Would actually listen to these in the radio… I wonder if this is somehow faked, sometimes it sounds like you still haering the reverb of voices at instrumental parts.
the hypotheticals on paper about “artfully moral” use of AI is endless, but the reality is, if there are tools to exploit to take shortcuts to get clout/fame/paid, people will take them, especially in todays culture. you can expect a shit show for sure. i dont care too much about synthfluencers, or “type beats” or any of this other consumption motivated fuckery that goes around these days. i look forward to AI music establishing a well defined boundary between musicians doing art and those making products for consumption. there may come a time where AI can produce some great music with just a few prompts - but it will inevitably lead to the internal question of “why do i listen to this?”
lets say you’re someone with little musical taste or knowledge. you like the “lo fi beats” playlist on spotify. you go find some tutorials on youtube, how to make “lo fi beats”, you mimic it, now you’re making lo fi beats. you get ambitious and do some research on “hacking” the algo to get plays and pay for some spotify playlist positions. you do some viral stuff on instagram and now you’re the “lo fi beats” king/queen. the most followers, the most views, millions of listens on your playlist placement etc.
AI enters the chat.
this is going to be the end of the world for those people i described above. i myself still find tons of original music thats not based on replicated/perfected aesthetic and its what i listen to. i dont know what the “type beat” producers are making or what the hottest trends are. im open to new music but im not going to waste my time listening to random shit that hits the aesthetic. but for those algo/view hustlers, AI is going to thin the herd enough to put it into another category of music (my theory, lol). there will be people that will want to hear the ultimate drop from a combination of the best producers and best techniques etc, and AI will provide them with some mind bending material once the tech gets to that point. then there will be people who like listening to their taste in music and what they think is good for whatever reason.
i dont think anyone who has a developed taste in artist/music is going to experience any impact from an AI revolution unless you’re already joyfully listening to the regurgitated shit that gets views that exists now. one does not simply say “i love nine inch nails” but they’d rather hear the AI version. whatever threat/market there is for AI is not going to make that much of an impact for the people that are legitimately into music imho. the real painful side of the future of AI music is going to be for the ones who play the algo systems and focus more on viralness/influence. they’re gonna get cooked. the rest of the musicians are under no obligation to “adapt” to all this stuff unless their goal is to fit in with the demographic i stated above.
you can get an AI version of tupac rapping with jimi hendrix while they both patch modular and maybe its even 100% on point and sounds amazing, but its never going to be tupac rapping with jimi hendrix while they both patch modular. its kind of a simple point but complicated to explain. whatever AI music brings to the table is going to be used to push consumption. its not going to change the views of a listener who truly appreciates music beyond a well done novelty here and there.
Well, there are different technologies possible using AI technology at different levels. The engineering has gone pretty far with deep learning technology, so deep learning is what current “AI” technology is assumed to be, not like the almost mechanical song composition you’d think of from the past. The deep learning technology relies on having databases of annotated example input and examples of expected outputs of some kinds of “magic black box” neural networks, that are then trained in deep learning processes by various stochastic algorithms that build up the “intelligence” kind of by sophisticated trial and error methods until the results match the expected quality.
This process is kind of quirky and error prone and needs lots of experimentations and is a huge waste of electric energy, because the deep learning training must be done with so much calculation power due to the stochastic (trial and error) methods for optimizing the networks. Then the resulting network would be rather compacted and performant, and is a black box where you can hardly see what is inside and what it would be doing…and the quality of the results greatly depends on how well it was designed and trained, and it is a process that can probably never be 100% perfected, only approached with increasing effort. Think of the early image generation results, with the distorted and twisted faces etc., and then after some years of research people were finally able to make generators giving convincing results, though such mighty AI usually requires a huge amount of effort, testing and revalidation work to make it real and work well enough and not give weird results. You can even deliberately mess up the training or parameters/design of the network to make an AI produce such frightening results, horror movies of the future will be something you need to be psychologically tested before to make sure you’re able to tolerate what you’d see. Somehow, in the training the network gets kind of an imprint of more or less logical abstraction of the data input/output set it was trained with, and can then also create similar results. However it does not truly understand the things it was trained with, it just abstracts them somehow, and is then able to mimick similar things in more or less convincing quality depending on the design and power used to generate the network.
