vlomme/again-melgan-voice-cloning

Русско-Английский вокодер на GAN

again-melgan-voice-cloning's Issues

Windows 10

Попробовал собрать под Windows 10. Train ни в какую не заводится. Похоже проблема с PyTorch и ее DataLoader
Перечисление тут не работает:
https://github.com/vlomme/MelGan-WavGan/blob/24f02c11c61aea7a5daf9b97776f733066033b69/wavgan.py#L120
Код отрабатывает без ошибок.
Есть советы как побороть?

Wavgan as a vocoder for espeak

Dear community.
I want to inform you that I was able to train wavgan vocoder with the output of the espeak (http://espeak.sourceforge.net/). And I got the following results:
Espeak output en
Wavgan output LJSpeech
Espeak ouput ru
Wavgan output R012

I also tried to train the melgan with the same approach. With the following sad result:
Melgan output LJSpeech

This approach shows that wavgan is relatively flexible concerning input audio. It should encourage the community to try out other approaches.

Thank you Vlomme!

Error while model training start

When I try to make training I get the error (I added torch.autograd.set_detect_anomaly(True) to get more details):

Загрузка чекпоинтов
Чекпоинт загружен: Эпоха 298, Шаг 518000
Warning: Error detected in WeightNormCudaInterfaceBackward. Traceback of forward call that caused the error:
  File "melgan.py", line 204, in <module>
    train()
  File "melgan.py", line 114, in train
    d_fake = D(fake_audio.cuda())
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/content/MelGan-WavGan/utils.py", line 278, in forward
    result.append(module(x))
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "/content/MelGan-WavGan/utils.py", line 259, in forward
    x = module(x)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 542, in __call__
    result = hook(self, input)
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/utils/weight_norm.py", line 55, in __call__
    setattr(module, self.name, self.compute_weight(module))
  File "/usr/local/lib/python3.6/dist-packages/torch/nn/utils/weight_norm.py", line 18, in compute_weight
    return _weight_norm(v, g, self.dim)
 (print_stack at /pytorch/torch/csrc/autograd/python_anomaly_mode.cpp:60)
Traceback (most recent call last):
  File "melgan.py", line 204, in <module>
    train()
  File "melgan.py", line 154, in train
    g_loss.backward()
  File "/usr/local/lib/python3.6/dist-packages/torch/tensor.py", line 198, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File "/usr/local/lib/python3.6/dist-packages/torch/autograd/__init__.py", line 100, in backward
    allow_unreachable=True)  # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [1, 1024, 3]] is at version 2; expected version 1 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!

Maybe I need to install correct pytorch version or there is some other issue with inplace variables. Was not able to address the issue.

In my experiment I just call
python melgan.py -r t -f mel

Add new voice

Hi, in case I want to train MelGan and WavGan on a new voice, what amount of input data do i need? 1H, 10Hrs or more for a new speaker?