+
f0_method="dio",
|
99 |
+
f0_up_key=0,
|
100 |
+
pad_time=0.5,
|
101 |
+
cr_threshold=0.02,
|
102 |
+
):
|
103 |
+
f0_min = 50
|
104 |
+
f0_max = 1100
|
105 |
+
f0_mel_min = 1127 * np.log(1 + f0_min / 700)
|
106 |
+
f0_mel_max = 1127 * np.log(1 + f0_max / 700)
|
107 |
+
f0_predictor = get_f0_predictor(
|
108 |
+
f0_method,
|
109 |
+
hop_length=self.hop_size,
|
110 |
+
sampling_rate=self.sampling_rate,
|
111 |
+
threshold=cr_threshold,
|
112 |
+
)
|
113 |
+
wav, sr = librosa.load(raw_path, sr=self.sampling_rate)
|
114 |
+
org_length = len(wav)
|
115 |
+
if org_length / sr > 50.0:
|
116 |
+
raise RuntimeError("Reached Max Length")
|
117 |
+
|
118 |
+
wav16k = librosa.resample(wav, orig_sr=self.sampling_rate, target_sr=16000)
|
119 |
+
wav16k = wav16k
|
120 |
+
|
121 |
+
hubert = self.vec_model(wav16k)
|
122 |
+
hubert = np.repeat(hubert, 2, axis=2).transpose(0, 2, 1).astype(np.float32)
|
123 |
+
hubert_length = hubert.shape[1]
|
124 |
+
|
125 |
+
pitchf = f0_predictor.compute_f0(wav, hubert_length)
|
126 |
+
pitchf = pitchf * 2 ** (f0_up_key / 12)
|
127 |
+
pitch = pitchf.copy()
|
128 |
+
f0_mel = 1127 * np.log(1 + pitch / 700)
|
129 |
+
f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
|
130 |
+
f0_mel_max - f0_mel_min
|
131 |
+
) + 1
|
132 |
+
f0_mel[f0_mel <= 1] = 1
|
133 |
+
f0_mel[f0_mel > 255] = 255
|
134 |
+
pitch = np.rint(f0_mel).astype(np.int64)
|
135 |
+
|
136 |
+
pitchf = pitchf.reshape(1, len(pitchf)).astype(np.float32)
|
137 |
+
pitch = pitch.reshape(1, len(pitch))
|
138 |
+
ds = np.array([sid]).astype(np.int64)
|
139 |
+
|
140 |
+
rnd = np.random.randn(1, 192, hubert_length).astype(np.float32)
|
141 |
+
hubert_length = np.array([hubert_length]).astype(np.int64)
|
142 |
+
|
143 |
+
out_wav = self.forward(hubert, hubert_length, pitch, pitchf, ds, rnd).squeeze()
|
144 |
+
out_wav = np.pad(out_wav, (0, 2 * self.hop_size), "constant")
|
145 |
+
return out_wav[0:org_length]
|
lib/infer_pack/transforms.py
ADDED
@@ -0,0 +1,209 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch.nn import functional as F
|
3 |
+
|
4 |
+
import numpy as np
|
5 |
+
|
6 |
+
|
7 |
+
DEFAULT_MIN_BIN_WIDTH = 1e-3
|
8 |
+
DEFAULT_MIN_BIN_HEIGHT = 1e-3
|
9 |
+
DEFAULT_MIN_DERIVATIVE = 1e-3
|
10 |
+
|
11 |
+
|
12 |
+
def piecewise_rational_quadratic_transform(
|
13 |
+
inputs,
|
14 |
+
unnormalized_widths,
|
15 |
+
unnormalized_heights,
|
16 |
+
unnormalized_derivatives,
|
17 |
+
inverse=False,
|
18 |
+
tails=None,
|
19 |
+
tail_bound=1.0,
|
20 |
+
min_bin_width=DEFAULT_MIN_BIN_WIDTH,
|
21 |
+
min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
|
22 |
+
min_derivative=DEFAULT_MIN_DERIVATIVE,
|
23 |
+
):
|
24 |
+
if tails is None:
|
25 |
+
spline_fn = rational_quadratic_spline
|
26 |
+
spline_kwargs = {}
|
27 |
+
else:
|
28 |
+
spline_fn = unconstrained_rational_quadratic_spline
|
29 |
+
spline_kwargs = {"tails": tails, "tail_bound": tail_bound}
|
30 |
+
|
31 |
+
outputs, logabsdet = spline_fn(
|
32 |
+
inputs=inputs,
|
33 |
+
unnormalized_widths=unnormalized_widths,
|
34 |
+
unnormalized_heights=unnormalized_heights,
|
35 |
+
unnormalized_derivatives=unnormalized_derivatives,
|
36 |
+
inverse=inverse,
|
37 |
+
min_bin_width=min_bin_width,
|
38 |
+
min_bin_height=min_bin_height,
|
39 |
+
min_derivative=min_derivative,
|
40 |
+
**spline_kwargs
|
41 |
+
)
|
42 |
+
return outputs, logabsdet
|
43 |
+
|
44 |
+
|
