Advances in AI for musical purposes
I'm really impressed with some recent developments. Also a bit horrified because it's the wild west right now. Better get on the wagon, right?
Hope to get a constructive discussion going, lets just forget the moral or whatever issues and focus on the practical aspects. Anyone up for it?
Comments
My latest hit https://app.suno.ai/song/4e3bbb13-06de-4a4b-83e2-59b95d50cb25/
googles MusicLM is pretty cool. Won’t be long before its everywhere. I find its synthesis kinda fascinating. I had it create various synth noise things and the way it does distortion is really interesting sounding. Sometimes it almost sounds like a piece of plastic wrap resting on the speaker if you know what kind of sound that makes. Its a really light BvvvBvvvvvvvv sound. Ive experimented a little in imitating it with some success.
Im probably gonna have to find an example now huh?
I think its gonna make the value of live music performed by human hands go up. Hearing music thats amazing won’t really be that impressive if you can’t do it live.
It's not going to affect me in the least. I will never use AI for making music. That would rob me of all the pleasure and defeat the whole purpose for me. I don't listen to music all that much any more, but honestly don't care if what I listen to is made by human or not as long as I like it. And I know what I like. I love watching people perform. Real people. Not AI generated videos of people. I love to watch and marvel at their human achievement.
I'm not against AI for music or for art in general. It's just irrelevant to me.
The one area I will happily use AI for is mastering. That's because I suck at it and for some reason am not particularly interested in making the effort to overcome that weakness. I don't see that as much different from sending something off to a mastering engineer though - it's just cheaper and more convenient.
I’m absolutely with you on this one @number37 .
@pedro it’s hard to talk about AI’s helpful hand without getting into morality although I wouldn’t overtly judge someone if they did
I’d never give bots the pleasure to write my riffs, melodies or lyrics because I just enjoy the craftsmanship of doing it myself. I could see deploying some kind of AI automaton to be my live drummer although the audience would never be able to tell the difference whether I’ve programmed something in or used AI, so other than a practical choice I wouldn’t see a point in that
I will always opt for a human equivalent given a choice. We are herd animals but I’m not sure if it’s the same feeling to herd up with a bunch of bots, not unless they start looking like us.
the point of having a mastering studio involved is having a second pair of ears in another room.
making decisions is so much easier when you have a second pair of ears that says this is good, and forget that is not good.
as musician you listened so long to your stuff, you are sitting to close to it sort of speak.
its the 2nd opinion and tips to go on you pay for I guess. 🤔
AI this Ai that.
Who cares if the muzak elevator music is made by some algo?
its not supposed to move you. 🤷🏻♂️
the only thing I saw changing with the new Ai stuff is advertisement.
every time I hear one of those computer voices I press skip, lol
Well, not sure I’d agree. My daughter has ears but I wouldn’t trust her mastering my tracks 😂
Seriously though. Human ear is fallible. It would be more accurate to use AI I think, especially as its anbilities improve with time.
hehe ,
trained professional ears, that are familiar with the genre, I should have said.
That's impressive!
that sounds like anything on the pop radio the last 20 years 🤷🏻♂️ Do we need more of that?!?
now you know what they trained the algo on? lol
but I can imagine it in a hip car commercial 😈
are the lyrics any better?
party , rhythm, smiles and da fonk , not exactly lyrical , everybody's little sister could come up with that 😂
👾
🤷🏻♂️
I mean its enough to fake your way through it,
for things like I have no $ and I need some plim plim background music ?
still haven't heard any "ai- music" that lifts my skirt and makes me think wow, what the hell was that.
practical aspects of AI?
find more or less similar sounding samples on my hard drive in Ableton live is great.
this is really a break through in advanced search. :)
hello AI, create some tacky junk for me, while creating headlines everywhere, isnt very impressive. meh.
We get it lala you don't think it's very impressive. I think it's amazing. I don't plan on using it really but it's fascinating what can be done. Have you heard those Ai cover songs where it makes a Beatles song performed by Queen or something like that?
