onejiin / CycleGAN-VC2

CycleGAN-VC2: Improved CycleGAN-based Non-parallel Voice Conversion
MIT License
41 stars 7 forks source link

Discriminator Loss is 0.000 #5

Open xyxxmb opened 4 years ago

xyxxmb commented 4 years ago

When training in my datasets, I find the print log "Discriminator Loss is 0.000". Will it affect my training process and convert result?

onejiin commented 4 years ago

I think so. Discriminator can always classify fake and real waveform created by generator. If this happens, learning will not proceed. Therefore, I recommend setting the hyper parameters appropriately(e.g., learning rate or learning scheduling).