Update cosyvoice/hifigan/generator.py

cosyvoice/hifigan/generator.py

@@ -672,6 +672,7 @@ class CausalHiFTGenerator(HiFTGenerator):
     def decode(self, x: torch.Tensor, s: torch.Tensor = torch.zeros(1, 1, 0), finalize: bool = True) -> torch.Tensor:
         s_stft_real, s_stft_imag = self._stft(s.squeeze(1))
         if finalize is True:
+            print('weight device {}'.format(self.conv_pre.weight.device))
             x = self.conv_pre(x.cuda())
         else:
             x = self.conv_pre(x[:, :, :-self.conv_pre_look_right], x[:, :, -self.conv_pre_look_right:])
@@ -713,8 +714,7 @@ 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)
+        f0 = self.f0_predictor(speech_feat, finalize=finalize).to(speech_feat)
         # f0->source
         s = self.f0_upsamp(f0[:, None]).transpose(1, 2)  # bs,n,t
         s, _, _ = self.m_source(s)