search

voicepaw / so-vits-svc-fork

so-vits-svc fork with realtime support, improved interface and more features.
Other
8.8k stars 1.18k forks source link

Correct configuration of the config.json file #639

Open laerciomonteiro opened 1 year ago

laerciomonteiro commented 1 year ago

Describe the bug My machine is an Asus Zenbook Duo 14 (UX482) with 16 GB RAM and dedicated Intel Iris / NVIDIA MX450 2GB graphics card. Specifications: https://www.asus.com/br/laptops/for-home/zenbook/zenbook-duo-14-ux482/. What would be the correct parameters to apply in the config.json?

Additional context My sample set is 105 .WAV files, some as long as 15 seconds. Would there be any problems? Also, how do I stop training and then resume where I left off? I noticed that some are facing this problem too.

billythemusical commented 8 months ago

Hi!

My suggestion would be to watch the video linked in the README. It helped me a lot.

There are only a few params that need adjusting based on one's system (mainly, batch_size, which should likely be reduced when not on a high performance GPU). You'll have to experiment with reducing it until you don't get errors. Try maybe halving it to 10 to start.

The vid also addresses your second question, as training will pick up where it leaves off, even if you decide to add additional epochs later.

Best of luck!