So an AI song generator could work in various ways, depending on what the network would produce. You can very well train it with whole songs and just let the network generate similar audio…then you have all the quality glitches etc. that can happen with a nonoptimal network. Or you can train networks to accomplish more isolated tasks, i.e. generating the lyrics and the singing track, generating the midi notes and instrument choice of all other tracks, then optimizing mixing, mastering, the AI could kind of generate the data for a DAW to produce a song that would no longer suffer from these glitches, or is a mix of AI and human interaction. Like you can train an AI to mimick whole songs, you can also train it to generate the best mixing and mastering settings for a given set of audio tracks, or whatever task you have. You just need enough good annotated example data, and knowledgeable manpower and processing power to create such a tool. Depending on how well it was made, it can then mix and master any tracks well that are similar to the input data, but for radically new styles you would probably have to adapt it and retrain it with new data.
1 Like
Exactly, this is why freedom of speech exists, but at the same time, I feel worried because my gen seems to have a tendency of discouraging it.
I know sharing thoughts in a good faith is always a better idea, but these days, some people these days can go quite extreme if you have different opinions. You don’t like a vst and share a post in Reddit? You got downvoted to hell, passive aggressively. You don’t support all of their opinions, even misinformation? They used any form of labels to dehumanise you and threat you to death, using their “Autobot vs Decepticon” kind of thinking, using their “knowledge” learnt from TikTok to “educate” you. Some kids feel lost and think them as a wolf or a doll rather than a human due to feeling detached from their peers? You can’t help to bring those lost kids back to the humanity because we “must” respect their decision, claimed by a short form video.
At the same time, Generative AI (wanna make this a bit relevant to the thread) seems to be a catalyst to worsen things because it is so easy to create fake information and fool all these people because they are not critically enough to question the validity, but the otherwise to reinforce their belief. I can foresee in the future, these people will use “because ChatGPT say so” into an argument rather than critically analyse a topic, while their thoughts are feed into the model, further firming the mis/disinformation.
Even though I am a technical person who know a bit about software engineering, I don’t really like some of the social media, especially the most popular one. I wish I can to do something for letting them quit the addiction, but I don’t think I can because it is all about self control, sadly.
Funny enough, Renoise forum, SunVox and tracker discord are the platforms which I feel comfortable to open up a bit for sharing my two cents; perhaps there aren’t many kids who love dramas and throwing tantrums since they have no interest what a music tracker is, so the average users are more mature.
Alright, let me go back to the topic.
but the rest is pretty much predefined selection from a huge database.
You are not wrong; in fact, this is how LLM works, pouring a huge amount of data into a model of neural network through multiple iterations, then output the “scrambled” result based on your description that sounds convincing.
There is no creativity or imagination during the generation, but a bunch of complex matrix calculation and threshold detection. (If I am not mistaken, as I have just studied this for a month)
These are indeed some good uses for AI (but technically, they are more about data science than AI itself), like how the last song of Beetle being restored, or how synthplant analyses the sound so that people can also learn from it. These help to rebuild something lost in the past, and drive people to learn more about their field, which I want to see in AI and they are some of the reason why I want to learn it.
However, the majority seems to prefer skipping the progress and getting the result instead, by batch generating some final products, and I could definitely tell that the majority will not fed up with those because they are basically instant rewards without any efforts, just like TikTok or junk food which are known as unhealthy, but they are still using / eating that anyways because of the short term excitement.
These are exactly why I won’t use any AI for my art even after I have learnt it. If we don’t give any meaning to the painting or music, it is really hard to call the product as art because of the lack of stories true off our chest, where is the “soul” came form. Similarly, no one really call a factory mass produced products arts, but decorations.
this is really my point, its really the difference between art and product. i heavily respect your position @Logickin. now dont get me wrong - we see how things work even today with influencers, copied aesthetics etc, i’m certain there will be a whole new era of shitstorm to follow once AI really gets tuned to this, but it just seems so opposite of a path to take for an actual creator working from the soul. for me, at the end of the day, even the most amazing AI work will be a novelty, and some of the most mediocre work from someones soul will connect with me in a way that is personal. on one hand you could make a faulty argument “people thought synthesizers were going to end studio musicians careers” but that didnt quite happen. with the AI thing, i think what really gets to a lot of people, myself included, is just how predatory and detached music has become with all of the influence and culture of “me me me”. the idea that AI would make it easier just spells out doom.
but lets be serious. if AI is capable of doing that, the big dogs with the $$ will be exploiting that so they wont have to rely on individual ideas/merit. if we wanna look at the cup half full, AI could actually cause the bigger players in the game to market solely to sheep/viral stuff and leave a whole world for us making art. if we wanna think of the cup half empty, well, you’re gonna have to do some soul searching on what it is you’re really doing. if the music/art you’re doing is something you NEED to do for yourself, or something you would do regardless, thats not going to change. if you have other ambitions beyond that, then its always going to be a matter of keeping up with the trends/aesthetics to stay relevant. the change/threat of AI really boils down to wether your art is being made for consumption or expression. the rest will sort itself out imho. the benefits of AI in the right hands will provide a lot of opportunity for true creatives and they may or may not use it as a tool. AI in the wrong hands is just going to saturate the existing turds with polished turds and appeal to the same audiences.
as far as learning it - AI is definitely part of our future. i have no hate towards the technology/ideas itself. i’ve just become a bit “anti” towards the current culture and trends, as well as the desire to replicate vs come up with original ideas. i think your way of describing it as “decorations” vs art is spot on.
