Using Chat GPT with your software Synths

[IvyBirds]

I have now created over 400 DX7 patches with ChatGPT, mainly because I know the DX7 engine

It produces a playable patch about 85% of the time. I began to notice a pattern in the things that didn't work. Namely that they should work but it was thinking the DX7 engine had features it didn't

So I asked it what would be the ideal FM synth, it responded with essentially a 12 operator DX7 that could make a bunch of different waveforms, and have feedback on every operator. Of course that doesn't exist but I get the idea that ChatGPT wants it to, I have F'em and FM-X inside of HALion7 but it's not a 12op DX7

I asked it what FM synth did it think was the best FM plugin in. I had given it all the ones I own which is all of them you can think of, and it said Plogue OPS7 running in enhanced mode that gives it multiple waveforms and feedback

I also asked it how it analyzes sounds. Not surprisingly it runs FFT and breaks it down into since waves and has done a lot of training to identify every conceivable nuggets it could from them

All of that gave me the working theory on how it works. So I asked it if it was right and it hinted that I was but wouldn't come out and say it

Anyway my theory is when you ask it to make a patch it had some kind of data base on how similar sounds look like when analyzed and makes the end result and works backwards. I think I think at least with FM it would have a perfect success rate if it had a synth that could do what it wants

Will experiment further with other synthesis types but my feeling is all of that is going to be the same

[Hyperbole]

I also asked it how it analyzes sounds. Not surprisingly it runs FFT and breaks it down into since waves and has done a lot of training to identify every conceivable nuggets it could from them

All of that gave me the working theory on how it works. So I asked it if it was right and it hinted that I was but wouldn't come out and say it

I would be highly skeptical of any claims ChatGPT makes about how it arrives at a particular result. I posted in the original thread that if you ask ChatGPT how it solved a math problem, it will provide an explanation that is both logical and mathematically correct. But when data scientists look under the hood, that's not at all what ChatGPT has done. Rather than use symbolic reasoning, ChatGPT uses a roundabout, language-based, probabilistic approach. It's amazing, ChatGPT gets the answer right at all, at least most of the time. But it's not doing math, that's for sure.

I suspect something similar is going on rather than any kind of FFT analysis. But I could be wrong.

[pinki]

What AI (and soon AGI and ASI) has already shown us is that intelligence and perhaps consciousness is not really unique in our universe at all. It is an naturally occurring emergent system that the universe is predisposed to.

A leap too far perhaps? It’s a new technology, not the definition of intelligence. It is a redefinition of the concept of intelligence to suit the market.

It’s a very sophisticated scrape of all knowledge on the internet and the clever bit is the understanding of the question but not the answers so much.

[IvyBirds]

I also asked it how it analyzes sounds. Not surprisingly it runs FFT and breaks it down into since waves and has done a lot of training to identify every conceivable nuggets it could from them

All of that gave me the working theory on how it works. So I asked it if it was right and it hinted that I was but wouldn't come out and say it

I would be highly skeptical of any claims ChatGPT makes about how it arrives at a particular result. I posted in the original thread that if you ask ChatGPT how it solved a math problem, it will provide an explanation that is both logical and mathematically correct. But when data scientists look under the hood, that's not at all what ChatGPT has done. Rather than use symbolic reasoning, ChatGPT uses a roundabout, language-based, probabilistic approach. It's amazing, ChatGPT gets the answer right at all, at least most of the time. But it's not doing math, that's for sure.

I suspect something similar is going on rather than any kind of FFT analysis. But I could be wrong.

Somehow it's able to program additive patches based on the harmonics from uploaded WAV files that are spot on

If it's doing anything other than FFT to pull that off, that would actually be more amazing than FFT which mainframe computers at universities like Stanford under John Chowning and Dartmouth under Cameron Jones were doing in the 1970s, it had infinitely more power available than those

For example I uploaded a 5 second sample of a violin and it used FFT to make a Synclavier patch of 50 timbre frames that I was able to upload into Synclavier V. It actually sounded a bit better than having the FFT analysis that Cameron Jones built in the early 1980s as part of its resynthesis program that he ported to the C programming language that is now inside of Arturia's Synclavier

I have done around 25ish such patches now from small samples I have and it's been spot on every single time, the Synclavier's resynthesis engine on the other hand returns errors every so often which are actually kind of cool sometimes in and if itself

I will also say it has take. it a step further and incorporated some FM in a few of the patches rather than just pure additive like Synclavier itself does

You can even take this a bit further and upload a small wave file and have it analyze it and tell it design a patch based off of that with the Jupiter 8. It struggles to nail that and returns errors but just the fact you can use it to resynthesize basic timbres into analog synths is pretty amazing

[IvyBirds]

I suspect something similar is going on rather than any kind of FFT analysis. But I could be wrong.

