deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 01:49:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BUG FIX] 使用 float64 避免精度误差问题，弃用 CPU 计算，避免拖累性能

Browse Source
This commit is contained in:
禾息
2026-01-30 18:10:36 +08:00
parent c93d3dda01
commit 0d990d6074
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
cosyvoice/hifigan/generator.py
View File

@@ -713,8 +713,8 @@ class CausalHiFTGenerator(HiFTGenerator):
@torch.inference_mode()
def inference(self, speech_feat: torch.Tensor, finalize: bool = True) -> torch.Tensor:
# mel->f0 NOTE f0_predictor precision is crucial for causal inference, move self.f0_predictor to cpu if necessary
self.f0_predictor.to('cpu')
f0 = self.f0_predictor(speech_feat.cpu(), finalize=finalize).to(speech_feat)
self.f0_predictor.to(torch.float64)
f0 = self.f0_predictor(speech_feat.to(torch.float64), finalize=finalize).to(speech_feat)
# f0->source
s = self.f0_upsamp(f0[:, None]).transpose(1, 2) # bs,n,t
s, _, _ = self.m_source(s)