Autechre production methods speculation

[xox]

just been listening to pendula hv moda while comparing it to steady gonk one... incredible!

anyone knows anything about algorithmic composition? i cant wrap my head around it  ... well, maybe bc i know shit about programming in general and zero about algorithmic music lol 

anyway...

how the fack they manage to make such musical masterpieces with algorithms?! everything is in its place, it's emotional, engaging, logical, transcendental... afaik with the algorithmic composition you need to know everything about the track in advance before you write an algorithm. amarite? 

cmon! i know we have clever ppl here... show your self! zeffolia? others?

[IOS]

algorithmic composition can be something as simple as

1. choose two notes on the piano
2. toss a coin
3. if it's heads play the first note you've chosen; otherwise, play the second
4. repeat steps 2-3 16 times

It's not necessarily tied to programming on the computer, and you don't need to know everything about the track in advance (quite the contrary really).

Edited June 3, 2019 by IOS

[xox]

^ok, i understand that but pendula hv moda and steady gonk one dont sound that simple and they surely sound deterministic in every way with a clear progression within them

[IOS]

of course, far from sounding that simple! My example was more to guide you to think about algorithmic composition in simple terms, so that you start wrapping your head round it

If you haven't already, it could be worth checking the "feed1 bass drum" thread, I've got a post about (what I believe is) the algorithmic process generating the bass drum pattern.

For the Quaristice tour, they had 127 patterns on the MD; 127 patterns on the MM; around 12 patches on the MPC that was also controlling the Nord G2. Of course, B&B were controlling various parameters live. A single "passage" (like "IO") in that live set could consist of say 8-12 different MD patterns, and the differences between patterns in one such passage could be very subtle, like an extra snare. Ae had presumably (obviously) spent ages composing, rehearsing, refining prior to the live shows; some patterns were composed in such a way so that they had internally a degree of variation e.g., an LFO would control a reverb decay. So when performed live they'd start with pattern A01, play it X number of times whilst jamming live, then move on to A02 etc.

Same thing happens in Max and the tracks you are referring to, only that Max allows you to create your own MD so there's many possibilities.

 

[mcbpete]

2 hours ago, xox said:

^ok, i understand that but pendula hv moda and steady gonk one dont sound that simple and they surely sound deterministic in every way with a clear progression within them

Another visualization is that of the double pendulum - Imagine the below but with one oscillating data value feeding into another (and possibly into another):

 

 

[Bodhidharma]

its not like they press bang and max msp churned out pendulu hv moda, its just shrouded in so much mystique talking about it fuddles it up even more. there could be loads of pre sequenced stuff  on hv moda, ableton shit maybe, idk. 

[flacid]

I don't believe that it is as deep as people (like to) think. 

The complexity of the music doesn't mean that it's necessarily complex to make. 

They're explorers. We get to listen to some of their experiments. 

[xox]

On 6/11/2019 at 12:28 PM, flacid said:

I don't believe that it is as deep as people (like to) think. 

The complexity of the music doesn't mean that it's necessarily complex to make. 

They're explorers. We get to listen to some of their experiments. 

Really?! And I thought it’s blue. 

[Chabraendeky]

On 6/11/2019 at 12:28 PM, flacid said:

I don't believe that it is as deep as people (like to) think. 

The complexity of the music doesn't mean that it's necessarily complex to make. 

They're explorers. We get to listen to some of their experiments. 

Sometimes it's reversed: seemingly simple sounding music is more complex (to make) as it sounds. And somedays it sounds complex and it is. But perhaps not for the same reason.

But "deep" and "musically complex" and "complex to make" are not the same either. And often don't correlate at all.

[Grain Bastard]

I loaded the MachineDrum patches they released from the live stuff and it sounds amazing but nothing more complex than excellent programming and sound design. They really know what they are doing with the synth for sure but breaking it down it’s nothing more than 2 bar 32 step patterns triggering sounds they have created from the onboard synths with automation (p-locks) and some clever choke groups.

