TypeError: 'dict' object is not callable

[dillfrescott]

      7 mel = mel_spectrogram_torch(wav, 2048, 256, 48000, 512, 2048, 0, None, False)
      8 mel = mel.cuda()
----> 9 out = hifigan(mel)
     10 
     11 wav_out = out.squeeze(0).cpu()

TypeError: 'dict' object is not callable

[dillfrescott]

Also, I get this error when trying to just load your model from hub:

Using cache found in /root/.cache/torch/hub/vtuber-plan_hifi-gan_main
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[<ipython-input-31-3ed915ea688c>](https://localhost:8080/#) in <module>
      1 import torch, torchaudio
      2 from hifigan.mel_processing import mel_spectrogram_torch
----> 3 hifigan = torch.hub.load("vtuber-plan/hifi-gan:main", "hifigan_48k")
      4 wav, sr = torchaudio.load("rec.wav")
      5 assert sr == 48000

3 frames
[/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py](https://localhost:8080/#) in load_state_dict(self, state_dict, strict)
   1603         if len(error_msgs) > 0:
   1604             raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
-> 1605                                self.__class__.__name__, "\n\t".join(error_msgs)))
   1606         return _IncompatibleKeys(missing_keys, unexpected_keys)
   1607 

RuntimeError: Error(s) in loading state_dict for Generator:
	size mismatch for conv_pre.weight: copying a param with shape torch.Size([512, 256, 7]) from checkpoint, the shape in current model is torch.Size([512, 128, 7]).

[FrostMiKu]

Hi @dillfrescott , the main brach is still under development, we'll let you know when we release an available version.

[dillfrescott]

Oh, okay!

[jstzwj]

I just uploaded the weights and code for version 0.2.1, the example in the readme works now. And the network is still in training, the output of this version may not be high quality.

[dillfrescott]

okay!

[dillfrescott]

Quick question. Could I use this for singing voice conversion?

[jstzwj]

We did take the singing voice conversion task into account, so a part of training dataset are singing voice, such as JSUT-Song and RAVDESS-Song. It is possible to put it into a singing voice conversion model. In future versions, we may add more singing data to fit the task better.

[dillfrescott]

Oh cool! Is it okay if I go ahead and try to run inference? Or is it not going to be ready?

[jstzwj]

This repository is the code of vocoder, the singing voice conversion model is in the vcvits repository and the model is still in development.

[dillfrescott]

Oh okay