45 |
+
def searchsorted(bin_locations, inputs, eps=1e-6):
|
46 |
+
bin_locations[..., -1] += eps
|
47 |
+
return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
|
48 |
+
|
49 |
+
|
50 |
+
def unconstrained_rational_quadratic_spline(
|
51 |
+
inputs,
|
52 |
+
unnormalized_widths,
|
53 |
+
unnormalized_heights,
|
54 |
+
unnormalized_derivatives,
|
55 |
+
inverse=False,
|
56 |
+
tails="linear",
|
57 |
+
tail_bound=1.0,
|
58 |
+
min_bin_width=DEFAULT_MIN_BIN_WIDTH,
|
59 |
+
min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
|
60 |
+
min_derivative=DEFAULT_MIN_DERIVATIVE,
|
61 |
+
):
|
62 |
+
inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
|
63 |
+
outside_interval_mask = ~inside_interval_mask
|
64 |
+
|
65 |
+
outputs = torch.zeros_like(inputs)
|
66 |
+
logabsdet = torch.zeros_like(inputs)
|
67 |
+
|
68 |
+
if tails == "linear":
|
69 |
+
unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
|
70 |
+
constant = np.log(np.exp(1 - min_derivative) - 1)
|
71 |
+
unnormalized_derivatives[..., 0] = constant
|
72 |
+
unnormalized_derivatives[..., -1] = constant
|
73 |
+
|
74 |
+
outputs[outside_interval_mask] = inputs[outside_interval_mask]
|
75 |
+
logabsdet[outside_interval_mask] = 0
|
76 |
+
else:
|
77 |
+
raise RuntimeError("{} tails are not implemented.".format(tails))
|
78 |
+
|
79 |
+
(
|
80 |
+
outputs[inside_interval_mask],
|
81 |
+
logabsdet[inside_interval_mask],
|
82 |
+
) = rational_quadratic_spline(
|
83 |
+
inputs=inputs[inside_interval_mask],
|
84 |
+
unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
|
85 |
+
unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
|
86 |
+
unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
|
87 |
+
inverse=inverse,
|
88 |
+
left=-tail_bound,
|
89 |
+
right=tail_bound,
|
90 |
+
bottom=-tail_bound,
|
91 |
+
top=tail_bound,
|
92 |
+
min_bin_width=min_bin_width,
|
93 |
+
min_bin_height=min_bin_height,
|
94 |
+
min_derivative=min_derivative,
|
95 |
+
)
|
96 |
+
|
97 |
+
return outputs, logabsdet
|
98 |
+
|
99 |
+
|
100 |
+
def rational_quadratic_spline(
|
101 |
+
inputs,
|
102 |
+
unnormalized_widths,
|
103 |
+
unnormalized_heights,
|
104 |
+
unnormalized_derivatives,
|
105 |
+
inverse=False,
|
106 |
+
left=0.0,
|
107 |
+
right=1.0,
|
108 |
+
bottom=0.0,
|
109 |
+
top=1.0,
|
110 |
+
min_bin_width=DEFAULT_MIN_BIN_WIDTH,
|
111 |
+
min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
|
112 |
+
min_derivative=DEFAULT_MIN_DERIVATIVE,
|
113 |
+
):
|
114 |
+
if torch.min(inputs) < left or torch.max(inputs) > right:
|
115 |
+
raise ValueError("Input to a transform is not within its domain")
|
116 |
+
|
117 |
+
num_bins = unnormalized_widths.shape[-1]
|
118 |
+
|
119 |
+
if min_bin_width * num_bins > 1.0:
|
120 |
+
raise ValueError("Minimal bin width too large for the number of bins")
|
121 |
+
if min_bin_height * num_bins > 1.0:
|
122 |
+
raise ValueError("Minimal bin height too large for the number of bins")
|
123 |
+
|
124 |
+
widths = F.softmax(unnormalized_widths, dim=-1)
|
125 |
+
widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
|
126 |
+
cumwidths = torch.cumsum(widths, dim=-1)
|
127 |
+
cumwidths = F.pad(cumwidths, pad=(1, 0), mode="constant", value=0.0)
|
128 |
+
cumwidths = (right - left) * cumwidths + left
|
129 |
+
cumwidths[..., 0] = left
|
130 |
+
cumwidths[..., -1] = right
|
131 |
+
widths = cumwidths[..., 1:] - cumwidths[..., :-1]