1 Like
Thankfully this won’t be the death of original music anytime soon, the crowd who actually enjoys making original music will shrink again and maybe become somewhat of an underground novelty.
We’ve already seen that fancy UIs and VSTs just make people that much lazier when it comes to technology, and definitely music (especially when you have a plugin for theory, one for mixing, one for making better drum patterns, etc), so I kind of embrace the mainstream’s idea of continuing to water themselves down. If they genuinely never get sick of it by the time shit like this is just flooding the radio, there’s just no hope for a particular type of person anyway. I mean, someone like that is still listening to the radio 
Does that make AI music ‘good’ or competitive with the human pop algorithm slop which already exists? That’s just a matter of perspective. If you like it, then AI is going to kick ass. If you’ve been sick of it for a long time, this is just more bullshit trickling down the pipe. It’s not change, it’s more of the same.
It’s kind of funny that art used to (at least sometimes) be about rebellion, innovation, saying “fuck you” to people who you know are never going to listen to your music anyway because it’s not the elevator variety that they enjoy. A whole generation of kids who want to make the most easy-listening music is going to feel so batshit crazy to my brain that I’m going to have to readjust – until I remember that soundcloud rap still exists
At this point I’m dead curious to know how exactly this AI thing comes up with a song in just 40 seconds?
I mean is it making it from ground up ? Making instrument lines according to genre or what? Does it involve a sort of music theory ? Like for example it chooses a chord progression then laying a bass pattern , and go on from there… adding effects and finally some vocals which I have to admit it does a briliant job on that …
And still mixing and mastering being even automatically done…
Where does this song structure come from?
I already saw complex songs being made by it.
Thinking like a musician?
Or as I said earlier it just tries to wrap up some already made audio files.
And all this in 40 seconds! Wow!
It depends on the technoligy. But complete deep learning AI solutions are really just a black box, a big crunched up numbers matrix that just needs to be calculated through for a while (40 seconds…) to produce the result. Well…it will directly write to the output sample buffer I guess. That’s what a black box ist - you push some buttons, configure the input, let it run and crunch numbers, and it will just fill a sample buffer with a ready made song. Not all AI solutions work like this, like I’ve written you can also use it to control or mimick traditional tools, but the complete AI solutions are just an artificial brain able to generate text/image/sound/video/whatever directly. Things hallucinate and consecutively build up to the final result on randomly generated variations or feedback of the networks. So this is why some older/less sophisticated AIs have glitchy results, i.e. also glitchy audio, because the networks are not powerful enough. The big players all have very refined networks, and then the results can sound (almost) like real material.
1 Like
Well another thing worth mentioning is there are advanced DAWs like Ableton , Reason , Logic and even Renoise which they’ve been actively developed by proffesional programmers since decades ago and still not have the tools to help us build songs easier like the AI thing.
And suddenly out of nowhere since like a year ago everyone is able to do anything they imagine at once . Is’nt it odd?
Sounds like some Alien invasion has occured:)
Can AI make a song? Yes.
It can generate the text, the melody, create the visual for the jacket, and a video clip.
So basically, yes it can.
Now is it good? I don’t know.
Can AI fill a stadium with thousands of people? Not yet.
Can AI invent a far new music style? Not yet.
Breaking news regarding AI music generation, must watch:
Quite impressive:
1 Like
The one who made the Depeche Mode Track is also around here… In case you are interested
. It switches to English later…
Here’s another crazy generator.
AI will eventually whirlwind through all industries which is a bit mind-blowing to think about.
1 Like
We need to make an AI to control renoise for composing tracks… It’s simple, you just need to record a lot of renoise sessions, you record all keyboard commands and ui and sound as a video. Then you train an AI with this that is able to control renoise like the trainers did when training the training data.
We now need to record a lot of renoise sessions to train the AI. We better hurry, AI can already make what a renoise would make directly, so we need to noise up the renoise obscenely with some noisy boys from the scene, and then people can watch the AI trying to use renoise making a lot of bullshit patterns and obscure native instrument effects routings.
