I've been thinking about the current possibilities of voicebanks a lot recently. We've been making huge steps forward in making voices sound as close to their source as possible but it's always been very.. clean voices. Unique ones, but they have one particular voice quirk or singing style that can be recreated more or less easily. But I've been wondering if this might also be possible now for voices and singing styles that are more.. all over the place without it just sounding weird?? As you guys might now I really love Fukuwa's voice and way of singing but he quickly switches from a pretty singing voice to a rough and and kinda scratchy sounding one, to growls and weird pronounciations etc. Back in the day I always thought you'd have to create multiple voicebanks to recreate this but maybe it would be possible now with AI and enough training? But I also wonder if it might get annoying if the style switches in places where you don't want to but you can't really do anything against it 'cause AI. Hmm. I'd love to see an experimental voicebank that does something like this. Even if it would never be released, just a test to see what is even able to be made currently.
This probably didn't make any sense at all lol