Intro
In this model has all the character, can achieve higher pitch. Gotou Futari is the cutest voice and able produce higher output quality even just 10 files at ~10 seconds each, Male Voice to Gotou Futari need to Crepe +6 pitch (0.5 octave) transpose (adjust depending how deep source is). Sample AI song are down below.
Dataset
Source dataset is from streaming services, model quality hard to achieve due to lossy codec (MP3/AAC/Opus), perhaps waiting BD Remux/FLAC release and rip from there, so processing, clean-up will be much better than ripping from the web
Infer
It has usual so-vits-svc-fork and rvc, I not sure which one is good, but rvc train much quicker but still artifact at breathing/whisper/low loudness
Post Processing
EffectPresets.xml included inside v2.1 zip file, read the readme.txt file please
Sample Song
Hyper Act. - Harapan ft. Nijika Ijichi
Language: ms-MY (Bahasa Melayu) [so-vits-svc-fork]
Kiroro - Mirai e by Hiroi Kikuri
Language: ja-JP (Japanese) [so-vits-svc-fork]
Description
Extended training at 30,000 epochs and include EffectPresets.xml for Adobe Audition to remove inference noise, it will clean-up when encode to lossy codec such as MP3, AAC, OGG