Autechre on Generative Music Techniques
In 2013 Sean and Rob of Autechre did a massive AMA on the We Are The Music Makers forum. Their answers have also been collected into a Google Spreadsheet.
Here are some highlights particularly about what they had to say about generative practices and software/programming matters.
when you say you make sequencers/synths, what’s your preferred way of development? do you prefer clicking boxes in max or juggling your RAM in C?
i was getting quite into c a couple of years ago and then gen came out so i dropped it
i way prefer using max to coding
Whilst being prolific in terms of bespoke software design for your own music (i.e. Max/MSP patches for Confield, Oversteps, etc. etc.), have you considered designing a manufactured hardware product? It’s been interesting to see what The Black Dog have done with the CS X51.
tbh it’s tricky cos part of the reason we make software is so that we can hack it easily, and save tons of versions. and the way everything integrates is bespoke as well, the protocols etc; that’s such a big part of it that it’s hard to make a one box solution or something that integrates with midi (or other equally lame or ancient protocols) that well.
we do bounce ideas a lot tho. we’d probably use fpaa and fpga pretty heavily if we did anything, so it could be part modular analogue but still have decent patch storage.
How much influence does randomness have on your music if any?
Aleatoric is the word I think
given that neither of us had any say in our genetic code, i would say quite a lot
Can you shed some light on the composition process for Oversteps? It almost seems as though the main melodies/chord progressions were composed traditionally with a generative process used to add flourishes to the existing melodic content—is there any truth behind this? It’s also quite a strange album in your discography since it doesn’t seem to share a similarity to anything before or after (besides Move of Ten, but I consider that and Oversteps to be a package deal); are there any plans to re-visit this sound or build upon it?
most of it’s markov chain stuff done in max, with conditionals forcing scales and harmonies
we still have those sequencers but we developed them a fair amount since, they’re not as trad as they were then
Oversteps really jumped out at me as having to do with counterpoint, the interplay between multiple voices talking to each other. I noticed this particularly in O=0 and redfall.
Is this something that you intended? Is this something you noticed?
yeah oversteps was programmed that way
What do you mean by ‘programmed that way’? Were the voices programmed to directly influence each other in a generative type of way? Were they programmed independently, but in such a way to compliment each other?
yeah the voices were independent but communicating
it’s just conditionals really nothing too fancy
I was listening to Oversteps today and I was wondering - did you have specific patches for those tracks that would generate them in an ongoing way, so you could leave them open as long as you wanted and have the melodies keep varying? Or were the songs set up to go through a certain number of variations / loops and then end? I just imagine it would be so sweet to pop open a patch and have it play a couple hours of pt2ph8 while you’re doing things around the house.
yeah exactly
the original takes are long as fuck. i was doing that most days, followed by loads of editing in the evenings
when making Oversteps, was there even a small element of it being a reaction against those folks who had been saying “they forgot how to make nice melodies”? Or were you just following your muse as per usual?
oversteps - basically we were making algorithms with a view to using them repeatedly, and seeing how flexible we could make them in terms of being able to reuse them in lots of different contexts (making diff styles of music) and it kind of grew out of that
i dunno if we were really successful in achieving our aims (not really, we got very sidetracked and went with it into a different zone) but we liked the result and ended up shaping it further into that kind of oversteps territory
i think we were concentrating on melody, harmony counterpoint and all that cos the rhythm stuff has been second nature for a while, in terms of programming and algos and such, and we thought we might learn more if we came at it from the opposite angle
how did you teach yourself? did you read up on computer programming? dsp reading? have you ever read old books like the allan strange synthesis book, did you have to re-learn Maths to develop your custom patches? Stuff like your aleatoric sound, did you just do it trial and error? Did you have to educate yourself in any specific way to build your sequencers?
i just learn stuff as i go, but patchily and not in terms of high level techniques at all. i did have to learn some new physics a few years ago but usually i just need tiny bits of info to get on with what i’m doing.lots of trial and error as well.
each of your albums have their own unifying 'vibe’ or identity that comes both from the hardware / software used as well as the compositional process:
untilted - densely programmed step sequenced kit.
