very bad quality of output inference #2841

Activity

1. Hayri-maker commented on Aug 10, 2026

   

   Hayri-maker

   on Aug 10, 2026 via email

   More actions

   Send me the datasetDen 8. aug. 2026 kl. 14.47 skrev TheAnkulai ***@***.***>:TheAnkulai created an issue (RVC-Project/Retrieval-based-Voice-Conversion-WebUI#2841) when i tried to inference I got this result with default settings not_normal.wav But with old rvc and same settings i got better result good_result.wav I dont know what's wrong, but that's sad —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>
2. TheAnkulai commented on Aug 12, 2026

   

   TheAnkulai

   on Aug 12, 2026

   Last edited by TheAnkulai

   Author

   More actions

   I don't know how, but it fixed by redownload and replace modules.py.
   In first attempts on all methods I got this error (first lines was different):

   [INFO]: device is not None, use cpu
     [INFO]    > call by:torchfcpe.tools.spawn_infer_cf_naive_mel_pe_from_pt
     [WARN] args.model.use_harmonic_emb is None; use default False
     [WARN]    > call by:torchfcpe.tools.spawn_cf_naive_mel_pe
   Traceback (most recent call last):
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\vc\modules.py", line 263, in vc_single
       audio_opt = self.pipeline.pipeline(
                   ^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\vc\pipeline.py", line 328, in pipeline
       self.vc(
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\vc\pipeline.py", line 222, in vc
       synthesized = run_cuda_graph(
                     ^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\tools\cuda_graph.py", line 190, in run_cuda_graph
       return function(*inputs)
              ^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\vc\pipeline.py", line 225, in <lambda>
       lambda phone, lengths, coarse, continuous, speaker: net_g.infer(
                                                           ^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\module\models.py", line 677, in infer
       o = self.dec(z * x_mask, nsff0, g=g, n_res=return_length2)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl
       return self._call_impl(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl
       return forward_call(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\module\models.py", line 484, in forward
       har_source, noi_source, uv = self.m_source(f0, self.upp)
                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl
       return self._call_impl(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl
       return forward_call(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\module\models.py", line 391, in forward
       sine_wavs, uv, _ = self.l_sin_gen(x, upp)
                          ^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl
       return self._call_impl(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl
       return forward_call(*args, **kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\module\models.py", line 335, in forward
       sine_waves = self._f02sine(f0, upp) * self.sine_amp
                    ^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\infer\module\models.py", line 315, in _f02sine
       rad += F.pad(rad_acc, (0, 0, 1, -1))
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "F:\Retrieval-based-Voice-Conversion-WebUI\.venv\Lib\site-packages\torch\nn\functional.py", line 4552, in pad
       return torch._C._nn.pad(input, pad, mode, value)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   RuntimeError: torch-dml currently doesn't support a mix of positive and negative padding values, must be all positive or all negative.
   

   But now after replace modules.py it starts work properly. So, at this time this issue can be considered irrelevant.
   If you want to get .pth and .index that i used for this, there it is: https://dropmefiles.com/zm3vS (i can't upload zip in this message, idk why. And you able to download it only for 1 week). This model was trained not by me, so I don't know exactly dataset that was used for train. This is all I can give at this momentzz