fumiama/Retrieval-based-Voice-Conversion-WebUI
1
0
Fork 0
mirror of https://github.com/fumiama/Retrieval-based-Voice-Conversion-WebUI.git synced 2026-06-07 19:20:28 +08:00
Code Activity

chore(format): run black on dev (#5)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
This commit is contained in:
github-actions[bot]
2024-06-07 00:44:05 +09:00
committed by GitHub
parent 5eed789fe7
commit ea66e6d28c
4 changed files with 60 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
infer/lib/infer_pack/attentions.py
View File

@@ -73,6 +73,7 @@ class Encoder(nn.Module):
x = x * x_mask x = x * x_mask
return x return x
""" """
class Decoder(nn.Module): class Decoder(nn.Module):
def __init__( def __init__(
@@ -158,6 +159,7 @@ class Decoder(nn.Module):
return x return x
""" """
class MultiHeadAttention(nn.Module): class MultiHeadAttention(nn.Module):
def __init__( def __init__(
self, self,

48
infer/lib/infer_pack/models.py
View File

@@ -60,7 +60,9 @@ class TextEncoder(nn.Module):
# skip_head: Optional[torch.Tensor] = None, # skip_head: Optional[torch.Tensor] = None,
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
return super().__call__( return super().__call__(
phone, pitch, lengths, phone,
pitch,
lengths,
# skip_head=skip_head, # skip_head=skip_head,
) )
@@ -80,8 +82,10 @@ class TextEncoder(nn.Module):
x = torch.transpose(x, 1, -1) # [b, h, t] x = torch.transpose(x, 1, -1) # [b, h, t]
x_mask = torch.unsqueeze( x_mask = torch.unsqueeze(
utils.sequence_mask( utils.sequence_mask(
lengths, x.size(2), lengths,
), 1, x.size(2),
),
1,
).to(x.dtype) ).to(x.dtype)
x = self.encoder(x * x_mask, x_mask) x = self.encoder(x * x_mask, x_mask)
""" """
@@ -193,17 +197,19 @@ class PosteriorEncoder(nn.Module):
self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1) self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
def __call__( def __call__(
self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
super().__call__(x, x_lengths, g = g) super().__call__(x, x_lengths, g=g)
def forward( def forward(
self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
x_mask = torch.unsqueeze( x_mask = torch.unsqueeze(
utils.sequence_mask( utils.sequence_mask(
x_lengths, x.size(2), x_lengths,
), 1, x.size(2),
),
1,
).to(x.dtype) ).to(x.dtype)
x = self.pre(x) * x_mask x = self.pre(x) * x_mask
x = self.enc(x, x_mask, g=g) x = self.enc(x, x_mask, g=g)
@@ -663,11 +669,12 @@ class SynthesizerTrnMs256NSFsid(nn.Module):
**kwargs **kwargs
): ):
super(SynthesizerTrnMs256NSFsid, self).__init__() super(SynthesizerTrnMs256NSFsid, self).__init__()
if isinstance(sr, str): sr = { if isinstance(sr, str):
"32k": 32000, sr = {
"40k": 40000, "32k": 32000,
"48k": 48000, "40k": 40000,
}[sr] "48k": 48000,
}[sr]
self.spec_channels = spec_channels self.spec_channels = spec_channels
self.inter_channels = inter_channels self.inter_channels = inter_channels
self.hidden_channels = hidden_channels self.hidden_channels = hidden_channels
@@ -776,9 +783,7 @@ class SynthesizerTrnMs256NSFsid(nn.Module):
m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths) m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g) z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
z_p = self.flow(z, y_mask, g=g) z_p = self.flow(z, y_mask, g=g)
z_slice, ids_slice = utils.rand_slice_segments( z_slice, ids_slice = utils.rand_slice_segments(z, y_lengths, self.segment_size)
z, y_lengths, self.segment_size
)
# print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length) # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
pitchf = utils.slice_on_last_dim(pitchf, ids_slice, self.segment_size) pitchf = utils.slice_on_last_dim(pitchf, ids_slice, self.segment_size)
# print(-2,pitchf.shape,z_slice.shape) # print(-2,pitchf.shape,z_slice.shape)
@@ -816,7 +821,9 @@ class SynthesizerTrnMs256NSFsid(nn.Module):
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
z = self.flow(z_p, x_mask, g=g, reverse=True) z = self.flow(z_p, x_mask, g=g, reverse=True)
o = self.dec( o = self.dec(
z * x_mask, nsff0, g=g, z * x_mask,
nsff0,
g=g,
# n_res=return_length2, # n_res=return_length2,
) )
return o, x_mask, (z, z_p, m_p, logs_p) return o, x_mask, (z, z_p, m_p, logs_p)
@@ -1000,9 +1007,7 @@ class SynthesizerTrnMs256NSFsid_nono(nn.Module):
m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths) m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g) z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
z_p = self.flow(z, y_mask, g=g) z_p = self.flow(z, y_mask, g=g)
z_slice, ids_slice = utils.rand_slice_segments( z_slice, ids_slice = utils.rand_slice_segments(z, y_lengths, self.segment_size)
z, y_lengths, self.segment_size
)
o = self.dec(z_slice, g=g) o = self.dec(z_slice, g=g)
return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q) return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
@@ -1014,7 +1019,7 @@ class SynthesizerTrnMs256NSFsid_nono(nn.Module):
sid: torch.Tensor, sid: torch.Tensor,
skip_head: Optional[torch.Tensor] = None, skip_head: Optional[torch.Tensor] = None,
return_length: Optional[torch.Tensor] = None, return_length: Optional[torch.Tensor] = None,
#return_length2: Optional[torch.Tensor] = None, # return_length2: Optional[torch.Tensor] = None,
): ):
g = self.emb_g(sid).unsqueeze(-1) g = self.emb_g(sid).unsqueeze(-1)
if skip_head is not None and return_length is not None: if skip_head is not None and return_length is not None:
@@ -1034,7 +1039,8 @@ class SynthesizerTrnMs256NSFsid_nono(nn.Module):
z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
z = self.flow(z_p, x_mask, g=g, reverse=True) z = self.flow(z_p, x_mask, g=g, reverse=True)
o = self.dec( o = self.dec(
z * x_mask, g=g, z * x_mask,
g=g,
# n_res=return_length2 # n_res=return_length2
) )
return o, x_mask, (z, z_p, m_p, logs_p) return o, x_mask, (z, z_p, m_p, logs_p)

