Algorithmic composition that doesn't suck

[vkxwz]

I've been thinking more about generative processes for composition, but have been very unimpressed with most of what I have heard, most of it is soulless, aimless, and generally not very interesting. However Autechre seem to have solved these issues completely, so I have attempted to figure out how this is possible:

There are 3 main issues I see. Firstly, randomness; in order to add variation a lot of people introduce randomness, this results in music that at least has variation and can morph over time, but there is no good reason for any of these changes, so its meaningless and uninteresting. The second issue is that most of the generation I've seen is only loosely tied to what happened before; people create large systems of routed LFOs and midi tools that spit out sounds, but it seems rare that the logic to create the sound for the current time step is actually reacting directly to the output from the recent time steps in the past, which is what we humans do to create music. The third issue is human input; the systems with no human input seem particularly uninteresting, but incorporating that input seems like a big challenge, if it's in real time then the user must effectively learn to play their system like an instrument, so traditional sequencing should be an option, but this seems rare too.

The solution to these problems is to effectively create an automaton, which takes input from the past, and uses a set of rules to generate the current time step. The user input should fill the place of the starting state, and a sequencer should be included so that the user can micro edit the substrate that the automaton reads from and writes to.

The next question is what exactly the substrate should be, my proposal would be in a continuous timeline which can be filled with abstract "event" objects. For example a drum hit would be an event, as would a synth note, a parameter change, etc. With a system like this you should be able to construct deterministic worlds by setting up the rules of the automaton, and then effectively create stories within this world by setting the starting conditions and adding specific events along the way.

So my questions are, has anyone here experimented with this approach? If so what were the results? Are there examples of this that I can find online that have been explained / open sourced? What software would best be used to construct such a system?

[psn]

This sounds surprisingly coherent:

 

[vkxwz]

50 minutes ago, psn said:

This sounds surprisingly coherent:

 

Yeah it is, and its a pretty cool project but imo its not very interesting to listen to after a few minutes. From the videos explaining how it works it seems to run without human input and has a lot of randomness, I'm talking about an approach with no randomness

[auxien]

7 hours ago, vkxwz said:

So my questions are, has anyone here experimented with this approach? If so what were the results? 

yeah, to some extent. results vary. like with any new venture of musical style, composition, genre, instrument, etc., early results will suck. often they will be pure trash. you deal with this by either abandoning the approach/etc., or iterating until you learn to carve out the statue from the piece of rock without destroying the medium (and hopefully along the way learning to use the inherent properties of that medium as a unique aspect of the final piece). the key, key, key part of that aspect is human interaction, direction, and definition at almost every step in the process. 

you could craft software to emulate similar choices made by humans based on past examples....but i'm not sure that this software would be any good at making new choices in similar ways to a human...because ultimately humans aren't exactly good at that either (look at bad albums by great artists). to craft software to just spit out music like humans that is also good and interesting in the way that humans are is a flawed premise...it would work at times, but curation/editing by humans would be necessary most or all of the time, and so ultimately isn't this just....human-crafted music? the audience is just as important as the performer, which seems to be almost never considered when i see this shit discussed. algorithms and generative sequencing trees are only just tools for humans to enjoy music, because the machines that make them can't enjoy it. (tangented here a bit, sorry, but i think it's relevant)

7 hours ago, vkxwz said:

Are there examples of this that I can find online that have been explained / open sourced?

idk, i'd say almost any live performance by a musician is to some extent imbued with some randomness/'algorithmic generation' and so i honestly think almost everything of your premise and assumptions outlined are incorrect, or at least approached from an unhelpful direction. 

i know what you're getting at is specifically regarding using computer tools to create algorithmic compositions, and i agree that a large portion of 'algo' stuff is boring/bland/soulless, but i'd say the same about almost every other genre or compositional classification system you could define. the difference is in the how a boring country song is boring is different than how a boring algorave song is boring. both still suck.

also, you defined at least one example in your first paragraph (Autechre). but idk that they'd define their compositions as 'generative'....i'm pretty sure Sean talked about this in depth during one or more of the Twitch AMAs, and i believe it was also discussed back in the AAA to some extent.

7 hours ago, vkxwz said:

What software would best be used to construct such a system?