Let's make Jim Jones sing the cool aid song, oh yeah. ;)
tasteless 💩
Sorry for starting this thread and then not participating for so long, I have had other things on my mind, but I’ve been following and have a couple of remarks.
Hope you don’t mind but I’ll just make this a long post instead of going back and replying one by one.
First off, stuff like google’s MusicLM and meta’s MusicFX, I totally agree with @Johnisfaster this is a fascinating way to get samples. I mean, it’s like a sample library that you can query! And have it adapt the sample if it’s not really what you wanted! It’s not only convenient but I have no moral issue with using it, I don’t see how it’s any different from using a sample-pack (unless you’re a purist that only uses homebrewed samples).
Stuff like Suno I find amazing, just got blown away by the results when like 3 months ago all I heard around was like meh…
But I’m never going to use it, because of the licensing model (you only own rights while you’re subscribed) and also because I agree with @wim that pressing a button to make music for you is the opposite of what I want from music making
But the free tier of suno can be used for a couple of things:
1. Inspiration: it generated a few songs I find myself later humming on. How is that for effective. Anyway, I can later use that inspiration for a vocal line or whatever, and don’t owe Suno anything
2. Stem extraction (which they don’t have yet) using some AI stem splitter, provided you can mangle it somewhat and they don’t have a way to prove it was extracted from their content
But for single samples, and loops, you can get away with a lot, if you include them in your compositions. These companies may well be fingerprinting their audio so be sure to mangle it beyond recognition.
Moving forward, I respect that “I will never use AI for making music” stance @wim, but what about ML models for automating simple tasks? Would you rather go back to doing tedious tasks by hand?
You guys mentioned AI mastering already, I'm totally for it. What’s wrong with using that instead of a super mastering plugin’s presets (which is basically what I ever did)?
And AMP or pedal emulations, or other IR tasks. That’s another field I see AI being very helpful. Can really open up models where you can build a complete synth from components and have it perform like a virtual analogue or something. Replicate classic synths designs, then just change that resistance to see how it sounds. I love this kind of thing, and I think this will be the next major advance in musical technology (and electronic music, as a consequence), you know like the advent of timestretch started a revolution of new genres.
And what about MIDI? that’s actually what interests me the most as I seldom use samples anymore. Stuff like Deep Bach I find amazing, but I don’t really need a fugue generator.
But have you heard about https://www.audiocipher.com/ ?
Who wouldn’t want to have this on iOS? Instabuy at desktop price ($30) for me. Wouldn’t you guys use a generative midi app because it’s cheating? Why does everybody keep asking for randomize functions in every app, then? Well this is generative that you can control, in a way that no amount of random self tuned rules are gonna cut it. You are in control of everything, the creative process is your, the AI is just there for inspiration or sorting the boring stuff (which a randomize button also does, only in a dumber more inefficient way).
And to finish, there’s now some concern about deepfakes in music (as well), stuff like deep drake. Well, my take on that, is that I find it deeply funny (ha), and it doesn’t really concern me, only concerns the artists or their heirs who manage the rights, so I say: bring it. Get on with the times, it’s not like you’ll be able to sue big-AI and the guys doing these deepfakes aren’t worth suing.
Anyway, sorry for the wall of text. I’ll try to keep my comments more fragmented from now on.
TL;DR: I don’t want AI to replace me, but I surely don’t mind it giving me a hand in being more productive.
@number37 sorry I wrote this post elsewhere and pasted here, just pinging you instead of @wim :)
coming up with my own sounds and loops and ideas and a vibe and putting it together is where all the fun is :)
if I was looking for "inspiration" I would look for a partner to work together with instead of downloading random loops of splice or see what junk the ai comes up with. ;)
remember we used to do music together and not alone infront of a computerscreen?
Technology isnt always the answer you are looking for. ;)
Everyone has their use case, but I was hoping this thread would evolve into what can we do with these tools that is fun, rather than to hear everyone’s specific reasons for not using it
Most guys seem to like it though.
