Youtube: smotto_ai Tiktok: smotto_ai

RVC v2 model

Don't forget to credit me @Smotto if you do use this model!

Data

• English Talking style taken from her live podcast. Cut and clean them manually, but didn't bother keeping clips with any background noise (too lazy to edit them out between words).
• 61.6 MB of data == 5 min and 36 seconds
• 48khz 16bit-depth audio files

Processing

• Split audio clips using whisperX
• Kim vocal 1 -> Reverb HQ -> Karaoke 2 (if * needed) -> DeEcho -> Denoise

Hyper-parameters

• mangio-crepe
• 6 batch size
• 16 pitch extraction hop-length
• 300 epochs