Autechre approach rhythms

How does Autechre approach beat programming and rhythm?

Hi everyone,

I’d like to understand more about how Autechre builds their beats, particularly the way different rhythmic elements seem to shift around each other while still feeling connected.

Sometimes I hear a relatively steady snare or pulse, while the kicks, percussion and other sounds seem to follow their own patterns. I find it difficult to tell whether I’m hearing different loop lengths, probability, changing timing, or something else entirely.

What techniques would be useful to explore to understand this approach?

  • Polymeters and patterns of different lengths?
  • Probability, conditional triggers and controlled randomness?
  • Generative sequencing or interacting sequencers?
  • Modulation of timing, sample selection and sound parameters?

I’m using Ableton Live and learning Max. I’m interested in understanding the underlying ideas and experimenting with them in my own music.

If anyone has practical examples, patch breakdowns, interviews or tutorials that explain these techniques, I’d love to hear your recommendations. It would also be helpful to distinguish what’s known about Autechre’s actual methods from what people have worked out by listening.

Thanks!

3 Likes

Depends on what era of Autechre you’re referring to, I guess. My view is that they would mostly be interacting with controlled chaos and probability. But I don’t know, that’s a guess.

1 Like

Combination of midi and maxmsp fuckery with probabilities and euclidean rhythm resampling…and other types of fuckery with granular synths and resampling and lfos.

1 Like
1 Like

nouserid’s Pinball sequencer is the customary starting point of deconstuction. If you’re learning Max it’s interesting in it’s own right and worth looking at.

In my opinion there are two things that you can’t extract from all the online bits - constraint and taste, and those are really the core of what Autechre’s about.

Any asshole can download a thing and make a bunch of “generative” music. There are countless unlistenable hours of the shit on the internet where it’s just ratcheting black-page percussion with some granular pads. It’s stupid easy to make that stuff nowadays, and without intervention about as artistic as pressing play on an ipod in 2005.

Those tools aren’t technically distant from how Autechre creates the basis for their tracks, except they spend an inordinate amount of time constraining how the probabilities interact and tailoring outcomes to be something they want and to be contextually interesting. The randomness isn’t nearly as random as people think.

Then they take the quality bits and stitch them up. It’s about being able to recognize when a thing is good, how to get at it, how to recreate it, how to use those things to drive a track.

All that said, you could probably just ask them. Sean is pretty responsive on social media.

6 Likes

Thanks for the detailed reply. Your point about constraint and taste makes a lot of sense. I enjoy experimenting with randomness, but I’d like to understand how to shape it into rhythms that feel coherent while still being unpredictable.

I tend to get bored with standard rhythms quite quickly, which is part of what draws me to Autechre’s approach. I’m also really curious whether anyone here has managed to create these kinds of evolving rhythms using only Ableton Live’s stock devices. If so, I’d love to hear how you approach it.

When you talk about constraining how probabilities interact, could you give a practical example? Would that mean allowing certain hits to move while keeping others fixed, or making one event influence what can happen next?

I’m learning Max, so I’ll definitely check out nouserid’s Pinball sequencer. I’m interested in understanding the principles and developing my own approach, rather than copying their sound.

2 Likes

The core of the systems you’re looking at is a Markov Chain matrix. If you’re not familiar, that just means a “memoryless” system that doesn’t care what’s happened before, just what’s happening right now (the formal mathematical definition is a bit more complicated, but you don’t really need it to create one).

Given the state of the system (lets say a sample just triggered), what can happen? There’s a 30% chance of silence, a 25% chance of a different sample, a 10% chance of the same sample, a 5% chance of a tempo change, etc. Those things then change the state of the system which have their own set of weights and triggers but don’t depend on what came before. It’s a big matrix of if/then statements where the conditionals are things that have or haven’t just happened.

There is, of course, nothing preventing the next thing triggered being an entirely different Markov chain… and you see where it can get complicated. Also, those numbers are arbitrary, but they’re the thing that matters. You’ll spend a couple weeks building out a systems and months or years tuning those numbers to create exactly what you want.

The size of the system is really up to you - are you just triggering drum samples? Other samples? Outputting MIDI to synths or controlling other sequencers? Are their distinct probabilities for velocity and microtimings and pitch and rests and S&H and everything else? Do they interact or determine each other? Do they feedback into each other?

In my experience, one of two things happen - either it’s incredibly boring or gets completely out of hand in about 10 seconds. And that’s where taste and constraints come in. How many samples are being triggered? Do they all work well together? Did you take the time to make sure all your bass hits are complementary so if they fire off in rapid succession they don’t sound terrible? Do you have a way to keep them from firing off in rapid succession? Those are matters of taste that you probably want to constrain - “if a sample from folder A is triggered, samples from folder B cannot be triggered until at least a measure has passed” and things like that. Maybe you have vocal samples but don’t want them just playing all the time or over each other, so you tie the probability of them triggering to complementary events, or exclude them if others have just played.

