Real Future Of AI in Music: Production Copilot Coming To Your DAW?

brisket · 2023-04-19T20:40:31.357Z

Humans are currently good at making distinctions between artworks. We used to be good at distinguishing trees, now most people couldn’t tell between a pine and a cypress. We make distinctions within the category of things we have adequate exposure to.

I suspect AI content will be preferentially pushed because it is algorithmically generated and thus meets the criteria of other AI as to what is worth showing. Think of those YouTubers who apologise sheepishly for their cheesy thumbnails but do the OMG thing anyway because it’s what the algorithm wants, not what they want to do or what their (still mostly human) audience enjoy.

If the public’s listening habits weren’t dominated by streaming services that serve up a proprietary mix of payola, in-house royalty-free Muzak, and whatever your Amazon cookies say you should like, maybe this dynamic wouldn’t work.

Anyway I hope I’m wrong. But what I see already is humans bending to make themselves more attractive to the content shitfunnel, not AI adapting to human preferences. Parasocial relationships with digital entities are already more important to some people than being a member of a human community; this is part of the same trend. Sure it’s just expressed preferences of individuals, but in toto it is a collectively shared episode of lazy, pathetic nihilism. Moloch wins because we want Moloch’s approval.

marigosa · 2023-04-19T21:37:19.781Z

Fantastic discourse between you two! Ya love to see it.

MythMaker · 2023-04-19T21:50:50.211Z

“The technology we use for beauty and creativity adds to the good of the future. The technology we use for utility and ego-fulfillment adds to the evil of the future.”
-Rudolf Steiner

I suppose this sort of tech can be used for either thing.

Follow the golden rule of artistry: Do unto your culture what you would have your culture do unto you. Or something like that. Basically, will your ancestors and descendants be proud of you or ashamed of you based on what you’ve done and how you did it?

Jukka · 2023-04-21T18:38:29.481Z

MusicRadar – 21 Apr 23

Almost 60% of musicians are already using AI in their productions, study...

28.5% say that they’ll never use it, though

SouvlakiPlaystation · 2023-04-21T19:49:25.511Z

I’m probably just rehashing what others have said, but AI will be a boon to a certain type of producer. IE: People who watch countless YouTube videos on “how to get that professional, radio ready sound”. These are people who are already following strict, cookie cutter formulas to get as many listeners as possible. That kind of shit may as well be made by robots anyways.

As for me I will continue to build everything from scratch for my 7 monthly listeners on Spotify.

Edit: I’ll add that I absolutely believe AI can be used in an authentic and human feeling way - as others have mentioned LFOs, arpeggiators, and dynamic EQ plugins like Soothe are already being used, and that hasn’t destroyed creativity. Much like sampling it will depend on the user to decide whether it’s a crutch or simply another tool in the toolbox.

onthebannedwagon · 2023-04-21T21:12:50.276Z

When AI starts telling “artists” what drugs to use I’ll start to worry.

Fin25 · 2023-04-21T21:57:24.053Z

Mescaline.

The answer is Mescaline.

onthebannedwagon · 2023-04-21T22:06:25.503Z

I won’t need a machine elf to tell me my AI assisted music sucks.

sacguy71 · 2023-04-21T22:12:47.043Z

I would love the perfect AI mastering recording tool that capture live performance jams and perfects to studio release ready albums. Would save me hundreds of hours of DAW work.

sleepside · 2023-04-21T23:31:43.893Z

Grumo · 2023-04-22T00:24:14.649Z

I’m with you and @m0ld, sometimes we are just working alone when we would need someone to help when making a decision.

The usefulness of AI tools for me would be giving some ideas when you need them, not to make the whole thing.

Grumo · 2023-04-22T00:56:08.431Z

Anyway I think that there are tools that make sense, like a wheel, a pulley, a lever… And other tools that completely take the fun out of an activity.

That’s the reason why people still paints without a computer for example but no one makes concrete music cutting tape by hand anymore.

At the beginning of Dall-E I was very excited with the results, now AI generated images are boring for me, you can be curious, but they have no real interest.

Maybe the imposture can become convincing for a lot of people in many areas, but a machine will never have “soul”, and surely this is going to show the value of the human when creating anything.

A machine could give you the music you prompt, but it will never be in the mood to create music, it won’t have any urgency to express anything.

This phenomena reminds me that sometimes we just need a friend with good judgement to count on when you need a second opinion.

earthangel · 2023-04-22T01:09:36.092Z

now AI generated images are boring for me, you can be curious, but they have no real interest.

I’m still pretty intrigued by the image generators. Messing around with re-training using a smaller subset of images and being really precise and iterative about the prompting has shown me that tools like Stable Diffusion continue to reveal hidden depth in their potential for output. Similar to the way that audio synthesizers do.

I don’t think I have much interest in asking a machine learning algorithm to write melodies for me but I probably won’t be able to stay away from future audio generators for the purposes of exploring new sound design possibilities.

Grumo · 2023-04-22T01:31:18.903Z

Of course if these tools are well used that can be awesome, I mean a lot of things I see that can be surprising in the first moment but then there are thousands like that.

I’m very curious about a lot of AI uses and tools, and how people will be using them to inspire themselves, for example.

Only a reminder: in the school they told me that luddites were against the technology, the machines, the progress… I think their problem was that the owners of those machines condemned them to misery, precarity and a new slavery.

So, is progress always good? Progress to where? Leaded by who?

chaocrator · 2023-04-26T07:08:52.977Z

there are two coexisting paradigms of «music»:

harmonies, melodies, all that stuff. i call it music for musicians.
«interesting textures», «advanced effect chains», «creative randomization» and all that stuff. i call it music for non-musicians.
currently this paradigm prevails, since either mainstream (hip-hop based) and underground (techno based) genres don’t require any tunes at all and can be entirely made just of beats, textures & effects.

AI sucks when it comes to making music from notes.
but AI certainly can and will rock in doing all that fancy technical stuff.

Volt_Thrower · 2023-04-29T07:28:13.193Z

I’m curious what are the best AI tools for music right now? The whole Drake/Weeknd thing is all over the news being touted as AI generated music. However, beyond deepfake vocals it seems like the track was created and produced by someone like any other track. I’m not sure how much of it is supposed to be AI or not and the media kind of misunderstands and dumbs down the idea to make it sounds like someone just types in a prompt.

trackerjack · 2023-04-29T08:47:01.188Z

Somebody please train an AI to compile all of @sezare56’s posts about the Octatrack into a nice manual.

That would be great.
Nothing against that Merlin guy…

MichalHo · 2023-05-08T17:21:43.334Z

Some generative dub, using an early version of an upcoming generative audio tool.

Taimi · 2023-05-09T10:30:13.291Z

Oh man, sounds good. What kind of tool is that nendo, what does it do exactly?

g3o2 · 2023-05-09T11:42:22.500Z

brisket:

it is algorithmically generated and thus meets the criteria of other AI as to what is worth showing

It is unfortunately more complex than that: the output is based on what the algorithm generates given the data on which it was trained and also given your input, if any is required. So if the train data set contains copyrighted material, that’s what it will potentially use for the output.

“Fixing the copyright issue” means that you would either need to feed the algorithm with data based on a free-to-use licence (e.g. creative commons), or have the algorithm check the output for similarity with copyrighted material and “punish” it when it gets too close.
The former reduces the quality of the AI’s output because there is less free data to train on, while the latter is not foolproof because it suffers from the lack of definition (how close is too close?) and the quality of the underlying database.