Vibe Coding Log - Sharing Journey - Git - Repeat Glitch/Time Warping FX - Implementing more MIT delays

[audiojunkie]

I've been reading and learning how to constrain results with agentic AI.

These are the main components of a proper generative AI prompt:

* Role/system
* Instruction
* Context
* Input data
* Examples
* Constraints
* Output format

It looks like you are pretty much doing all of these things! At the level of complexity you are working at, it has got to be difficult to get it to stay focused and on task.

[Touch The Universe]

Adding Github desktop to my workflow. This seems handy, though in some ways, working with agents, the builds are all kept locally so it's already a sort of git. But nonetheless, there are other benefits.

[Touch The Universe]

I've been reading and learning how to constrain results with agentic AI.

These are the main components of a proper generative AI prompt:

* Role/system
* Instruction
* Context
* Input data
* Examples
* Constraints
* Output format

It looks like you are pretty much doing all of these things! At the level of complexity you are working at, it has got to be difficult to get it to stay focused and on task.

I'm definetly not doing all that, haha. I wouldn't mind sharing the gpt prompt pages so you can see the level of my prompts if desired, some are very tedious, but i do not always have examples NEVER constraints.

Here is my Chat discussion for that harmonic generator synth.

https://chatgpt.com/g/g-p-6a0e6eb055088 ... fd7de25def

Here are examples of my latest prompt inserting new delays MIT code on one page to make sure it seperates the knobs and treats it as unique - basically just have unique plugin parameter ids for the knobs etc.

PROMPT 1

Let's work with the Caradag delay, but remember what we're talking about a while before, how we have, instead of making a new tab for every delay that we're making so it's entirely unique, like we just want to have, we want to isolate the code so that all the knobs correspond only to the delay. So in the menu, when we're selecting the Carad delay, it'll re-populate an entire new set of knobs that are only directly linked to this delay that we're adding. For instance, if the Coco delay has 20 knobs, then when we set Carad delay, if the default is 12 knobs, then it'll have 12 knobs and there'll be 8 not even existing, showing up. Because we want to minimize, we want to have zero possibility whatsoever of any knobs being shared between different modules so that they do not get mixed up. Kind of like how I have no problems whatsoever having the OSC vital knobs mixed up with any other page, like the OSC master section or the thorn section. They're completely separate, but this might get complicated because these delays are on the same page, but they need to be isolated in the code as though they're on different pages and tabs.

PROMPT 2

And for the delay routing and the explicit names, whatnot, like how are we gonna do this? Is this like, we are creating a new effects insert, delay, or whatever, and then, like, I wanna make sure as we're adding new things that the audio routing is being updated intelligently and it's going well so everything's explicit. Um, because we were going off about that before, um, but I don't even, you'd have to search through the thing to, because I don't even know what prompt to give you to kind of ensure that, but the idea is that for CPU reasons, we want things to be explicit. Um, but I don't even, yeah.And I'm just so, I don't know if we're updating the right files. Did the delays that we're adding, Coco, and then the wet delay, were those just added into the plugin editor and plugin processor.cpp, or did we want to have a new type of like delay.cpp, so we're just handling smaller files? But I do know that we need to kind of update the plugin editor as well eventually, but so, I mean, I, I don't know, how do we go deal with that front? And remember, on this page, we have the visual interactive graphic display editor there. We want to keep that, so that is being shared. And the new knobs that we're adding need to affect the controls there eventually at some point as well. So, but still, at the same time, the code needs to be isolated.And so, you're right, how do I create a new .git right now? Does like a before karag delay, like, would it only let me create a milestone unless I add new things? So you need to walk me through that process.

Prompt 3

Oh, and also, um, we're not recreating the delay from scratch, is that right? Do you understand that? Like, we are implementing the exact and only the exact code that's inside of the source code. We're not making things, taking shortcuts or facsimiles or this or that. We're implementing it. We've gone through the trouble of testing each of these .htmls. I did, we tested a whole bunch. I narrowed down these ones, so I don't want to go through the trouble of, um, that's the whole point of narrowing down these delays to implement. I've already tested and narrowed these things already, so I don't wanna just recreate another delay from scratch, even if it's a similar to that or whatever. Um, for right now, I just want to implement these delays as is. We can always add features on top of it, but for right now, let's just make things nice and simple, copy and paste and using the exact source code. It's all MIT. Hopefully that's clear.

