Im talking about what the biggest barrier will be to developing a DAW with LLMs. Buying the sounds doesnt let you redistribute them like that, you need commercial redistribution / software-embedding rights. The sounds people demand are going to be a bigger barrier than an agent coding it.
Your options are
1. synthesize the sounds yourself in code. LLMs aren't very good at this. Try it - you will play your sound next to a real TR808 recording and and think "this isnt even close".
2. Record a real machine yourself. Expensive and outside the domain of someone who only has coding experience. And with the 909 there may not even technically be a legal way to do it since some of the sounds are recorded and stored on the device, so you are essentially copy their recording. Many DAWs do do this but its murky.
3. License samples from a sound company. Normal sound pack licenses only cover making music with the sounds, not distributing them in a DAW. You’d need a custom OEM deal (contacting the company, negotiating, paying probably a lot, lawyer to check contract, etc).
4. License directly from Roland (not a thing as far as I know)
You need a real sound designer. But you could make a serviceable DAW with an LLM without having experience writing DAWs.
This is the same problem seen elsewhere, like video games. AI can do the code. It's the media that it can't really understand and someone without domain experience doesn't have the expertise, and doesnt even know what they don't know.
Maybe I'm undercutting my own point - I thought DAWs typically had these sorts of sounds but now that I think about it I recall hearing that sometimes people use plugins for this sort of thing.
And I agree that the assets are a big barrier. If you look at Logic Pro X (my tool): it needs some 80GB for the full install without getting anything else. And without other software, you can expand it with extra loops and sample packs. I guess the interface to those can be reverse-engineered, but even that's a barrier.
On the other hand, refining/tweaking over time may get you closer. I'm quite far down this road, having created a DAW of my own flavor.
It's absolutely not perfect, nor is it a production-level app or even clone, but I trust it for electronic music production because it, like many other programs, does the basics well.
That's another point worth making about the assets. They're a barrier as someone builds them over time. You might have 2000 (example) custom presets and audio file alterations in Ableton and use them as your "sound" as a producer.
Thinking names like Kaskade, Deadmau5, etc. Plenty of others.
Going back to the code, reverb is the example that illuminates the difficulty of cloning in my day-to-day. Sometimes quite explicit in songs, and sometimes hiding just enough in the background. But IMO - stands out immediately when it sounds "off" as I lack a better description.
Anyhow, here's my song from my DAW in case you're curious! Note that it's not mastered so you'll need to crank your volume somewhat.
Thanks for sharing the song. This is material proof that your program can generate a composition. Did you come up with some new ideas about how to create and produce?
Blending lots of ideas from other places (unsurprising I'm sure). I first used FL Studio, and am more recently an owner of Ableton. Some things I like and dislike about both, so what I tried to do with my app is blend what I like.
E.g. FL Studio = step sequencer. Ableton = more intuitive (I think) for whole song composition. Of course, now Live 13 has a Step Sequencer...
The other areas I'm exploring is sound design. One goal in this effort is to remove dependencies on other tools as much as possible. So every sound you heard in my song is from the app alone - not some external sample or other synth/VST tool.
So, then suppose you sit down with a blank slate and draw a wave. Preview it. Now add bits and pieces on top. But more of a visual design canvas rather than the Reason-style rack wiring.
PS: Thank you for checking out the song!
But answering here if I understand correctly: I use (at the moment) a mostly serial chain architecture.
I think what you're asking whether I'm using signals from one instrument (or other component) to drive another. Not as much as a first class feature similar to what you'd find in Ableton.
So, for example, I don't have today something that will drive low-pass cutoff frequency on a synth based on the kick drum sine wave decay.
Same idea you're mentioning?
On my list to get to but not there yet! I tend to add things as they come up in the process of making music, rather than approach this as a software-first-music-second project.
I mean there are some sounds in the public domain or things that you can synthesize yourself. But I'm talking about the basic sounds that people ahve come to expect from DAWs, like TR808 or 909. You probably don't have those, have your own versions which don't sound quite right, or versions you cannot redistribute in a DAW. If you have solid versions please tell me, because I would like to learn.
BTW I'm not saying YOUR app needs those for YOUR purposes but I'm asking to help clarify my original claim that the sound design will be the bigger bottleneck.
Sound design is laborious. My understanding (perhaps wrong) is the TR808/909 sounds are high-quality audio samples that the DAW maker licenses.
I'm slowly picking apart the 808/909 sounds to attempt to get close, but if I'm being honest, I don't love those sounds as the basis for everything I'm doing! My way of saying "my versions that don't sound right"
So - I spend more time on what I create from scratch!
To be clear, I think sounds synthesized by LLMs and public domain stuff goes a very long way. But when you want to create something very specific you're going to run into a wall. Same for 3d models, video, etc.
Oh I'm sure - but that's not the weakness of an LLM. An LLM can handle all sorts of complicated signal processing.
It cannot reproduce the sound of a 808. And if you give it a benchmark and a candidate it cannot hear the difference in a meaningful way nor determine if it sounds "good." It cannot do it. Only a person with extensive domain experience can.
When I think of a DAW, I think of the interface, the VST support, and the built in plugins.
I have so many sound assets that I've accumulated over the years, I don't need or want more.
Not saying it wouldn't be important to some people, but that's not important to me and I don't think I'm alone.