Update ONNXVITS_infer.py
Browse files- ONNXVITS_infer.py +127 -8
ONNXVITS_infer.py
CHANGED
@@ -1,6 +1,102 @@
|
|
1 |
import torch
|
2 |
import commons
|
3 |
import models
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
class SynthesizerTrn(models.SynthesizerTrn):
|
5 |
"""
|
6 |
Synthesizer for Training
|
@@ -26,6 +122,7 @@ class SynthesizerTrn(models.SynthesizerTrn):
|
|
26 |
n_speakers=0,
|
27 |
gin_channels=0,
|
28 |
use_sdp=True,
|
|
|
29 |
**kwargs):
|
30 |
|
31 |
super().__init__(
|
@@ -50,16 +147,21 @@ class SynthesizerTrn(models.SynthesizerTrn):
|
|
50 |
use_sdp=use_sdp,
|
51 |
**kwargs
|
52 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
-
def infer(self, x, x_lengths, sid=None, noise_scale=1, length_scale=1, noise_scale_w=1., max_len=None):
|
55 |
from ONNXVITS_utils import runonnx
|
56 |
|
57 |
-
|
58 |
-
x, m_p, logs_p, x_mask = runonnx("ONNX_net/enc_p.onnx", x=x.numpy(), x_lengths=x_lengths.numpy())
|
59 |
-
x = torch.from_numpy(x)
|
60 |
-
m_p = torch.from_numpy(m_p)
|
61 |
-
logs_p = torch.from_numpy(logs_p)
|
62 |
-
x_mask = torch.from_numpy(x_mask)
|
63 |
|
64 |
if self.n_speakers > 0:
|
65 |
g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
|
@@ -151,4 +253,21 @@ class SynthesizerTrn(models.SynthesizerTrn):
|
|
151 |
o = runonnx("ONNX_net/dec.onnx", z_in=(z * y_mask)[:,:,:max_len].numpy(), g=g.numpy())
|
152 |
o = torch.from_numpy(o[0])
|
153 |
|
154 |
-
return o, attn, y_mask, (z, z_p, m_p, logs_p)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import torch
|
2 |
import commons
|
3 |
import models
|
4 |
+
|
5 |
+
import math
|
6 |
+
from torch import nn
|
7 |
+
from torch.nn import functional as F
|
8 |
+
|
9 |
+
import modules
|
10 |
+
import attentions
|
11 |
+
import monotonic_align
|
12 |
+
|
13 |
+
from torch.nn import Conv1d, ConvTranspose1d, Conv2d
|
14 |
+
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
|
15 |
+
from commons import init_weights, get_padding
|
16 |
+
|
17 |
+
class TextEncoder(nn.Module):
|
18 |
+
def __init__(self,
|
19 |
+
n_vocab,
|
20 |
+
out_channels,
|
21 |
+
hidden_channels,
|
22 |
+
filter_channels,
|
23 |
+
n_heads,
|
24 |
+
n_layers,
|
25 |
+
kernel_size,
|
26 |
+
p_dropout,
|
27 |
+
emotion_embedding):
|
28 |
+
super().__init__()
|
29 |
+
self.n_vocab = n_vocab
|
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.emotion_embedding = emotion_embedding
|
38 |
+
|
39 |
+
if self.n_vocab!=0:
|
40 |
+
self.emb = nn.Embedding(n_vocab, hidden_channels)
|
41 |
+
if emotion_embedding:
|
42 |
+
self.emo_proj = nn.Linear(1024, hidden_channels)
|
43 |
+
nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
|
44 |
+
|
45 |
+
self.encoder = attentions.Encoder(
|
46 |
+
hidden_channels,
|
47 |
+
filter_channels,
|
48 |
+
n_heads,
|
49 |
+
n_layers,
|
50 |
+
kernel_size,
|
51 |
+
p_dropout)
|
52 |
+
self.proj= nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
53 |
+
|
54 |
+
def forward(self, x, x_lengths, emotion_embedding=None):
|
55 |
+
if self.n_vocab!=0:
|
56 |
+
x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h]
|
57 |
+
if emotion_embedding is not None:
|
58 |
+
print("emotion added")
|
59 |
+
x = x + self.emo_proj(emotion_embedding.unsqueeze(1))
|
60 |
+
x = torch.transpose(x, 1, -1) # [b, h, t]
|
61 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
|
62 |
+
|
63 |
+
x = self.encoder(x * x_mask, x_mask)
|
64 |
+
stats = self.proj(x) * x_mask
|
65 |
+
|
66 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
67 |
+
return x, m, logs, x_mask
|
68 |
