Originally Posted by dcuny
... First, some clarification on SynthV. A voice provider is basically given a "script" of words to sing at different pitches. English has a lot of phonemes, so there are lots and lots of words to sing. The original version of SynthV didn't use AI at all, it just reassembled the voice similar to how BiaB works. Later, they trained a neural network because it was better at connecting all these phonemes.

The accent of the voice provider will come through. Here's an example: https://soundcloud.com/dcuny/ive-been-down

And yes, there are a number of mis-pronounced words.

There's no way to remove that accent from that voice provider.

Most of the English voice providers for SynthV have been North American, so there aren't really many UK voices. With some effort, I do some substitutions and edits and try to approximate a Liverpool accent, but it would sound very much like an American trying to do a (poor) British accent.

There are other ways to go about this. For example, RVC is the process of training a neural network on a specific voice, so it will take an input voice and replace it with the timbre of another voice. That will get you somewhat in the ballpark, but the original singer should optimally have something close to that accent in the first place.

I suspect that eventually there will be more UK-flavored voices for SynthV.

I have ACE Studio (but the voices are IMO, not great), and none of the voices sound British to me. I just loaded the MIDI output of this song onto ACE Studio, and verified the voices aren't great. They aren't terrible, they just seem to be missing phonemes and not sound especially realistic. They're pleasant enough to listen to, though.
Thanks David for all the info on AI vocals and responding to my points. I've never used AI vocals and probably won't at this juncture, but I need to 'do my research'. This has been really interesting.

BTW - I LOVE that song 'I've been down' - full of quirky energy! I don't think I detected any maj7 chords though!
Andrew