Originally Posted By: Byron Dickens
I think it is that they are too pitch perfect and lack the various inflections that real singers have. But I also think that because they are initially generated pitch perfect they also lack some of the weird artifacts that heavily auto-tuned vocals have.

There's actually an option to "auto-tune" the vocals, that does the exact opposite of what you might think. In the Vocaloid and UTAU community (which is where SynthesizerV came out of) the term doesn't refer to pitch correction - the pitch is already "correct", as you hint at.

"Tuning" in this context refers to the process of adding the elements of performance such as vibrato, over and undershoot, and the "imperfections" of a real vocal performance.

SynthesizerV is trained on example songs in the singer's style, so it's able to apply the performance "tuning" automatically. At least, as far as the pitch is concerned.

So the vocal actually does have the imperfect pitches inflections that a real singer has.

There are a lot of other factors that you can control in SynthesizerV, such as Loudness, Tension, Breathiness, and so on. But there's no obvious way to observe these features - and thus train a neural network on them - so only Pitch Deviation is the only "automatic" feature in SynthesizerV. The rest would have to be tweaked by hand - which in this case, I didn't do.

-- David Cuny
My virtual singer development blog

Vocal control, you say. Never heard of it. Is that some kind of ProTools thing?