Aaaaand of course there will be a commercial point to this vision, not to say that the boys recording the noise would be sold out, because the renoise AI can just autocomplete your beat, so when you’re lazy you could just load a pattern and a bassdrum and set a one two three four and a hihat, and the noiseboys would make a complete bullshit scene intro song from this, and then you repeat, and then they make another one which is quite a bit like the first but you know, like, unique, and each would immediately let you win you any scene compo if you don’t forget to let it make some ascii art in the patterns and generate intellegible speech just from 303 filter modulations. No bullshit, people will pay for this and compos will be a whole new game. Over are the days of sweat and knuckles, nowadays the renoise demoise are trained in line and in shackles, and their masters will like just cast another tune into the world until nobody wants to listen to that crap any longer.
So on another hand, maybe better let’s protect the human creativity and turn back to straight up procedural algorithms coded in handmade assembly, so we at least still have some control over our machines and creativity. Those keys were made to bang them up alright in line, you know, each time a bassdrum, each time the producer hit a key or did a copy and paste, and then he molded his whole soul and mind state into the whole soundscape. This is why when people listen, there is life in the music, there is the stories of the producers and those listening to it in the music, some real life shit that is alive. One day, in the age of machine music, people would only listen to empty music, and all the life around it, is also empty, hollow, the voices in it all only talk empty lies. I can feel this vibe already today when I turn on the radio and listen to the muzak, there’s no more life in much of that crap, and it’s not even yet AI generated. Must’ve something to do with the wave of AI entrained/controlled music producers, while there wasn’t enough human resources and capabilities where a deal was made to serve all this hollow bullshit.
Now off with this bullshit, wish I could find back some peace of mind and could take my laptop and some power source and a GUITAR or some bullshit and books to read with a backpack and a tent to some green place in nature or maybe where some hippies are, then repelling mosquitos with herbal incenses and tracking beats all around the night time… But then when you play back such tracks on mainstream radio, people couldn’t comply, it’d just shatter their plastic faces and make then buy less artificial augmentations of reality. Who need hippies with chopped up mosquitos stuck in the cooling fan of their laptops nowadays, when we already have the clean shit from the box just with a push of a button? I tell you, in this clean shit, it’s not even alive, it’s just outright all emotions scattered together, it doesn’t uplift you but make you hollow - and within it is all filth and crap, once people realize that they will turn back to making art themselves and shoot anyone who still uses that crap and forces people to get their wits fucked by even just watching and listenting to the results.
AI makes generic music very easily, and so stock musicians might be in danger. For now I don’t think AI can really nail human expressiveness in the same way a real artist or band could. It’s always lacking something and while technically impressive, really doesn’t move me.
I don’t care if AI can write a song or paint a picture but I do care if AI decides who lives, dies, or gets to enjoy the benefits of society. Already AI is being used to make rash decisions about human lives but everywhere you go online it’s people desperately worrying about the artists. The AI conversation so far goes like this: “Oh won’t someone please think of the albums we could release and no one will listen to?!” or “Great now no one will buy my furry erotica!!!”
Maybe off topic but I think we need legal protections for more than intellectual property as it’s one component to what is going to be seen as the next industrial revolution. A lot is going to change in the next 10 years.
Back on topic these tools are excellent for sampling. Really surprised hiphop producers aren’t latching on more. Now you have unlimited exclusive vintage soul samples. Everyone is Moby now! You’re a Moby! I’m a Moby!
Personally I have about 5 albums from the past that did alright and I can’t put them anywhere because I can’t afford to clear the samples. Most are undetectable and short but I don’t risk it. If I were making them today, I’d have AI reinterpolate the samples and release the mambajambas. Suck it Harry Fox.
1 Like
So much wise words. I’m glad to see there is so much reason and common sense here in Renoise community.
I do agree with pretty much everything I’ve read on this thread. There’s however one thing I’ve personally found AI very useful for. I’ve done digital painting since 2010. About a year ago I had a serious moment of frustration with my progress. I tried AI art creator just because I was basically angry at myself. I kept producing images with it for about 6 months until I had enough. I noticed what was going on. So easy, so fast… I felt constantly rewarded without an effort. Not healthy. However I realised what was the true value of the whole thing for me. I studied the “paintings” AI created for me I kept learning more and more what was wrong with my approach in painting. I understood certain things about the painting process in a way I didn’t before. If the work of AI can open some locks in us to move on to next level in whatever it is that we do, then it’s good and useful, but only in that case I’d say.
If AI supplents the popular pop-artist’s, slaves of record companies, who are already making music very robotically without true passion, I don’t think anything signifigant happens. Nothing essentially changes there.
1 Like
I recommend for everyone to read what Nick Cave wrote on this subject.
2 Likes
renoise does in fact have the inputs, that’s where the OSC, LUA and open format XML come in for the translation. That Trimurti is where the “Guru levitating, producing without wires” connects.