So-Vits-Svc Base Model V1
The base model to generate new voices with so-vits-svc voice lab.
The dataset was comprised of 278 english speaking people. 4 datasets where used:
- Genshin Voice: Only speakers with more than 30min of audio
- VCTK
- Vocalset
- Private scraped dataset
The model was trained for around 4 days and 16 hours on a single rtx 3090 (61 epochs / 430k steps)