Not that I'm already using it while composing (I still find it too much fun to come up with my own riffs) but I've learned quite a bit from AI so far.
Building my own melody generator in Drambo has taught me a lot about what makes a melody sound good and what doesn't. Complete randomness is definitely not beautiful 😄
Writing code with the help of AI is a different thing. So far I've had AI write about 10 python programs so far, and it's impressive how little I had to tweak to make it work as intended.
Exciting times!
And we should keep the discussion relevant to drambo. For instance we can download free sounds already, what if in the future theytrain a model on free sounds only and make it public domain? Wouldn’t you use it? Or if giku gets a sudden AI streak someday and makes an AI powered euclidian? Man I just hope I’ll still be here by then 😅
If you guys want the drama then please go here https://forum.audiob.us/discussion/60176/your-a-i-generative-music-sound-experiments#latest
Within this safe space let’s please keep positive thoughts only 🙏
Did you know that automatic synth preset creation is perfectly within reach in Drambo today?
The Modulator => Meta => Meta Randomizer can randomize any amount of parameters in any desired range.
I've tried it with Zeeon and a choice of 15 exposed AUv3 parameters, randomized within reasonable ranges that I thought should work.
It didn't take long and after 10 minutes I had 20 presets that sounded like a "Classic Analog Polysynth Preset Collection" 😊
I love the randomizer! But imagine a randomizer that is self-aware! 😂
Random is neat but where do I type the text prompt? /s
Sometimes I use Freesound.org right inside Drambo to find a sound I want. I’ll search for a crash or snare and then scroll through to find one I like. If there isn’t an Ai sample generator out already there will be soon. I see no reason not to ask it to generate snare drums for me and then dig through to find the one I like. The only difference from Freesound would be that it has an endless library and no sound it doesn’t have. I want to access that site through Drambos sampler browser.
I like to imagine you can abuse the generators a little by given them strange enough prompts. Like a vibrating maraca wrapped in tin foil falling down stairs. Sure I could break out the vibrator and maracas myself but I kinda wanna see what Ai thinks that would sound like.
@rs2000 I mean a randomizer you can instruct, as opposed to keep pressing a dice button till you find inspiration
This always happens in these discussions where people stick around just to keep sharing why they won't use it or why it's trash. It's like saying you won't use Bing search but then sticking around to keep telling everyone why you won't use Bing. Like, just don't use Bing.
@pedro How would you like to instruct it?
Moving forward, I respect that “I will never use AI for making music” stance @wim, but what about ML models for automating simple tasks? Would you rather go back to doing tedious tasks by hand?
What kinds of simple tasks are you thinking of @pedro ? Besides mastering, I can't really think of any simple tasks that I find tedious and would want automated. I'm open to new ideas though.
Maybe it's because my musical background has been on instruments hundreds of times more tedious to learn and play. I mean, I even find some level of pleasure in tuning my guitar and changing strings for Pete's sake. 😂
Who wouldn’t want to have this on iOS? Instabuy at desktop price ($30) for me. Wouldn’t you guys use a generative midi app because it’s cheating?
Cheating? Not at all! I just find that expressing something that came from my own soul is more satisfying. Sure, it's also satisfying to repurpose something made by someone else or an AI into one's own creation. I get that.
Yes, yes, I know I'm borderline running afoul of the developing "only one side of the conversation" sentiment here, but being negative isn't my intent. I'm genuinely interested in counters to my way of thinking. I'm looking specifically for what people find satisfying, since I find it lacking myself.
I only posted this because I was directly tagged. I really don't have more to contribute, but am reading with interest.
Apparently Ai Foley sounds are a thing already. I was just messing with Myedit.online but I used up my free uses. Does anyone know or can test if you can use that site from inside Drambos sample browser?
what the hell did you expect? ;)
a bunch of roll your own ... modular freaks raving about "I have no clue what im doing but I need a piece music now tec" ?
wrong peer group ;)