Ok so I have Midi files from my sheet music scores over the years, for things like my ensemble, orchestral ,and chamber compositions… and what I’m wondering… well I mean is what I want is to process them through something like an AI Orchestrator? So MIDI>Audio, instant like, every part of the score/orchestra is tracked out, given patches/VSTs/instruments then automated according to dynamic expressions and tempo changes, etc. Like I’d do it myself but I got other shit to do man. I don’t want to go through the process of picking all the instruments and tweaking them, plus all my orchestra VSTs are shit right now.
Goddamn, that sure would be helpful for me if something like this existed!
YEs I could put together an ensemble/orchestra/choir to PERFORM it, that is idea., but this is more so I can give decent quality(ish) audio version of each person’s part because honestly that’s how a lot of musicians do it, if any of you here play in “high art” music situations/read scores.
TL:DR – Save me time, make a robot orchestrate my scores for me.
Ah, you’re looking for modular. You just don’t know it yet
Doing this all-freeware is the real trick. I have setups that’ll pretty much just play shit like domino rally but they cost a pretty penny. Tempo changes and very meticulous things aside.
A solid DAW template is your next best bet. Have a slot for basses, string instruments, pianos, guitars, drums, etc. One automation lane for tempo changes, etc. Everything can have its dedicated VST and then you’re kind of set. Use markers for launching specific songs and all of that. LABEL EVERYTHING (that’s the real kicker, innit)
I understand all the stuff already… and I have wanted a modular a long time, don’t I know it
no… I really am INTERESTED whether or not this kind of thing exists. I know all those other things, my point is to see if someone has already designed a bot that can orchestrate scores for you easily on the fly.
imho it’s a nice idea and I haven’t seen something like that yet. might exist already or be in the works somewhere though, development of these tools is kinda fast these days…
I’m not aware of anything out there yet that can do this, though i would imagine under the hood that Suno or Cassette might be doing this it’s just not available to all to put in the actual midi score. I’ve played with ChatGPT for maybe like 20 minutes with a friend to see what midi stuff we could get it to do. I didn’t try uploading anything for it that time, like examples, etc.
but ChatGPT isn’t going to be able to pick instruments and generate the audio for you. it’s audio abilities are largely limited to the PyDub library and the operations it’s supports.
I thought Suno might be able to do it but yeah not quite there yet. I think the classical/orchestral/scoring world could really benefit from this in a lot of ways. I’m just now remembering an app where you can scan scores with your phone camera and have it automatically played but the quality is shit
yeah the tools still have to be built for ai to call in order to complete this type of a task. you could probably show it how to take a lot of different midi files and assemble them into one file you can open at this point but i don’t know that it would be much of a timesaver. but it’s a task that would need to be done prior to selecting sounds none the less.
i’m regretting not engaging more with CrewAI at the conference I was at last week. can’t seem to get an account to play with their stuff to see if it’s worth using.
I haven’t been that hot on the uses of generative AI, but this actually seems like a perfect use case for it, letting composers prototype their scores.
All the current tools have done a lot of the hard work - training the neural network on a shitload of music to be able to predict ‘next in sequence’. But it looks like they’re all focused on spitting out tunes for a subscription fee instead of exposing the interesting parts like inputting the score and then tweaking the output. Because AI is now for-profit, none of these look to be open where you could hijack the API to leverage their training.
What you’re asking for isn’t terribly hard from a programming/training standpoint (though also not trivial from scratch, with some potential big gotchas in the output section depending on how tunable you want it), but we’re unfortunately in a moment where interest in AI completely outstrips resources and everyone is chasing dollars and unicorns in lieu of useful tools like what you want.
If you’d asked for this a few years ago, some CS grad student would have made it for you pro bono and then written a dissertation on it. Nowadays you’re competing with Big Tech and Walmart and probably the Vatican for Nvidia cards and compute space and programmers so they can build a microwave that will predict whether you want a baked potato or a pop tart the next time you open it.
tl;dr - I don’t think what you want exists because the money is pointed elsewhere. It’s not that hard to make it exist, but everything’s tied up with the money right now.