quaristice - looser & sparser live jams with hardware.
oversteps - evolving generative melody (i’m assuming sequenced via Max?).
i’m curios what the process/vibe was for Exai? what was the hard/software situation that the album grew out of and how were those early experiments refined and evolved to get the 'Exai sound’ ?
the software used for exai grew out of the software we made for oversteps mostly realtime stuff so, long jams edited down
not all of it tho, some of it was worked on more compositionally (the software can do both)“
Is this software that’s part of something else (Max/Msp or whatever) or software that you two’ve created entirely on your own?
we use max as a hub really, with c externals and gen, few bits of hardware
we don’t program in OSX Assembler (would that count?)
You guys are huge on generative music; ever considered releasing an album of puredata patches or something along those lines, rather than prerecorded audio? I recognize the issues with doing so, but it’d probably blow some peoples minds to have that happen.
yeah this comes up every few years
maybe in the future
do you guys still build synths in Max or just use it for sequencing hardware?
we use it for both, and as a hub for connecting other bits of software, and sometimes hardware but increasingly less the line between sequencing and synthesis is pretty much gone now. textures are sequences, sequences are like harmonies. it’s all the same thing when you get down to it.
When programming, do often spend a long time trying to make more general purpose tools with a ride range of uses, or is it usually more ad-hoc? What are some of the more interesting/unexpected design challenges for you when making these tools? I’m particularly curious about how you decided to handle recording/storage and playback of continuous parameter changes, i.e. stuff you control with knobs/faders.
i might start out making a specific thing but then i get distracted by an idea for something else do it becomes the new thing
i don’t really care if i get to finish the thing i started
everything is continuous basically, there is no recording as such, just a stream of data being generated strictly
Over-thought question time: one of the main differences between "early Ae” and “"later Ae”“ is the tension between rhythm and arrhythm - at least that’s one of the main things I get. Obviously "flutter” was consciously in that vein but it wasn’t a major part of the toolkit until maybe cichlisuite / chiastic / envane. At least that’s how it seems to me. Obviously we all like a massive 4/4 boom-bap, and that’s like junk food when you guys do it well (e.g. 6IE.CR, IO, 1 1 is, etc) but my very favourites (e.g. Pencha, Osla for n) seem to scatter beats around like confetti while a steady pulse runs invisibly through it.
Oh yeah, the question. How does that come about? Is it usually programmed? Some kind of time-scatter algorithm that you push right up to the bleeding edge of rhythm v chaos? Micro-editing? Played live and tweaked? The reason I ask is that it’s probably the number one thing I like about Ae and the most difficult to describe.
various ways, sometimes it’s all programmed step by step, sometimes played in, and sometimes defined in advance and then run off in realtime (i guess you could call that algorithmic)
my spellcheck was turning realtime into teatime then
would u plees name one ambient music recording which u fancy?
god that word ambient is such a loaded term (thanks brian) - i’m gonna say phaedra
Are you interested in artificial intelligence in the realistic sense? Do you think machines can become conscious, if not very intelligent? It seems to me we are well on the way to machines learning pattern recognition and becoming smart. Check out “Qualcomm Zeroth”, a neural processor. It actually doesn’t have a von neumann architecture. It learns by feeding it info and it did image recognition without being programmed to do so. I wonder if in a few decades we’ll have convincing robots etc.
ai is a funny thing, i mean, if i was to use the average game programmer’s definition then we’re already using ai in our work, but i would never make that claim outright cos i think describing a series of if statements as ai is a bit reaching and self-congratulatory
but i’m pretty interested in it from some angles, but most of them to do with analysis and mining tbh, and that’s not really the way i like to make things
it would be fun tho if one day we just turned a machine on and it wouldn’t make us a track cos we had ignored it for a fortnight, and just started spitting out really bad trance tracks out of spite
In the past you expressed discomfort with not being in total control with regards to generative music. Have you changed your mind or have your methods/algorithms for making generative music changed?
nah we never said that. iirc i said that there were always details we’d want to edit
the way we were working then was like, make an algorithm for each track, whereas since 2006 we worked on a more general system that we can expand as we go, and those kind of problems are solvable in the long term with that strategy
It’s pretty well-known that you guys use MAX. I was wondering what some of the other languages you guys use are.