As a concrete example,you could certainly do what you’re describing in Autechre’s music - set a snare to 100% probability on the 1 and 3, no drift, no velocity changes, you’ve basically made a basic drum machine. Then everything else revolves around that because the snare is the probability driver for the kick and high hat and weird sound #3 and whatever, and those don’t have 100% probability of velocity or timing or even sample and shape, so they’ll drift in and out and around that solid snare element.

3 Likes

Nail on the head.

When I was much younger, I listened to a lot of experimental music. In those days before DAWs and general computer use, a lot was going on with tape manipulation and random fuckery with all kinds of sound sources. For an outsider, I was listening to noise that didn’t make sense and they would just put everything in the same (garbage) bag.

What made/makes an artist or band stand out for me was/is the actual musicality that they generated out of the chaos. A great example is Brume aka Christian Renou who uses (he still creates music!) anything he puts his hands on, à la early Neubauten, for example.

Another favorite of mine is S.B.O.T.H.I. - musicality borne out of pure noise sources.

So I guess what I’m trying to say is that randomness is good, as long as it is controlled by someone who understands musicality, has talent, and hears music differently. Not many are able to pull that off :laughing:

2 Likes

Plugin-based modular or livecoding, good buddy :grinning_face:

Orca was just mentioned in another thread, and that can be a really fun place to start. Syncs to the DAW like glue, too.

2 Likes

I didn’t address this and feel like I probably should. The answer is a resounding ‘no’. Live is a lot of things, and in my opinion does some of it very well, but it just doesn’t have the sort of built in logic to handle things like this. The best you’re going to get is some basic probability triggers in the stock devices, but no way to cascade them into each other or trigger between them the way you can programmatically.

Visual languages like Max and Pd, and more traditional programmatic environments like SuperCollider tend to handle it very well, and the upside is it’s pretty trivial to feed data into Live for capture from those sources.

3 Likes

Multiple swing dividers in series. The beat will feel really wonk, but it’s always locked on to the clock, so you can find all sorts of happiness amongst the inbetween.

2 Likes

Could you walk me through one concrete example of two swing dividers in series: the input clock, each divider’s settings, and the resulting trigger timings over one bar? I’m also missing the connection to my original question: where does probability enter this setup, and what rule constrains how those probabilities interact? As described, it sounds deterministic—what am I missing?

1 Like

I’m not going to speak for Manton as he’s got his own ideas and I’ll probably learn something by hearing how he does things.

My own take, if we’re talking about Autechre-style “slippery rhythms”, I agree that timing/swing is definitely a thing to look at. I know there’s a whole Groove system in Live, but I never gelled with the process of extracting and applying agr files. It’s useful in some situations but always seemed pretty limited as an extensible creative tool.

What you absolutely should be looking at is phasor-driven timing in Max. phasor~, what~ and subdiv~ are a great basis for very exacting control over timing. I don’t know if I’ve ever run one into another since doing it from scratch in Max means you don’t really need to (you can just branch/copy the output), but it does allow you to do some pretty crazy stuff by modulating the phasor itself, which creates all sorts of “swing”.

The brilliance here is twofold - first, the phasor is clocked to the transport tempo, which means you can drive it externally, internally, whatever, but there’s always a source clock to refer back to and base other elements off of (like if you want your synth line to actually be in normal time). Second, you can abuse the shit out of that phasor in any number of ways - do some simple calculus to change it from a ramp to a curve (steadily speeding or slowing over the period), drive the modulation with sample and hold noise, n-multisilders into a list object, or a collection of external data changing the curve/slope of your phasor.

Then you can modulate your subdiv~ and what~ crossing points and it’s basically random noise at that point so you delete everything and start over… :stuck_out_tongue: But you can maybe see where you can fit in probability for things like whether to grab another S&H datapoint to change the swing curve or tempo so you get this endlessly morphing timing.

And that’s really just scratching the surface. Think different phasor curves stored in pattr objects that the Markov chain interpolates between. Or a collection that changes timing/tempo/instruments/timbre all at once. Way down deep at the bottom, literally everything is Max is just a number, just data, and it’s data you can store and edit and recall, so there’s not much you can’t do and the ability to have one thing influence another is just a matter of sorting the numbers out and keeping track of it all.

2 Likes

Thanks for taking the time to explain this in such detail. I’ve only just started learning Max, so I haven’t explored much yet, and many of these objects and concepts are still new to me.

Could someone translate these ideas into an Ableton Live workflow using stock devices? For example, how would you approach changing the timing curve, using probability to decide when it changes, and keeping everything tied to a shared clock? I’m interested in which parts can be approximated in Live and how you’d set them up. A written explanation would be great—I’m not asking for project files.

1 Like

To my knowledge, you can’t, at least not in the same ways with the same level of control.

You can automate the song tempo, and even draw automation curves to get a “swing” feel, but as far as I’m aware there’s no way to trigger those changes programmatically or based on chance - you draw them in and that’s it.

As far as I know stock Live doesn’t have much in the way of general probability built in. Session view has Follow Actions and you can do velocity ranges and note on/off chance in MIDI. Maybe there’s more that I’m not aware of.