'best' is defined by what works for you specifically, what you enjoy or gravitate to, or that you can operationally handle well. i would first say software isn't the only possibility, but it is by and large of course....so what software do you use to create music? 

[mlon]

Synalegg sounds pretty dope to me.

[thawkins]

15 hours ago, vkxwz said:

I've been thinking more about generative processes for composition, but have been very unimpressed with most of what I have heard, most of it is soulless, aimless, and generally not very interesting. However Autechre seem to have solved these issues completely, so I have attempted to figure out how this is possible:

There are 3 main issues I see. Firstly, randomness; in order to add variation a lot of people introduce randomness, this results in music that at least has variation and can morph over time, but there is no good reason for any of these changes, so its meaningless and uninteresting. The second issue is that most of the generation I've seen is only loosely tied to what happened before; people create large systems of routed LFOs and midi tools that spit out sounds, but it seems rare that the logic to create the sound for the current time step is actually reacting directly to the output from the recent time steps in the past, which is what we humans do to create music. The third issue is human input; the systems with no human input seem particularly uninteresting, but incorporating that input seems like a big challenge, if it's in real time then the user must effectively learn to play their system like an instrument, so traditional sequencing should be an option, but this seems rare too.

The solution to these problems is to effectively create an automaton, which takes input from the past, and uses a set of rules to generate the current time step. The user input should fill the place of the starting state, and a sequencer should be included so that the user can micro edit the substrate that the automaton reads from and writes to.

The next question is what exactly the substrate should be, my proposal would be in a continuous timeline which can be filled with abstract "event" objects. For example a drum hit would be an event, as would a synth note, a parameter change, etc. With a system like this you should be able to construct deterministic worlds by setting up the rules of the automaton, and then effectively create stories within this world by setting the starting conditions and adding specific events along the way.

So my questions are, has anyone here experimented with this approach? If so what were the results? Are there examples of this that I can find online that have been explained / open sourced? What software would best be used to construct such a system?

I think first of all it's safe to assume that every example of generative music you have heard is heavily curated-mixed-edited. By that I mean there is a human who selects the sounds, the balance and a bunch of other things, like whether the result sucks ass or is it good to post online.

This means that really no example we can find online isn't really pure unadultered generative, but it's probably true that the generative composition process played a big part in creating the piece.

 

The second thing that comes to mind is: what does it mean when you say "meaningless and uninteresting variations"? I am thinking here that a lot of atonal-inharmonic music relies heavily on the ostinato concept, i.e. even the weirdest thing starts being catchy if you repeat it enough times. This is kind of at odds with the concept of meaningless-uninteresting variations, because you can't really predict beforehand what variations will suck ass and what variations will trigger the ostinato/pattern-matching part of the human brain.

And personally I have had this happen many times in my generative experiments - record a randomized 4-bar MIDI loop with the most basic piano sound, let it play for a while and you kind of start hearing patterns and melodies going on, even though it's a simple one shot random thing.

 

But I think what I did with generative kind of makes sense with your idea. I was not smart enough (or had time) to figure out some system that feeds on itself, so I really simplified my approach. I figured that if I fade one repeating random loop into another slowly enough, then it would be a fluid change and I could trick the listener to believe that hey this one melody just morphed into another. I also figured that if I separated the rhythm-pitch parts of a MIDI loop I could replace the rhythm but keep the melody/pitch progression the same or vice versa: replace the melody but keep the rhythm (spoiler: I didn't end up doing that trick much). Finally, I picked a set of pitches for each element (bass, mids, high, slow ambient) so that a) they would be in the same scale and b) each would be in their own frequency region and the elements wouldn't fight with each other for space.

Anyway you can hear the result in my signature. I will be honest though - the raw output of that generative algorithm needed a lot of tweaking to get to the end result. It was a really fun experiment though, and I am planning to try and work more like this in the future.

[cern]

 See my other topic and read what other members have to say here: Linear vs Non Linear  many posts there are very interesting. 

The problem with algorithmic composition that follows a non linear way of making music is that is had to become more randomness to the pieces.
If it's not following a timeline and a special note structure and so on it has to become more experimental and random. 
Therefor it has to focus more on the sound sculpturing and that part is still tricky. Make the sounds flow with each other, make fades, fx etc..
 

Edited October 9, 2022 by cern

[TubularCorporation]

Like 90% of Iannis Xenakis' music.

