mirror of
https://github.com/fumiama/Retrieval-based-Voice-Conversion-WebUI.git
synced 2026-06-09 04:29:50 +08:00
optimize(uvr5): remove redundant files
This commit is contained in:
源文雨
@@ -1,85 +1,100 @@
|
||||
import layers
|
||||
import torch
|
||||
import torch.nn.functional as F
|
||||
from torch import nn
|
||||
|
||||
from . import spec_utils
|
||||
from . import layers
|
||||
|
||||
|
||||
class BaseASPPNet(nn.Module):
|
||||
def __init__(self, nin, ch, dilations=(4, 8, 16)):
|
||||
super(BaseASPPNet, self).__init__()
|
||||
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
|
||||
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
|
||||
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
|
||||
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
|
||||
class BaseNet(nn.Module):
|
||||
def __init__(
|
||||
self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
|
||||
):
|
||||
super(BaseNet, self).__init__()
|
||||
self.enc1 = layers.Conv2DBNActiv(nin, nout, 3, 1, 1)
|
||||
self.enc2 = layers.Encoder(nout, nout * 2, 3, 2, 1)
|
||||
self.enc3 = layers.Encoder(nout * 2, nout * 4, 3, 2, 1)
|
||||
self.enc4 = layers.Encoder(nout * 4, nout * 6, 3, 2, 1)
|
||||
self.enc5 = layers.Encoder(nout * 6, nout * 8, 3, 2, 1)
|
||||
|
||||
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
|
||||
self.aspp = layers.ASPPModule(nout * 8, nout * 8, dilations, dropout=True)
|
||||
|
||||
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
|
||||
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
|
||||
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
|
||||
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
|
||||
self.dec4 = layers.Decoder(nout * (6 + 8), nout * 6, 3, 1, 1)
|
||||
self.dec3 = layers.Decoder(nout * (4 + 6), nout * 4, 3, 1, 1)
|
||||
self.dec2 = layers.Decoder(nout * (2 + 4), nout * 2, 3, 1, 1)
|
||||
self.lstm_dec2 = layers.LSTMModule(nout * 2, nin_lstm, nout_lstm)
|
||||
self.dec1 = layers.Decoder(nout * (1 + 2) + 1, nout * 1, 3, 1, 1)
|
||||
|
||||
def __call__(self, x):
|
||||
h, e1 = self.enc1(x)
|
||||
h, e2 = self.enc2(h)
|
||||
h, e3 = self.enc3(h)
|
||||
h, e4 = self.enc4(h)
|
||||
e1 = self.enc1(x)
|
||||
e2 = self.enc2(e1)
|
||||
e3 = self.enc3(e2)
|
||||
e4 = self.enc4(e3)
|
||||
e5 = self.enc5(e4)
|
||||
|
||||
h = self.aspp(h)
|
||||
h = self.aspp(e5)
|
||||
|
||||
h = self.dec4(h, e4)
|
||||
h = self.dec3(h, e3)
|
||||
h = self.dec2(h, e2)
|
||||
h = torch.cat([h, self.lstm_dec2(h)], dim=1)
|
||||
h = self.dec1(h, e1)
|
||||
|
||||
return h
|
||||
|
||||
|
||||
class CascadedASPPNet(nn.Module):
|
||||
def __init__(self, n_fft):
|
||||
super(CascadedASPPNet, self).__init__()
|
||||
self.stg1_low_band_net = BaseASPPNet(2, 16)
|
||||
self.stg1_high_band_net = BaseASPPNet(2, 16)
|
||||
|
||||
self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
|
||||
self.stg2_full_band_net = BaseASPPNet(8, 16)
|
||||
|
||||
self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
|
||||
self.stg3_full_band_net = BaseASPPNet(16, 32)
|
||||
|
||||
self.out = nn.Conv2d(32, 2, 1, bias=False)
|
||||
self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
|
||||
self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
|
||||
class CascadedNet(nn.Module):
|
||||
def __init__(self, n_fft, nout=32, nout_lstm=128):
|
||||
super(CascadedNet, self).__init__()
|
||||
|
||||
self.max_bin = n_fft // 2
|
||||
self.output_bin = n_fft // 2 + 1
|
||||
self.nin_lstm = self.max_bin // 2
|
||||
self.offset = 64
|
||||
|
||||
self.offset = 128
|
||||
self.stg1_low_band_net = nn.Sequential(
|
||||
BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
|
||||
layers.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0),
|
||||
)
|
||||
|
||||
def forward(self, x, aggressiveness=None):
