HDVM retakes feature flaws and Feature suggestion.

Hello, I Have a rather annoying Bug with the HDVM feature. First of all, I know it is not perfect since it use AI randomness. And I just found an idea to make it a bit better.

The issues :
In certain take, There can be a small drop out in waveform which cause kevin (I only have him as full voicebanks, I don’t know for the others voices) often say “Wo-ords” instead of “Words” like a badly configured UTAU oto (it happen randomly, It is not just specific takes/lyrics)

My Feature suggestion for this :
A Seed feature for AI HDVM Pitch/Dynamics feature, Instead of generating Multiples take and choose the ones which sound good, We can Randomize a Seed number Until the Pitch/Dynamics sound good.

takeglitchaudio.zip (316,0 Ko)


We can Randomize a Seed number Until the Pitch/Dynamics sound good.
Isn’t that what the re-take feature is? Just the seed itself is obscured to us.

I did some brief testing, and I managed to recreate the break. I found it wasn’t the pitch takes doing it, it was dependent on the expression takes; sometimes you just hit a bad one. It can also be fixed by using alternate pronunciations in the note properties.

Unfortunately, I think these are pro only features. I did manage a work around just by splitting the note into 2 and then just making the first [were] and the second [erds], same phonemes but gave a different output.
Note this didn’t work by just using [words] [-] or even just manually changing the phonemes to {w er} {er dz}; you need to actually change the note into something else for it to re-calculate the AI takes.

1 Like

Thank for the solution for the waveform break bug, But I already found a solution for it.

What I mean about the seed number feature, Is to make it Visible and be able to Edit the number Instead of generating multiple take and try one by one for the best takes.
Best regards.