I’m a Linux head and tend to fiddle around with PD when I’m not using Renoise.
just c, used to use lisp a bit (kyma and symbolic composer), and as a kid i was all about bbc basic and dabbled with pascal
i don’t even use c much since gen arrived tho
I’ve been using Reaktor for a few years now and find it hard to leave cause it’s got a very nifty randomize 1-100% button. I’m getting tired of being stuck at my computer all day though. There any good beginner piece of hardware that would enable me to get the depth and randomization ability? You guys ever mess around with any Reaktor ensembles?
u ever played with synplant?
it’s well basic but the way it works is good fun, for about 10 minutes
can’t u just get a demo
i dunno, prob not, i used it for about a day and didn’t record anything
but i liked the idea of it
u mean a button that moves all parameters a bit more. 1-100. yeah ok that is diff a bit, but what if your really close to something quite good, and u press that, u may never find what u were after - and be further from it than b4. we call that 'pot’ luck
Yeah exactly. It’s risky. Like u said, I could be on to something. Usually if I can feel something sounds decent I’ll begin the manual tweaking and cleaning.
Synplant looks like a quirky thing. I could mimic that using a more complex patch with hundreds more values and randomize everything in 1% increments.
I use this method mostly when I have a blank slate and am feeling lazy. It can be tough jumping into things cold when there’s no idea of what sounds I want to get into. That being said, I am gearless thus subject to the electronic abyss. Shit can be scary. The elektron looks like fun id like to get my hands on one
more like u should have 2x reaktors , then make the one u like less, become more like the one u like, i guess thats what synplants more about.
Just how “generative” (not sure this is the right word for this) were some of the tracks on Confield? Tracks like Sim Gishel, Parhelic Triangle and Uviol, for example.
I’ve read so many rumors and theories about this, that I don’t know what to believe at this point. I mean, it sure doesn’t sound like you just hit record and then left the computer running for a few minutes, since all the tracks seems to go somewhere. Were you adjusting stuff on the fly or?
parhelic triangle wasn’t algo at all
sim gishel was slightly
uviol was a lot
but none of it’s in the way you prob think of 'algorithmic’ with loads of randomness. it was all planned out, with bits of it being executed after intervals etc
like a track but codified
that’s the thing when u say algorithmic people think you mean unplanned and sprawling. doesn’t mean that at all in our case, just means procedural
Either of you have any interest in Supercollider anymore, or does Smalltalk code do your head it too much? I love Max to death, but often feel deflated when some algorithmic process takes an extraordinary amount of patching in Max, but is only several lines of code in Supercollider.
i never used SC. no good at coding really.
What are your opinions on the use of fractal formulas for generating beats or melodies?
Any particular favorite types for particular applications? (i.e. L-Systems for melodies… Mandelbeats…etc)
i prefer moire effects in music more than fractal - in aesthetic ways anyway - tv transmissions recorded off tv with video etc… fences phasing each other on motorways so on, its a bit more wild but simpler too, and is more like things i find IRL on larger scales. i know theres loads of args to shut me up there, but mbe we o’d d on fractals as a generation.
really like L-systems for graf/ix, tho’
yeah (altho some people prob already know this) i’m a huge fan of simulating fluid dynamics to make things move around in a controllable but beautiful way (using navier-stokes equations)
For generating sounds? Can you also do turbulence?
sort of yeah
it’s a lot easier since gen came out as well
I think that your albums have different character, but for me Confield is special because is based on generative sequences, how did you avoid using random number generators?
i dunno it’s easy enough not to use random in max
not that i’m against it any more (at least in principle - as long as it’s done carefully noise can be really useful), but on confield we weren’t doing that
UnderBOAC has one of my favorite complex beats, did you program a polyrhytm sequencer in max?
no that was all programmed in logic audio