if I use my own dataset, does that mean I have to train ParallelWaveGAN first?

Tinglok / CVC

CVC: Contrastive Learning for Non-parallel Voice Conversion (INTERSPEECH 2021, in PyTorch)

MIT License

57 stars 12 forks source link

if I use my own dataset, does that mean I have to train ParallelWaveGAN first? #6

Closed chikiuso closed 3 years ago

chikiuso commented 4 years ago

Hi, I see there is a pretrained checkpoints of ParallelWaveGAN which I have to download first before training, does that mean I have to train ParallelWaveGAN first if I use my old dataset? thanks.

Tinglok commented 4 years ago

Hi @chikiuso, Yes, you were right. The neural vocoder does not have the speaker-independent ability so far. If you want to convert your own dataset, please see this URL for training ParallelWaveGAN from scratch.

chikiuso commented 4 years ago

Hi @Tinglok , thanks for your reply. ParallelWaveGAN seems not a non-parallel model, does that mean I have to train it with a parallel voice dataset?

Tinglok commented 4 years ago

Hi @chikiuso, I am not sure about the training details of the ParallelWaveGAN vocoder, but I think you are right.