SCP-GAN train loss NAN

[taqta]

Hello! I have encountered an issue and would appreciate your help.

when I try to train SCP-GAN with Voicebank-Demand dataset, the Loss become NAN after several batch as the fig below:

the command is below:
python main_gan -a scp --config= ./config/scp.yaml -j 8 --world-size=1 --multiprocessing-distributed

Thank you very much for your help! I look forward to your response.

[Wang-Jingrun]

Hello! I have encountered an issue and would appreciate your help.

when I try to train SCP-GAN with Voicebank-Demand dataset, the Loss become NAN after several batch as the fig below:

the command is below: python main_gan -a scp --config= ./config/scp.yaml -j 8 --world-size=1 --multiprocessing-distributed

Thank you very much for your help! I look forward to your response.

I also have a similar issue. It seems to occur after computing 'est_prime_spec', but I haven't identified the root cause of the error yet. I'm not sure if you've resolved it?

[taqta]

I didn't resolve this problem, but I reproduce CMGAN using this code https://github.com/SherifAbdulatif/CMGAN. By the way, CMGAN will perform better if you delete C-NORM when you use https://github.com/SherifAbdulatif/CMGAN.

[Wang-Jingrun]

Thank you for your response. I'll give it a try.