Voice inference model, RVC v2, 200 epochs