deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 18:09:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix lint

Browse Source
This commit is contained in:
lyuxiang.lx
2024-12-16 09:54:24 +08:00
parent 6b5931dc70
commit ac70560364
8 changed files with 33 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
cosyvoice/flow/decoder.py
View File

@@ -49,7 +49,7 @@ class CausalBlock1D(Block1D):
class CausalResnetBlock1D(ResnetBlock1D):
def __init__(self, dim: int, dim_out: int, time_emb_dim: int, groups: int=8):
def __init__(self, dim: int, dim_out: int, time_emb_dim: int, groups: int = 8):
super(CausalResnetBlock1D, self).__init__(dim, dim_out, time_emb_dim, groups)
self.block1 = CausalBlock1D(dim, dim_out)
self.block2 = CausalBlock1D(dim_out, dim_out)
@@ -70,12 +70,11 @@ class CausalConv1d(torch.nn.Conv1d):
dtype=None
) -> None:
super(CausalConv1d, self).__init__(in_channels, out_channels,
kernel_size, stride,
padding=0, dilation=dilation,
groups=groups, bias=bias,
padding_mode=padding_mode,
device=device, dtype=dtype
)
kernel_size, stride,
padding=0, dilation=dilation,
groups=groups, bias=bias,
padding_mode=padding_mode,
device=device, dtype=dtype)
assert stride == 1
self.causal_padding = (kernel_size - 1, 0)
@@ -124,7 +123,8 @@ class ConditionalDecoder(nn.Module):
input_channel = output_channel
output_channel = channels[i]
is_last = i == len(channels) - 1
resnet = CausalResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim) if self.causal else ResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim)
resnet = CausalResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim) if self.causal \
else ResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim)
transformer_blocks = nn.ModuleList(
[
BasicTransformerBlock(
@@ -138,14 +138,16 @@ class ConditionalDecoder(nn.Module):
]
)
downsample = (
Downsample1D(output_channel) if not is_last else CausalConv1d(output_channel, output_channel, 3) if self.causal else nn.Conv1d(output_channel, output_channel, 3, padding=1)
Downsample1D(output_channel) if not is_last else \
CausalConv1d(output_channel, output_channel, 3) if self.causal else nn.Conv1d(output_channel, output_channel, 3, padding=1)
)
self.down_blocks.append(nn.ModuleList([resnet, transformer_blocks, downsample]))
for _ in range(num_mid_blocks):
input_channel = channels[-1]
out_channels = channels[-1]
resnet = CausalResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim) if self.causal else ResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim)
resnet = CausalResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim) if self.causal else \
ResnetBlock1D(dim=input_channel, dim_out=output_channel, time_emb_dim=time_embed_dim)
transformer_blocks = nn.ModuleList(
[

5
cosyvoice/flow/flow.py
View File

@@ -202,7 +202,6 @@ class CausalMaskedDiffWithXvec(torch.nn.Module):
embedding = self.spk_embed_affine_layer(embedding)
# concat text and prompt_text
token_len1, token_len2 = prompt_token.shape[1], token.shape[1]
token, token_len = torch.concat([prompt_token, token], dim=1), prompt_token_len + token_len
mask = (~make_pad_mask(token_len)).unsqueeze(-1).to(embedding)
token = self.input_embedding(torch.clamp(token, min=0)) * mask
@@ -211,7 +210,7 @@ class CausalMaskedDiffWithXvec(torch.nn.Module):
h, h_lengths = self.encoder(token, token_len)
if finalize is False:
h = h[:, :-self.pre_lookahead_len * self.token_mel_ratio]
mel_len1, mel_len2 = prompt_feat.shape[1], h.shape[1] - prompt_feat.shape[1]
mel_len1, mel_len2 = prompt_feat.shape[1], h.shape[1] - prompt_feat.shape[1]
h = self.encoder_proj(h)
# get conditions
@@ -229,4 +228,4 @@ class CausalMaskedDiffWithXvec(torch.nn.Module):
)
feat = feat[:, :, mel_len1:]
assert feat.shape[2] == mel_len2
return feat, None
return feat, None