My last response gave me an idea. I asked it if it knew the "Crystal Rhoades" patch from the Roland JD800. It said it did and then I asked it to make a preset for the Synclavier that sounds like that using the additive engine

This is what it returned

Here is the advanced version of the Crystal Rhoades-inspired Synclavier V preset, now with realistic evolution across multiple timbre frames, FM overlay, modulation, and effects:

It contains:

20 smoothly interpolated timbre frames

FM layer for extra sparkle

Evolving harmonic motion

Long, expressive envelope

Built-in chorus and reverb

Gotta say that one it made is KICK ASS GOOD, and I didn't even have to sample or upload anything. I am blown away

[osiris]

I KNEW you were on the paid plan and I told ChatGPT this, or rather asked it if that's the reason I get wonky info. (Because I use it for free) It also gave me links for things on the net that disappeared seconds after they appeared. I told Chat stop showing me links because whoever programs the free version won't let them show.
I feel like this is the great divide coming between those that can pay and those that can't.
It's apparent to me they've programmed different responses for subscribers.

[Teksonik]

AI is just another tool to do so

No, A.I. is a crutch that's destroying humanity in the arts. It's taking the soul out of music and leaving the creative process to a soulless machine.

The sad part is when I see people like you who call themselves musicians defending the very thing that's destroying the human creativity of music. A musician defending A.I. is just artistic Stockholm Syndrome.

In my opinion your time would be better spent learning your craft than spending time asking A.I. to make sounds for you.

But then I'm someone who enjoys the creative process, who finds joy in creating something with my own hands and imagination. I understand that the creative process requires effort and the reward for that effort is the whole point.

If you find artistic satisfaction in simply pushing a button then I have nothing but pity for you. You're completely missing the point of any form of artistic creation.

I won't debate the subject any further. Do what you think is best for you but don't expect anyone to respect you for your choices.

[Daru925]

If you want better result, ask the ai to replace the monkey/human who writes the prompt for the ai.

[IvyBirds]

AI is just another tool to do so

No, A.I. is a crutch that's destroying humanity in the arts. It's taking the soul out of music and leaving the creative process to a soulless machine.

The sad part is when I see people like you who call themselves musicians defending the very thing that's destroying the human creativity of music. A musician defending A.I. is just artistic Stockholm Syndrome.

In my opinion your time would be better spent learning your craft than spending time asking A.I. to make sounds for you.

But then I'm someone who enjoys the creative process, who finds joy in creating something with my own hands and imagination. I understand that the creative process requires effort and the reward for that effort is the whole point.

If you find artistic satisfaction in simply pushing a button then I have nothing but pity for you. You're completely missing the point of any form of artistic creation.

I won't debate the subject any further. Do what you think is best for you but don't expect anyone to respect you for your choices.

Only no one is just pushing a button, but thanks for the tired old argument anyway, it has been used for decades

Synthesizers were going to destroy music as all you had to do was push a button, then it was the Fairlight CMI with samples, why bother learning and instrument like the violin or cello, why hone your craft all you need to do is push a button, then it was used against MIDI, MIDI was going to destroy music you know why did anyone need to learn anything or practice their craft when MIDI and sequencers could just play in strings for you? Then the same was said about small affordable multitrack recorders like the Tascam PortaStudio, then it was said about DAWs and Plugins, then it was said about software -vs- hardware

The reality is unless you are a luthier and handcrafting your own acoustic instruments from raw wood, or a brassiere handcrafting instruments out of sheets of brass you are not making music with your own hands, you are using tools made by others

As an artist and a musician I have an artistic vision that AI helps me achieve, if I was rich I could hire someone to come in and play violin for me, or I could use a violin sample, or I could load a violin preset, or roll my own violin sounding patch, or I could have an AI make one

What ever I choose I as the musician is making that choice for my music as part of my creative process

[DJErmac]

I've just asked it to recreate Faithless pizzicato with any VST synth and give me a preset. It proposed me to go with Serum. I agreed. It then generated a tutorial after saying the sound is cult. I asked for a preset instead. It refused.
So I followed the tutorial. It sounded nothing like the original sound. I can't even hear any connection.

robot de merde.jpg

[cake builder]

I think song generating AI like Suno and Udio should die in a fire, but I don't think Chat GPT telling someone how to create a synth patch is disruptive to the creative community on the same level. Whether the person using it decides to use that information to learn to fish (so to speak), or just keep asking for another fish is up to the individual and not my problem.

