vibecoding codebox

sousastep's icon

so ai can implement any academic paper or open-source effect in any language now..

and I've always wanted a whole library of fx to choose from for my rnbo rig, so I generated 200+ effects while developing this workflow for researching, prompting, verifying, testing, and utilizing vibecoded rnbo codebox: https://github.com/Sousastep/sousa_fx_generator

they can be viewed and downloaded individually here: https://aifx.sousastep.quest/

Mitch Triplett's icon

Thanks for posting all this!

Are you doing any signal processing pre-pitch tracking that improves pitch tracking output? eg, compression? gating? etc?

sousastep's icon

I'm not doing any signal processing pre-pitch tracking, as I've always been of the opinion that pitch trackers work best when given direct mic audio. YinFFT has a "confidence" output that can essentially act as a gate. Now that you mention it though I do wonder if it'd become spurious on a loud live stage..

saw your forum post about rnbo envelopes; there are seven of them under ai_tuba_env that may be worth checking out :)

Mitch Triplett's icon

Yeah, I'm with you on the clean input. My question isn't about "effects" pre-pitch tracking, as much as it is about ways one might "normalize" the direct input to get better output. Your V2 patch is quite good, but I'm still getting some artifacts (yes, I am adjusting technique). So looking for ways to minimize the artifacts... other than "play differently/better" :)

I'll be checking out the envelopes in due time. I'm currently using some DataKnot patching that I already had set up.

Thanks again.

Cheers!!

Roman Thilenius's icon

it would be quite hard to decide how a generic building block for "pitch tracker" could look like, there is too many options what you could do derive a control signal of certain specs, and there are many different applications beside the classic "human singer correction".

for the me the smallest building block is simply the analytic frequency of the music signal.

a common alternative is autocorrelation (which is required for polyphonic tracking.)

pre- and post-processing around those are variables.