Description

Trained on 90 minutes of clean audio direct from game files. Works exceptionally well on mid-range to lower voices (not too deep though) using either rvmpe+ or mangio-crepe as low as 100 (in fiddling, lower is probably fine too). Works great female > male down an octave or so without needing any formant shifting either. < There is another model which is done well, however, it sounded too nasally and boomy for the kind of singing covers I wanted to do, though I'm sure it sounds excellent with talking though! It is likely just the difference between mangio-crepe and rvmpe, each to their own. NOTE: Astarion's charismatic way of speaking is still dependant on the voice you are applying model to, so don't expect them to sound as charming as he does otherwise :)