Avenue for exploration / vqgan + crepe

[johndpope]

https://github.com/marl/crepe

This image represents audio pitch detection via crepe. Is it possible to use something like vqgan to glue this and unagan together ?
you could basically explore the latent space by drawing different squiggles

I have lots of vocal stems to help out testing.

[johndpope]

I found this - https://github.com/v-iashin/SpecVQGAN
It's not quite solving what I want above - but potentially simplifies the problem.
Take a picture - and spit out audio - change the squiggle - get a different rendition of vocals.