Using Chat GPT with your software Synths

[Examigan]

I asked it for a Diva vst patch in the style of Bladerunner and it spelled out how to do that in text. It also asked if I want a preset file for Diva to download (made from the same steps it gave me). If it is just the main lead sound everyone knows, I can create that, but not some of the others.

[osiris]

The problem with AI is it can't hear. I asked it for something audio wise and it told me that. Then I asked it how it hears what I speak through the mic and it told me that it was translated to text for it to 'hear' what's being said.

[jules99]

ChatGPT also spits out MIDI now.

Care to elaborate? I tried, prompted it to create a Boom Bap beat in 90 Bpm. It showed me all kinds of right and wrong research and where else you can create said beat, but in the end CGPT saif it wasn't possible.

[jules99]

The problem with AI is it can't hear. I asked it for something audio wise and it told me that. Then I asked it how it hears what I speak through the mic and it told me that it was translated to text for it to 'hear' what's being said.

I'll play devil's advocate here. Don't we, in that sense, also not "hear" sound, but our brain translates what the ossicles send it (in a very simplified manner) into meaningful sonic experiences?

[Hyperbole]

it's a glorified search engine.

Lord knows LLMs have serious limitations, not the least of which is making authoritative statements that a five-year-old immediately knows are false.

But calling ChatGPT “a glorified search engine” sells it short. It’s what search should have been all along. (I guess in that sense it is “glorified” search.) Rather than return a list of dumb links, LLMs will synthesize all available information in one concise (or lengthy if you specify) summary. That alone is a game changer for me. I use ChatGPT for almost all of my search needs these days.

Also, Google search can only dream of generating patch files.

[Examigan]

ChatGPT also spits out MIDI now.

Care to elaborate? I tried, prompted it to create a Boom Bap beat in 90 Bpm. It showed me all kinds of right and wrong research and where else you can create said beat, but in the end CGPT saif it wasn't possible.

I asked it to create a MIDI drum pattern for me that alternates between 7/16 and 5/8.
Here's what happened:

"Here’s your MIDI file with the alternating 7/16 and 5/8 drum beat:

Download 7-16_5-8_drum_beat.mid

You can load this into any DAW or drum sampler to hear the groove. Let me know if you want variations, fills, or a longer loop!" ​

It worked, not too well though. I had to go back and ask it to add hihat and other things to make it use more of the kit. You have to be a bit more specific to get good results.

[IvyBirds]

The problem with AI is it can't hear. I asked it for something audio wise and it told me that. Then I asked it how it hears what I speak through the mic and it told me that it was translated to text for it to 'hear' what's being said.

Beethoven struggled with hearing and for most of his adult life it got progressively worse until he reached near total deafness, where he could only hear if something was a low tone but not what that tone was or it's timbre. if it was a high tone and again couldn't tell you what note that time was of its timbre

Yet some of his best, most revolutionary, pieces like the 9th were written while he was deaf and couldn't hear them

If you ask it, it will tell you it uses

"FFT (Fast Fourier Transform) to analyze its frequency content. FFT breaks a signal into its component sine waves, letting me see:

Harmonics (fundamental + overtones)

Timbre characteristics (e.g., a violin vs. a square wave)

Spectral peaks (formants, resonances)

Noise vs. tone content

How things evolve over time (with a spectrogram)"

In many ways it "hears" better then we do

It also has a very deep understanding of various Synthesis methods

So when you ask it to design a synth patch, it's able to figure out what the end product should "sound" like using FFT and then is able to use a Synthesis method to achieve that, then is able to use the options available in various instruments to make that happen

In my experience it's been getting right 80-90% of them time

[IvyBirds]

Nice. Do you use any specific GPT models for MIDI creation? I’ve never managed to generate a working MIDI file in ChatGPT. It always outputs empty download links or attempts to connect to external services.

I use 4o. There is known issue with using Android devices returning empty download links for some reason. For me on Android it returns a "file not found" error many times, but it always works on Windows

[jamcat]

Beethoven struggled with hearing and for most of his adult life it got progressively worse until he reached near total deafness

That’s not true. That’s just what he told his wife.

[IvyBirds]

Beethoven struggled with hearing and for most of his adult life it got progressively worse until he reached near total deafness

That’s not true. That’s just what he told his wife.

I do the same every time she tells me I don't listen to her, at least I think that is what she said, I don't know I wasn't really paying attention

[Examigan]

Beethoven struggled with hearing and for most of his adult life it got progressively worse until he reached near total deafness

That’s not true. That’s just what he told his wife.

[Touch The Universe]

Okay back to legitimate chat GDP meanderings, lol would anyone happen to know how far chat GTP can actually take you and programming vstiis?, of course I expect at some point I would need to bring in a developer to kind of finalize things or maybe even pieced together separate bits of code, possibly I mean more than likely probably a gooey artist, and a few other things but I'm wondering if like the bulk of the coding and the hard work can actually be done automatically through artificial intelligence, I asked him that question and it sounded like that's the case like it might be even optional to need even someone to design the layout and a piece things all together but that's so what claims, has anybody heard anything like somebody just designing something from scratch with AI

[jamcat]

ChatGPT can design, write, test, and debug entire code. Which is why it's such BS that there are some developers who can't seem to get their plugins ported to Apple Silicon.

[seafire]

So many experts...

[jamcat]

So many experts...

I am