+
|
69 |
+
class PosteriorEncoder(nn.Module):
|
70 |
+
def __init__(self,
|
71 |
+
in_channels,
|
72 |
+
out_channels,
|
73 |
+
hidden_channels,
|
74 |
+
kernel_size,
|
75 |
+
dilation_rate,
|
76 |
+
n_layers,
|
77 |
+
gin_channels=0):
|
78 |
+
super().__init__()
|
79 |
+
self.in_channels = in_channels
|
80 |
+
self.out_channels = out_channels
|
81 |
+
self.hidden_channels = hidden_channels
|
82 |
+
self.kernel_size = kernel_size
|
83 |
+
self.dilation_rate = dilation_rate
|
84 |
+
self.n_layers = n_layers
|
85 |
+
self.gin_channels = gin_channels
|
86 |
+
|
87 |
+
self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
|
88 |
+
self.enc = modules.WN(hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels)
|
89 |
+
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
|
90 |
+
|
91 |
+
def forward(self, x, x_lengths, g=None):
|
92 |
+
x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
|
93 |
+
x = self.pre(x) * x_mask
|
94 |
+
x = self.enc(x, x_mask, g=g)
|
95 |
+
stats = self.proj(x) * x_mask
|
96 |
+
m, logs = torch.split(stats, self.out_channels, dim=1)
|
97 |
+
z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
|
98 |
+
return z, m, logs, x_mask
|
99 |
+
|
100 |
class SynthesizerTrn(models.SynthesizerTrn):
|
101 |
"""
|
102 |
Synthesizer for Training
|
|
|
122 |
n_speakers=0,
|
123 |
gin_channels=0,
|
124 |
use_sdp=True,
|
125 |
+
emotion_embedding=False,
|
126 |
**kwargs):
|
127 |
|
128 |
super().__init__(
|
|
|
147 |
use_sdp=use_sdp,
|
148 |
**kwargs
|
149 |
)
|
150 |
+
self.enc_p = TextEncoder(n_vocab,
|
151 |
+
inter_channels,
|
152 |
+
hidden_channels,
|
153 |
+
filter_channels,
|
154 |
+
n_heads,
|
155 |
+
n_layers,
|
156 |
+
kernel_size,
|
157 |
+
p_dropout,
|
158 |
+
emotion_embedding)
|
159 |
+
self.enc_q = PosteriorEncoder(spec_channels, inter_channels, hidden_channels, 5, 1, 16, gin_channels=gin_channels)
|
160 |
|
161 |
+
def infer(self, x, x_lengths, sid=None, noise_scale=1, length_scale=1, noise_scale_w=1., max_len=None, emotion_embedding=None):
|
162 |
from ONNXVITS_utils import runonnx
|
163 |
|
164 |
+
x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths, emotion_embedding)
|
|
|
|
|
|
|
|
|
|
|
165 |
|
166 |
if self.n_speakers > 0:
|
167 |
g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
|
|
|
253 |
o = runonnx("ONNX_net/dec.onnx", z_in=(z * y_mask)[:,:,:max_len].numpy(), g=g.numpy())
|
254 |
o = torch.from_numpy(o[0])
|
255 |
|
256 |
+
return o, attn, y_mask, (z, z_p, m_p, logs_p)
|
257 |
+
|
258 |
+
def voice_conversion(self, y, y_lengths, sid_src, sid_tgt):
|
259 |
+
from ONNXVITS_utils import runonnx
|
260 |
+
assert self.n_speakers > 0, "n_speakers have to be larger than 0."
|
261 |
+
g_src = self.emb_g(sid_src).unsqueeze(-1)
|
262 |
+
g_tgt = self.emb_g(sid_tgt).unsqueeze(-1)
|
263 |
+
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src)
|
264 |
+
# z_p = self.flow(z, y_mask, g=g_src)
|
265 |
+
z_p = runonnx("ONNX_net/flow.onnx", z_p=z.numpy(), y_mask=y_mask.numpy(), g=g_src.numpy())
|
266 |
+
z_p = torch.from_numpy(z_p[0])
|
267 |
+
# z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True)
|
268 |
+
z_hat = runonnx("ONNX_net/flow.onnx", z_p=z_p.numpy(), y_mask=y_mask.numpy(), g=g_tgt.numpy())
|
269 |
+
z_hat = torch.from_numpy(z_hat[0])
|
270 |
+
# o_hat = self.dec(z_hat * y_mask, g=g_tgt)
|
271 |
+
o_hat = runonnx("ONNX_net/dec.onnx", z_in=(z_hat * y_mask).numpy(), g=g_tgt.numpy())
|
272 |
+
o_hat = torch.from_numpy(o_hat[0])
|
273 |
+
return o_hat, y_mask, (z, z_p, z_hat)
|