21
infer/lib/infer_pack/models_onnx.py
View File

@@ -1,7 +1,15 @@
import torch import torch
from torch import nn from torch import nn
from .attentions import TextEncoder, ResidualCouplingBlock, PosteriorEncoder, Generator, SineGen, SourceModuleHnNSF, GeneratorNSF from .attentions import (
TextEncoder,
ResidualCouplingBlock,
PosteriorEncoder,
Generator,
SineGen,
SourceModuleHnNSF,
GeneratorNSF,
)
class SynthesizerTrnMsNSFsidM(nn.Module): class SynthesizerTrnMsNSFsidM(nn.Module):
@@ -29,11 +37,12 @@ class SynthesizerTrnMsNSFsidM(nn.Module):
**kwargs **kwargs
): ):
super(SynthesizerTrnMsNSFsidM, self).__init__() super(SynthesizerTrnMsNSFsidM, self).__init__()
if isinstance(sr, str): sr = { if isinstance(sr, str):
"32k": 32000, sr = {
"40k": 40000, "32k": 32000,
"48k": 48000, "40k": 40000,
}[sr] "48k": 48000,
}[sr]
self.spec_channels = spec_channels self.spec_channels = spec_channels
self.inter_channels = inter_channels self.inter_channels = inter_channels
self.hidden_channels = hidden_channels self.hidden_channels = hidden_channels

25
rvc/utils.py
View File

@@ -2,11 +2,12 @@ from typing import List, Optional, Tuple
import torch import torch
def call_weight_data_normal_if_Conv(m: torch.nn.Module): def call_weight_data_normal_if_Conv(m: torch.nn.Module):
classname = m.__class__.__name__ classname = m.__class__.__name__
if classname.find("Conv") != -1: if classname.find("Conv") != -1:
mean=0.0 mean = 0.0
std=0.01 std = 0.01
m.weight.data.normal_(mean, std) m.weight.data.normal_(mean, std)
@@ -15,8 +16,10 @@ def get_padding(kernel_size: int, dilation=1):
def slice_on_last_dim( def slice_on_last_dim(
x: torch.Tensor, start_indices: List[int], segment_size=4, x: torch.Tensor,
) -> torch.Tensor: start_indices: List[int],
segment_size=4,
) -> torch.Tensor:
new_shape = x.shape new_shape = x.shape
new_shape[-1] = segment_size new_shape[-1] = segment_size
ret = torch.empty(new_shape) ret = torch.empty(new_shape)
@@ -28,10 +31,13 @@ def slice_on_last_dim(
def rand_slice_segments( def rand_slice_segments(
x: torch.Tensor, x_lengths: int = None, segment_size=4, x: torch.Tensor,
) -> Tuple[torch.Tensor, List[int]]: x_lengths: int = None,
segment_size=4,
) -> Tuple[torch.Tensor, List[int]]:
b, _, t = x.size() b, _, t = x.size()
if x_lengths is None: x_lengths = t if x_lengths is None:
x_lengths = t
ids_str_max = x_lengths - segment_size + 1 ids_str_max = x_lengths - segment_size + 1
ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long) ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
ret = slice_on_last_dim(x, ids_str, segment_size) ret = slice_on_last_dim(x, ids_str, segment_size)
@@ -53,8 +59,9 @@ def convert_pad_shape(pad_shape: List[List[int]]) -> List[int]:
def sequence_mask( def sequence_mask(
length: torch.Tensor, max_length: Optional[int] = None, length: torch.Tensor,
) -> torch.BoolTensor: max_length: Optional[int] = None,
) -> torch.BoolTensor:
if max_length is None: if max_length is None:
max_length = int(length.max()) max_length = int(length.max())
x = torch.arange(max_length, dtype=length.dtype, device=length.device) x = torch.arange(max_length, dtype=length.dtype, device=length.device)