About the models
These two models are originally Japanese text-to-speech (TTS) voices, which I was able to find in an online TTS website.
List of voices
- Haruka: Typical anime girl voice. Good for cute/kawaii characters.
- Hikari: For everything else. Soft voice tone, ideal for news and/or other characters.
Training details
The two voices were trained using a 20-minute dataset, with 250 epochs and RMVPE as the pitch extraction method. However, the original streaming audios were 22 kHz, 48 kb/s MP3 files, so the AI "learned" some of the artifacts. I don't have access to higher-quality versions of these voices (there aren't ways to get them) and if I had, these RVC models wouldn't exist in the first place.
Final words
Nothing. Enjoy the models, and let me know if you make something nice with them!
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API:
The model has no library tag.