Training epochs

[dinashi]

How many epochs are needed for training melgan? With the default value of 3000 for the number of epochs, the result generated wav only contains noise. Also could you provide your pretrained models?

[dwgnr]

Hi, 3000 epochs shouldn't be needed when you train on all speakers. You should get good results at around 1000 epochs already. I uploaded the pretrained MelGAN and the pretrained VQ-VAE + GAN decoder for audio synthesis here.

[RichardGajda]

Hello, can you please share how many epochs were needed for the MelGAN to be trained into the shape that is used in the demo ?

Thanks!

[Ujali049]

Hello, I was training MelGAN for all speakers. I have trained this for 2000 epochs, still results generated are still noise. Can you please tell is there anything we need to make sure while training MelGAN.

Thanks.