mostafaelaraby / wavegan-pytorch Goto Github PK

How do I generate audio with a trained model?

Fails with last Step

0% 1/250 [00:27<1:52:18, 27.06s/it, Loss_D WD=-0.02465313859283924, Loss_G=-0.41385307908058167, Val_G=-0.030089471489191055]Traceback (most recent call last):
  File "train.py", line 223, in <module>
    wave_gan.train()
  File "train.py", line 199, in train
    save_samples(fake, iter_indx)
  File "/content/wavegan-pytorch/utils.py", line 172, in save_samples
    librosa.output.write_wav(output_path, sample, sampling_rate)
AttributeError: module 'librosa' has no attribute 'output'
  0% 1/250 [00:28<1:57:56, 28.42s/it, Loss_D WD=-0.02465313859283924, Loss_G=-0.41385307908058167, Val_G=-0.030089471489191055]

Thank you for making the repository public. I am trying to implement the model in Python 3.9. It showed a package error when using Python 3.11. Now that I am trying to train the model, it shows the below-mentioned error. Please let me know where I am going wrong. I have installed all the required packages.

Can you upload some pre-trained models

Thanks very much for making this repo!

Quick comment is that, as I understand it, the WavGAN paper actually does not recommend nearest neighbor upsampling, and instead better results were obtained by the standard ConvTranspose approach. However, in model.py, the class "Transpose1dLayer" has a comment saying that upsampling IS recommended and the class "WaveGANGenerator" has upsampling enabled by default.

I don't understand how I can generate Audio from my trained model, other than the samples automatically saved during the training process.

Hi,

I try to run MacBook Pro 2015 computer which has no GPU unit. Successfully downloaded piano dataset but while I try to start train, get these errors:

I was trying to get the original Wavegan to run and tearing my hair out. Thank you for making it into a usable colab notebook, you saved me days of frustrated effort.

Dear mostafaelaraby,

Hi, thank you for your pytorch implementation.

I've done your codes on sc09 speech dataset with some hyperparameters, but I couldn't get any plausible sound results (no sound).

Are these codes available on sc09 dataset?
If you have succeeded on sc09, please tell me the hyperparameters.

Regards,

Hello.

I'm currently training wavegan-pytorch with a set of files (<300) and I'm getting silence files every iteration - it generates 10 .wav but it's silence, 0 values all duration. During the installation, I had some problems with librosa in my machine and changed line 172 on utils.py to:

    #librosa.output.write_wav(output_path, sample, sampling_rate)
    sf.write(output_path, sample, sampling_rate) #'PCM_24')

sf as of soundfile library. Might this be the problem? Is there a script here to navigate the latent space and debug this?

Thanks in advance.
Luis

How would one go about converting the model to torch script?

Hi,

Is there any way to increase output file duration. It is now just 4 seconds and that is really short for me.