bacow/GPT-SoVITS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update_infer

Browse Source
This commit is contained in:
Watchtower-Liu
2024-02-16 16:53:57 +08:00
parent 41041715a4
commit 1803729360
6 changed files with 88 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
GPT_SoVITS/module/models.py
View File

@@ -228,6 +228,7 @@ class TextEncoder(nn.Module):
)
y = self.ssl_proj(y * y_mask) * y_mask
y = self.encoder_ssl(y * y_mask, y_mask)
text_mask = torch.unsqueeze(
@@ -958,11 +959,13 @@ class SynthesizerTrn(nn.Module):
@torch.no_grad()
def decode(self, codes, text, refer, noise_scale=0.5):
refer_lengths = torch.LongTensor([refer.size(2)]).to(refer.device)
refer_mask = torch.unsqueeze(
commons.sequence_mask(refer_lengths, refer.size(2)), 1
).to(refer.dtype)
ge = self.ref_enc(refer * refer_mask, refer_mask)
ge = None
if refer is not None:
refer_lengths = torch.LongTensor([refer.size(2)]).to(refer.device)
refer_mask = torch.unsqueeze(
commons.sequence_mask(refer_lengths, refer.size(2)), 1
).to(refer.dtype)
ge = self.ref_enc(refer * refer_mask, refer_mask)
y_lengths = torch.LongTensor([codes.size(2) * 2]).to(codes.device)
text_lengths = torch.LongTensor([text.size(-1)]).to(text.device)