|
||||
mix = x.detach()
|
||||
x = x.clone()
|
||||
self.stg1_high_band_net = BaseNet(
|
||||
2, nout // 4, self.nin_lstm // 2, nout_lstm // 2
|
||||
)
|
||||
|
||||
self.stg2_low_band_net = nn.Sequential(
|
||||
BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
|
||||
layers.Conv2DBNActiv(nout, nout // 2, 1, 1, 0),
|
||||
)
|
||||
self.stg2_high_band_net = BaseNet(
|
||||
nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2
|
||||
)
|
||||
|
||||
self.stg3_full_band_net = BaseNet(
|
||||
3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm
|
||||
)
|
||||
|
||||
self.out = nn.Conv2d(nout, 2, 1, bias=False)
|
||||
self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
|
||||
|
||||
def forward(self, x):
|
||||
x = x[:, :, : self.max_bin]
|
||||
|
||||
bandw = x.size()[2] // 2
|
||||
aux1 = torch.cat(
|
||||
[
|
||||
self.stg1_low_band_net(x[:, :, :bandw]),
|
||||
self.stg1_high_band_net(x[:, :, bandw:]),
|
||||
],
|
||||
dim=2,
|
||||
)
|
||||
l1_in = x[:, :, :bandw]
|
||||
h1_in = x[:, :, bandw:]
|
||||
l1 = self.stg1_low_band_net(l1_in)
|
||||
h1 = self.stg1_high_band_net(h1_in)
|
||||
aux1 = torch.cat([l1, h1], dim=2)
|
||||
|
||||
h = torch.cat([x, aux1], dim=1)
|
||||
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
|
||||
l2_in = torch.cat([l1_in, l1], dim=1)
|
||||
h2_in = torch.cat([h1_in, h1], dim=1)
|
||||
l2 = self.stg2_low_band_net(l2_in)
|
||||
h2 = self.stg2_high_band_net(h2_in)
|
||||
aux2 = torch.cat([l2, h2], dim=2)
|
||||
|
||||
h = torch.cat([x, aux1, aux2], dim=1)
|
||||
h = self.stg3_full_band_net(self.stg3_bridge(h))
|
||||
f3_in = torch.cat([x, aux1, aux2], dim=1)
|
||||
f3 = self.stg3_full_band_net(f3_in)
|
||||
|
||||
mask = torch.sigmoid(self.out(h))
|
||||
mask = torch.sigmoid(self.out(f3))
|
||||
mask = F.pad(
|
||||
input=mask,
|
||||
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
|
||||
@@ -87,37 +102,32 @@ class CascadedASPPNet(nn.Module):
|
||||
)
|
||||
|
||||
if self.training:
|
||||
aux1 = torch.sigmoid(self.aux1_out(aux1))
|
||||
aux1 = F.pad(
|
||||
input=aux1,
|
||||
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
|
||||
aux = torch.cat([aux1, aux2], dim=1)
|
||||
aux = torch.sigmoid(self.aux_out(aux))
|
||||
aux = F.pad(
|
||||
input=aux,
|
||||
pad=(0, 0, 0, self.output_bin - aux.size()[2]),
|
||||
mode="replicate",
|
||||
)
|
||||
aux2 = torch.sigmoid(self.aux2_out(aux2))
|
||||
aux2 = F.pad(
|
||||
input=aux2,
|
||||
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
|
||||
mode="replicate",
|
||||
)
|
||||
return mask * mix, aux1 * mix, aux2 * mix
|
||||
return mask, aux
|
||||
else:
|
||||
if aggressiveness:
|
||||
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
|
||||
mask[:, :, : aggressiveness["split_bin"]],
|
||||
1 + aggressiveness["value"] / 3,
|
||||
)
|
||||
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
|
||||
mask[:, :, aggressiveness["split_bin"] :],
|
||||
1 + aggressiveness["value"],
|
||||
)
|
||||
return mask
|
||||
|
||||
return mask * mix
|
||||
|
||||
def predict(self, x_mag, aggressiveness=None):
|
||||
h = self.forward(x_mag, aggressiveness)
|
||||
def predict_mask(self, x):
|
||||
mask = self.forward(x)
|
||||
|
||||
if self.offset > 0:
|
||||
h = h[:, :, :, self.offset : -self.offset]
|
||||
assert h.size()[3] > 0
|
||||
mask = mask[:, :, :, self.offset : -self.offset]
|
||||
assert mask.size()[3] > 0
|
||||
|
||||
return h
|
||||
return mask
|
||||
|
||||
def predict(self, x, aggressiveness=None):
|
||||
mask = self.forward(x)
|
||||
pred_mag = x * mask
|
||||
|
||||
if self.offset > 0:
|
||||
pred_mag = pred_mag[:, :, :, self.offset : -self.offset]
|
||||
assert pred_mag.size()[3] > 0
|
||||
|
||||
return pred_mag
|
||||
|
||||
Reference in New Issue
Block a user