Honestly, I use Live as a glorified tape recorder. It’s where audio gets recorded, processed and mixed. There are whole swaths of Live’s toolset that I’m not familiar with (most of the advanced MIDI stuff) because I do those things in other programs and pipe it into Live. Could be it has capabilities I’m not aware of, but a cursory search of the docs doesn’t bring much up.

To be fair, DAWs in general don’t do what we’re talking about. It’s not something the general music-making public wants or needs, it’s incredibly complicated to make work under the hood, and most people that want it end up in a more generalized system like Max, Pd or SC and build it themselves.

2 Likes

Thanks, that helps put things into perspective. If I want to start learning Max, where would you recommend I begin?

I understand that I won’t be building anything like a Marko Prša or Autechre patch straight away. But I’d like to start experimenting with some of the underlying ideas they use—things like probability, changing timing, and interconnected sequencers—so I can gradually develop my own approach.

What would be a good learning path for a complete beginner with those interests? Are there any tutorials or small, manageable patch examples you’d recommend as a starting point?

1 Like

I’d start with the official tutorials. Cycling has done amazing work making their product well documented and approachable despite the complexity. Everything has practical and interactive examples to work with and is explained in a straightforward way. They’ll ease you into the ecosystem.

Start with the basic Max tutorials to understand the layout and common objects. Everything through Numerical User Interfaces. You can skip the drawing and movie stuff if you want, come back to it later if it’s relevant. Encapsulation, Remote Messaging and Abstractions are very important down the line as organizational tools but you can get by without them early on.

Those first few sections will get you familiar with Max in general - making objects, hooking things up, data flow etc. I would do the MIDI section while you’re there.

Finish out the Max tutorials or skip them for now, move on to the MSP tutorials. This is all audio rate things - synthesis, samples, DSP, etc. None of this is strictly about what you’re asking, but for me it’s where Max’s power is and the parts I gravitate towards. At the very least you’ll be able to set up simple noise makers so you have some output to listen to.

What you’re specifically looking for is Audio-rate sequencing. That in conjunction with everything else on the list should give you a starting point for what we’re talking about (timing manipulation, ramps, probability, and so on).

It’s worth pointing out because it’s not immediately obvious - even if you have zero interest in video, the JIT environment and objects can be useful. They’re built to handle large amounts of vectorized data and can be useful in storing and processing large list of numbers (like FFTs of audio data). It’s fair to make it last on the list, but a passing familiarity with JIT can be handy if you’re really into Max data.

Beyond the tutorials, the gold standard for deep Max learning is the book Generating Sounds and Organizing Time (usually referred to as GO in Max circles). It is written as straightforward as the Cycling tutorials, but the subject matter is much more in depth and complex, so it’s harder to approach in some sections. But it’s good. So, so good. Explicitly, Chapters 2 and 3 are a very comprehensive look at what we’re talking about regarding “Autechre-style rhythms”.

Past that, you just branch out in the usual ways - Max Facebook groups are surprisingly active and people share patches. The Cycling Discord and forums have years of knowledge and people to answer questions as well as patches to dissect. There are some youtube channels that do Max related content, though I’m not much of a video watcher so I’m not terribly familiar with them.

Lastly, some practical advice. Read the section on Abstractions. Create one that has an dac with your preferred output, a limiter (limi~), and a volume slider (live.gain~) set to halfway. Use this in all your patches while you’re learning, and feel confident to experiment. The limiter and volume slider means you won’t blow your ears or speakers, so you can hook up anything to anything without fear. If you find yourself wondering “what happens if I do this?”, try it and find out.

2 Likes

Thank you so much for taking the time to write such a detailed and thoughtful reply. As someone who’s only just starting with Max, having a clear starting point and an order to work through makes a big difference.

I’ll start with the official tutorials and build towards audio-rate sequencing. Your explanation helps me see how the basics connect to the techniques I’m interested in. I’ll also keep Generating Sounds and Organizing Time in mind for when I’ve got a better grasp of the fundamentals.

Thanks again for sharing your knowledge—and for the practical advice about setting up an output abstraction. There’s a lot to explore, and this gives me a clear direction!

1 Like

Can anyone explain what’s happening here from a rhythmic and structural perspective?

I’m curious about the techniques being used to create these shifting rhythmic patterns. Is this based on Euclidean sequencing, or are we hearing loops of different lengths gradually drifting in and out of alignment?

Could it also involve an unconventional way of subdividing the rhythmic grid, or is there something else going on entirely?

I’d really like to understand the underlying principles and how this kind of rhythmic structure is constructed.

1 Like

It sounds like a sampler triggering different loops in a semi random sequence layered up against each other…

But also consider occams razor…

Taking different loops and layering the audio clips in an arrangement by hand

That’s what I did with this

No coding was involved for this…I just ran different loops through a granular synth and made some noises based off of fm synth presets and spliced, layered and reversed all the different samples together by hand and eardrum.

It is possible…but it does take trial and error and constant listening to fine tune things.

1 Like