|
132 |
+
|
133 |
+
derivatives = min_derivative + F.softplus(unnormalized_derivatives)
|
134 |
+
|
135 |
+
heights = F.softmax(unnormalized_heights, dim=-1)
|
136 |
+
heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
|
137 |
+
cumheights = torch.cumsum(heights, dim=-1)
|
138 |
+
cumheights = F.pad(cumheights, pad=(1, 0), mode="constant", value=0.0)
|
139 |
+
cumheights = (top - bottom) * cumheights + bottom
|
140 |
+
cumheights[..., 0] = bottom
|
141 |
+
cumheights[..., -1] = top
|
142 |
+
heights = cumheights[..., 1:] - cumheights[..., :-1]
|
143 |
+
|
144 |
+
if inverse:
|
145 |
+
bin_idx = searchsorted(cumheights, inputs)[..., None]
|
146 |
+
else:
|
147 |
+
bin_idx = searchsorted(cumwidths, inputs)[..., None]
|
148 |
+
|
149 |
+
input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
|
150 |
+
input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
|
151 |
+
|
152 |
+
input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
|
153 |
+
delta = heights / widths
|
154 |
+
input_delta = delta.gather(-1, bin_idx)[..., 0]
|
155 |
+
|
156 |
+
input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
|
157 |
+
input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
|
158 |
+
|
159 |
+
input_heights = heights.gather(-1, bin_idx)[..., 0]
|
160 |
+
|
161 |
+
if inverse:
|
162 |
+
a = (inputs - input_cumheights) * (
|
163 |
+
input_derivatives + input_derivatives_plus_one - 2 * input_delta
|
164 |
+
) + input_heights * (input_delta - input_derivatives)
|
165 |
+
b = input_heights * input_derivatives - (inputs - input_cumheights) * (
|
166 |
+
input_derivatives + input_derivatives_plus_one - 2 * input_delta
|
167 |
+
)
|
168 |
+
c = -input_delta * (inputs - input_cumheights)
|
169 |
+
|
170 |
+
discriminant = b.pow(2) - 4 * a * c
|
171 |
+
assert (discriminant >= 0).all()
|
172 |
+
|
173 |
+
root = (2 * c) / (-b - torch.sqrt(discriminant))
|
174 |
+
outputs = root * input_bin_widths + input_cumwidths
|
175 |
+
|
176 |
+
theta_one_minus_theta = root * (1 - root)
|
177 |
+
denominator = input_delta + (
|
178 |
+
(input_derivatives + input_derivatives_plus_one - 2 * input_delta)
|
179 |
+
* theta_one_minus_theta
|
180 |
+
)
|
181 |
+
derivative_numerator = input_delta.pow(2) * (
|
182 |
+
input_derivatives_plus_one * root.pow(2)
|
183 |
+
+ 2 * input_delta * theta_one_minus_theta
|
184 |
+
+ input_derivatives * (1 - root).pow(2)
|
185 |
+
)
|
186 |
+
logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
|
187 |
+
|
188 |
+
return outputs, -logabsdet
|
189 |
+
else:
|
190 |
+
theta = (inputs - input_cumwidths) / input_bin_widths
|
191 |
+
theta_one_minus_theta = theta * (1 - theta)
|
192 |
+
|
193 |
+
numerator = input_heights * (
|
194 |
+
input_delta * theta.pow(2) + input_derivatives * theta_one_minus_theta
|
195 |
+
)
|
196 |
+
denominator = input_delta + (
|
197 |
+
(input_derivatives + input_derivatives_plus_one - 2 * input_delta)
|
198 |
+
* theta_one_minus_theta
|
199 |
+
)
|
200 |
+
outputs = input_cumheights + numerator / denominator
|
201 |
+
|
202 |
+
derivative_numerator = input_delta.pow(2) * (
|
203 |
+
input_derivatives_plus_one * theta.pow(2)
|
204 |
+
+ 2 * input_delta * theta_one_minus_theta
|
205 |
+
+ input_derivatives * (1 - theta).pow(2)
|
206 |
+
)
|
207 |
+
logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
|
208 |
+
|
209 |
+
return outputs, logabsdet
|