verachen/VoxCPM
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

FIX:When a prompt is present, concatenate two patches as the context for VAE decoding

Browse Source
This commit is contained in:
刘鑫
2025-12-15 20:37:02 +08:00
parent b3a2d95fec
commit ee5f2567ac
1 changed files with 0 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/voxcpm/model/voxcpm.py
View File

@ -452,7 +452,6 @@ class VoxCPMModel(nn.Module):
patch_len = self.patch_size * self.chunk_size patch_len = self.patch_size * self.chunk_size
for latent_pred, _ in inference_result: for latent_pred, _ in inference_result:
decode_audio = self.audio_vae.decode(latent_pred.to(torch.float32)) decode_audio = self.audio_vae.decode(latent_pred.to(torch.float32))
print(decode_audio.shape)
decode_audio = decode_audio[..., -patch_len:].squeeze(1).cpu() decode_audio = decode_audio[..., -patch_len:].squeeze(1).cpu()
yield decode_audio yield decode_audio
break break