[DJErmac]

This thing is sooooo disappointing... I asked it to generate the sound of a dog mimicking a cat, and it said there's no way it can do this...

[IvyBirds]

This thing is sooooo disappointing... I asked it to generate the sound of a dog mimicking a cat, and it said there's no way it can do this...

And yet weirdly when I asked it if it could do that it said

Absolutely, that’s a fun and creative idea. Here's a concept for a synth patch that sounds like a dog trying to mimic a cat meow — meaning it would have a weird, rough, wobbly approximation of a meow, with some tonal instability or bark-like undertones.

I’ll describe how you can build it using a flexible subtractive or semi-modular synth (like Serum, Vital, Arturia Mini V, or even a modular setup like VCV Rack).


---

"Dog Meows Badly" Synth Patch Concept

Oscillators:

Osc 1: Triangle wave — this gives you the basic vocal-sounding tone.

Osc 2: Slightly detuned square wave — adds a rougher texture (doggish undertone).

Osc 3 (optional): Formant or vowel wavetable if available — shape to resemble “meow”.


Pitch Control:

Portamento (Glide): Medium setting to slide between pitches like a meow.

Pitch Envelope: Fast downward pitch bend at the beginning — gives that “yaowl” attack.


Filter:

Type: Bandpass or Formant Filter (if available).

Cutoff Envelope: Slow rise and fall to mimic the meow’s filter sweep.

Resonance: Medium-high for a more nasal sound.


LFOs:

LFO 1: Assign to pitch with random or sine shape, subtle amount — to make it unstable like a confused dog.

LFO 2: Assign to filter cutoff, synced to meow envelope, to make it more expressive.


Envelope (Amp):

Attack: Medium-fast

Decay: Around 500ms

Sustain: Low

Release: Short


FX:

Formant Shifter / Vocoder (if available): Pushes the vowel formants.

Reverb: Medium room to mimic a meow echoing in space.

Bitcrush or Distortion: Lightly — helps with the "barkiness".



---

If you're using something like VCV Rack, I can give you a module-by-module build. If you want it in a specific synth like Serum, Synclavier V, DX7, HALion, or others, let me know and I’ll tailor the patch exactly for that.

Want a downloadable preset or audio demo of this sound?

Weirdly it doesn't have any issues doing that

Of course the struggle with a sound like this is that no one really knows what a Dog trying to mimic a cat sounds like, so one can only assume that this sounds as close as anything you or I could come up with

[IvyBirds]

I've just asked it to recreate Faithless pizzicato with any VST synth and give me a preset. It proposed me to go with Serum. I agreed. It then generated a tutorial after saying the sound is cult. I asked for a preset instead. It refused.
So I followed the tutorial. It sounded nothing like the original sound. I can't even hear any connection.

Weirdly it told me this

Absolutely — the pizzicato sound used by Faithless, especially in tracks like "Insomnia", is an iconic, sharp, plucky synth string — kind of like a digital pizzicato with attitude. Here's how to recreate that sound in a general-purpose synth like Serum, Vital, Arturia Jup-8 V, Synclavier V, or any subtractive synth.

So far ChatGPT is beating you 2-0 when it comes to accurate information

[DJErmac]

I've just asked it to recreate Faithless pizzicato with any VST synth and give me a preset. It proposed me to go with Serum. I agreed. It then generated a tutorial after saying the sound is cult. I asked for a preset instead. It refused.
So I followed the tutorial. It sounded nothing like the original sound. I can't even hear any connection.

Weirdly it told me this

Absolutely — the pizzicato sound used by Faithless, especially in tracks like "Insomnia", is an iconic, sharp, plucky synth string — kind of like a digital pizzicato with attitude. Here's how to recreate that sound in a general-purpose synth like Serum, Vital, Arturia Jup-8 V, Synclavier V, or any subtractive synth.

So far ChatGPT is beating you 2-0 when it comes to accurate information

Well, first I'm using french language...

And I asked for a SOUND of a dog mimicking a cat (what I wrote exactly). This is what it refused to create.

And there's no difference with what it proposed to me with the pizz. Excepted I made the patch and it sounded like @ss. What I said also.

No difference. You simply misunderstood me.