As you can see, it's kind of messy, just voice to speech as I'm talking.

[Touch The Universe]

From last sessions progress, I had 1 OSC from the monster synth that is per voice filter that I couldn't play a single note for some reason. I extracted that from the 1.7mb file to a .50kb file. I added cpu saving stuff, then more features and detuning, which sounds lovely. I have a menu that one can select from a dozen detune laws, spreading, where even filter/phase,subtle distortion can be spread in the voices, even an eq and white noise for insanely lush full spectrum saws. I can play 16 unision 8 poly note chords and it's around 20%. If I utilizer per voice filter spread, it's 40%. I transplant this renewed 1 OSC, beefed up back end, see similar improvements, plus a little overhead 10% from presumably the other gui controls. The gui open was causing a doubling of the cpu open versus closed, but that has been fixed. I forget what the fix was, will see if I can document that.

This is an encouraging workflow. If I can extract any page, tab or section, isolate into a new synth or fx, then fix, reimport it, I can either save all my work, one way or the other, and refine problem areas in isolation with a smaller footprint code for agents to do with. This in my mind just encourages me to keep adding to this thing because there is a secondary branch paths available. I can always turn just one of the synth pages into its own thing. How they all play together, is another matter, yet I don't want to hallucinate problems where there are none. I believe I assumed there was issues with signal paths and explicit things but only because I wasn't aware and intentionally adding them, so I assumed I missed an important step, yet the synth works, the cpu issues are not related to this issue. There are a few areas that I made complicated, that arbitrarily bypasses the filter on this page and the amp here, but that can be remedied. Eventually, I'll do an optimization to ensure things make sense explicity wise with the routings, but for now, I'm going to focus on adding all the features I want, even if it's madness to keep adding to it. It's not bigger a monster than unify, which in my opinion is a similar type of monster synth, though it hosts vst's instead of embedding them, or perhaps they do embed them, regardless.

[Touch The Universe]

Useful prompt from going to html testing to css.

Implement this HTML as a behavioral spec, not just a UI reference. Before coding, audit every control and every state variable in the HTML. For each knob/menu, identify: parameter range, default, mapping curve, dependent variables, smoothing, modulation phase, stereo behavior, feedback routing, output routing, and any hidden interactions with other controls. Then implement the C++ version only after producing a checklist proving each HTML behavior is either implemented, intentionally skipped, or deferred. Do not simplify stereo/decorrelation, feedback routing, smoothing, or hidden modulation unless I approve it.

[Touch The Universe]

[Touch The Universe]

Work in Progress in HTML not incorporated yet.



On a secondary note, I head back from a developer and there is a chance I might get a chance to incorporate some old code from a now discontinued though once beloved synth - and some effects from a daw.

[Touch The Universe]

I explore in depth all the delays, some I notice are a bit broken cpu wise though i have my buffer set at 512. Maybe I'll go back and check out delay cocoa because it's one of the best ones. The IV-X delay should have a stable tone but i noticed it's jacked in this one. It's already fixed. Next time....

[Touch The Universe]

IVXDelay_TapeEcho_HTML_Lab_v3_ScrapeLFO.html.zip

Here is an html delay to test if anyone is interested. I'm trying to transplant it into css/juice but I'm not quite there yet. I'm pretty sure the html is a facsimile of the GPL3 code so we might be able to implement it exactly, especially since I added 5 knobs to affect the tone scrape itself, yet GPT won't let me. I might need to verify if I can use the code or not in a deeper dive though haven't thought of a way to do this.

Edit... have claude looking into it...

I'm only interested in implementing these knobs from the HTML: scrape, scrape tone, scrape res, scrape amount LFO, scrape tone LFO, and scrape LFO rate. I'm pretty sure those last five knobs besides scrape were the ones that I myself implemented inside of the text, so I can use that. It's my own unique kind of addition to this.But I just need to verify from the source code itself—we changed it enough that I can implement it directly from the HTML, from the original source code that's GPLv3, and I'm trying to implement it into my commercial synth. I need to be completely sure before you do that.Compare which knobs are common and the knobs that are kind of borrowed over. Can you verify that they're perfectly implemented? Or if the HTML is unique enough for me to implement as is just those five knobs, that's the only thing I'm concerned about.