[flacid]

Unfortunately a large % of people interested in 'generative' lack the creativity, patience or brains to use it in a musically interesting way. 

The best combinations contain live human interaction, the 'correct' generative variables with pre planned composition or editing the jam afterwards.

[thawkins]

Basically I think for a generative piece to "work" well, you need to tune both the sound/mix aspects and the generative process for that particular piece. You can probably have a general approach on how you set up a generative process, but it's not going to be possible to just copy/paste the generative algorithm to a new sound/mix and just press play and have it not suck ass.

[auxien]

1 hour ago, thawkins said:

Basically I think for a generative piece to "work" well, you need to tune both the sound/mix aspects and the generative process for that particular piece. You can probably have a general approach on how you set up a generative process, but it's not going to be possible to just copy/paste the generative algorithm to a new sound/mix and just press play and have it not suck ass.

idk man, Aleksi just keeps pumping out hours of jams that all sound the same from batch to batch and people just keep throwing money at it. i think 98% of it sucks but obv others don’t so…

but is he actually using generative processes on any of it? i’d just assumed so since it all sounds very samey but i don’t know….i’m sure someone here does

[o00o]

48 minutes ago, auxien said:

idk man, Aleksi just keeps pumping out hours of jams that all sound the same from batch to batch and people just keep throwing money at it. i think 98% of it sucks but obv others don’t so…

but is he actually using generative processes on any of it? i’d just assumed so since it all sounds very samey but i don’t know….i’m sure someone here does

he basically does his own version of lofi beats which you also can create without end once you got the formula nailed. 

I worked a lot with Algorithmic composition and its basically like sampling. You have to find a tool that does what you want. then you have to find a way that it creates what you want and then you have to take a lot of care and time to select stuff you like to create a decent amount of output so you can select at least some stuff thats working. In addition you always want to make it better and find new things so you never stop with finetuning and researching better generators that have to be as flexible and as customizable as possible. 

if they work well you still have to figure out great chord progressions and tunings which do not fall from the sky. also generators are never a one click here goes the music thing. they are always a more intelligent midi editor as this is not very intelligently solved in most commonly used DAWs and the way you put in notes in your daw is at a stoneage level in 90% of the DAWs if you think about it

 

Edited October 10, 2022 by o00o

[EXTRASUPER81]

Fwiw William Fields' work in this vein is very good imo, as is Topdown Dialectic. The actual method and nature (wrt how much is generative Vs human controlled in the course of composition) is a mystery to me though.

Of course Sean Booth has talked a fair amount about the use of Artov Chains to generate melodies/rhythms (almost inseparable terms in the case of autechre). But I'm sure everyone chatting here knows way more about this sort of thing than me anyway.

[cern]

This shiet is all about Autechre and you know that too 

When you start taking that algoritm path you gonna be in that "Wannabe Autechre- category" and you all know that. 

[thawkins]

2 hours ago, EXTRASUPER81 said:

Artov Chains

Wait, is that just a Markov Chain in a trenchcoat and mustache? Just goes to show that IDM may be intelligent dance music, but its fans don't seem very smart at all.

[d-a-m-o]

9 minutes ago, thawkins said:

Wait, is that just a Markov Chain in a trenchcoat and mustache? Just goes to show that IDM may be intelligent dance music, but its fans don't seem very smart at all.

 

[EXTRASUPER81]

1 hour ago, thawkins said:

Wait, is that just a Markov Chain in a trenchcoat and mustache? Just goes to show that IDM may be intelligent dance music, but its fans don't seem very smart at all.

Derp

Edited October 10, 2022 by EXTRASUPER81

[EXTRASUPER81]

1 hour ago, thawkins said:

 

And yeah, I'm dumb as a rock.

But I do try to treat other people online with the courtesy I would afford them IRL.

[thawkins]

13 hours ago, EXTRASUPER81 said:

And yeah, I'm dumb as a rock.

But I do try to treat other people online with the courtesy I would afford them IRL.

Sorry this was not aimed as an insult to you at all.

I think people who use the acronym IDM un-ironically (instead of EDM) are kind of low-key full of themselves - the connotation being that "IDM is for smart people, if you don't get it then it means your dumb". IMO the point is to enjoy the music, it's not important at all to be able to play galaxy brain 4D chess while rotating cubes in your mind in order to truly uNdErStAnD it.

I do think it's funny that often you can have EDM artists do a word play on some fancy sounding technical terminology and it's like yeah congrats guys you got to page 10 of the Calculus 101 textbook in your quest to sound deep. It kind of cheapens the art itself too, but I try to ignore it since in the end it should be all about the music.

[Nil]

Using algorithmic and generative processes for the only sake of it will just result in bland, utterly boring music. IMO what sets Ae, William Fields, Synalegg and a few other ones head and shoulders above most is that they use their computers (and what computers have to offer) to voice their creativity, to express something... tl;dr : to make actual bits of music. They are fantastic curators of their own arts too.

I'm super lucky to have ongoing discussions with Synalegg and WF : William keeps on refining his instrument - FieldsOS - and practices it as much as possible, just like a guitarist would practice the guitar. Synalegg has his own music system he can use in many different ways and an overall (fascinating) approach he keeps on questioning.
No matter their set-up though, they'd still make fantastic tunes, no matter what they use or how they make it : they create music first and foremost. And they do it according to their own, personal vision. They just happen to use and master algo/generative techniques. Same could be (obviously) said about Ae.

I'll paraphrase WF (from memory) : thanks to his system, he never starts from a blank page, and I think it's crucial here. And his system has the ability to generate countless starting points by pressing a button on his Lemur UX. He stores ideas he might re-use later, disregard the rest and then starts performing his instrument / recording whenever he feels inspired. You can't remove the performance from the equation. 

As I'm humbly patching my own tools in similar fashion, I find it especially challenging and compelling to channel and interact with what the computer can generate into something musical and personal. To easily isolate, replicate, alter... the generated data, to shape these into musically relevant ideas. I'm happy I'm (slowly) getting there.

Oh and you don't need anything nerdy to compose algorithmically : just set compositional rules and write your score accordingly. Not sure it will sound good as is though.

[thawkins]

I second what Nil said.

A bit off topic but there's also the legal aspect: remember some folks tried to copyright all the chords or something. Here https://hackaday.com/2020/03/05/brute-forced-copyrighting-liberating-all-the-melodies/

This is a kind of generative music. The catch though is that I recall some court ruling - about AI art - that said this kind of stuff is not copyrightable. Turns out it was the US Copyright Office that said only people can be authors of things.

What I conclude here is that if you go 100% generative and admit that you did not do any creative work with your music, then whoops you can't copyright it. So there always has to be a human element of putting creativity or some sort of work to get the result. Basically music a from a fully perfect generative music tool can not be copyrighted. Might be important to people who want to make money from their stuff.

[IOS]

but.. but Autechre have perfected it to the point where it sounds like it's not even generative anymore ugh

[Brisbot]

So far generation to me is only good for giving you ideas. Even that Djent thing, even if it's coherent ideas 24/7, you still have to wade thru it for usable ideas. Then you are gonna want to change it up. So it's like having a band member improv some stuff, then you come in and reply to that with your ideas. I remember listening to a number of songs and was impressed by like one 20 second stretch. But the rest was just whatever.

I have never heard anything impressive for more than a few seconds at a time from music Ai yet. IT's because of the coherence of ideas. It'll come up with something cool, then quickly just go back to basic improv feel.

Edited October 13, 2022 by Brisbot

[thawkins]

On 10/11/2022 at 10:24 PM, IOS said:

but.. but Autechre have perfected it to the point where it sounds like it's not even generative anymore ugh

As a person who has never gotten into listening Autechre, I will interpret this post in a specific way.

[thawkins]

8 hours ago, Brisbot said:

I have never heard anything impressive for more than a few seconds at a time from music Ai yet. IT's because of the coherence of ideas. It'll come up with something cool, then quickly just go back to basic improv feel.

Even if it's not impressive, would you say that AI music is better or worse than some random beginner who posts crude jams on the internet? I don't mean to disparage people just starting out making and sharing their music: it's an interesting point of comparison because hey now we have an AI that can make music, even though if it is bad/unlistenable (which is anyway a subjective thing).

And you know, generative music does not have to be super technical - a LFO hooked to a filter cutoff is also generative and I bet people could listen to that for hours, given an interesting sound.