I guess the MonoMachine stuff is basically the same principle, so nothing more ‘complex’ than programming step sequencers, triggering sounds they have crafted from the onboard synth engines - it’s not a load of effects or custom software for this era of their output just good old fashioned knowledge of the gear they’re using and a great musical vision.

I would guess at the Max/MSP patches they use are are extensions of the same ideas, so they have programmed their own sequencers (to get greater controls, more ways of manipulating etc)  to trigger their own created synths/samplers that have the level of control they require and the real ‘complexity’ is in there musical vision.  Unless you program/play the max patch as they do, I don’t imagine it would sound any more ‘Autechre-y’ than using the Elektron boxes.

 

just MHO

Edited June 13, 2019 by Grain Bastard

[Nil]

Using TidalCycles has really opened my eyes regarding the way Ae *might* compose. Simple algorithms can go a long, long way.

The adequate combo of a few conditions / variables can lead to infinite variations of a (simple) pattern, variations that you can still keep in control if wanted.
Apply a function to the whole (or another variable in parallel) and you can make things evolve drastically in a second.

With TC you can compose a lush melody in the most traditional way and modify it afterward.

Now add Sean and Rob's talents and skills to the equation...

[xox]

Nah you’re just complicating it to much! If anything is simple in music land it’s autechre

[sweepstakes]

On 6/14/2019 at 4:33 PM, xox said:

Nah you’re just complicating it to much! If anything is simple in music land it’s autechre

If anything I thin what Nil is saying confirms this! My experience with TC so far has been really eye-opening - remove the friction imposed by traditional sequencing interfaces and throw down just a pinch of code and you're off to the races. That is, less complexity, not more - just expressed in code.

[zlemflolia]

On 6/11/2019 at 3:28 AM, flacid said:

I don't believe that it is as deep as people (like to) think. 

The complexity of the music doesn't mean that it's necessarily complex to make. 

They're explorers. We get to listen to some of their experiments. 

I think that they actually do have some pretty advanced and unmatched stuff.  I vaguely remember them talking about using fluid dynamics simulations as a trigger mechanism or envelope generator or something.  This would give a very organic yet completely generative feel

On 6/13/2019 at 9:39 PM, Stickfigger said:

Sometimes i think people want it to be more complex and mystical than it really is 

If it's so simple show me some music that matches it.  I'm not trying to fanboy but I'm legitimately asking here because nothing else is as good.  I'd love if you could post it and be really happy about the musical discovery, but I don't think anyone else makes anything as good as it

Is it mystical?  No.  It has an implementation and it's understood fully by AE since they created it.  But nobody else matches it, either due to a lack of maturity of synth patches and experimentation methods, or a lack of trying.  Theoretically could someone else do it?  Yeah but nobody has

I'd love to be wrong because I want more AE-like artists

[Freak of the week]

2 hours ago, Zeffolia said:

If it's so simple show me some music that matches it.  I'm not trying to fanboy but I'm legitimately asking here because nothing else is as good.  I'd love if you could post it and be really happy about the musical discovery, but I don't think anyone else makes anything as good as it

You should listen to academic electronic. Here are some specific albums:

Xenakis - Electronic Music

Jean-Claude Risset - Mutations

Barry Truax - Digital Soundscapes

Mike Dred / Peter Green - Virtual Farmer

Hecker - Acid In The Style Of David Tudor

IMHO this stuff is more advanced than anything AE have ever made (not including stuff like Bine and Gantz Graf).

 

 

[zlemflolia]

"Hecker - Acid In The Style Of David Tudor"

 

holy shit

[zlemflolia]

On 6/3/2019 at 3:20 AM, xox said:

just been listening to pendula hv moda while comparing it to steady gonk one... incredible!

anyone knows anything about algorithmic composition? i cant wrap my head around it  ... well, maybe bc i know shit about programming in general and zero about algorithmic music lol 

anyway...

how the fack they manage to make such musical masterpieces with algorithms?! everything is in its place, it's emotional, engaging, logical, transcendental... afaik with the algorithmic composition you need to know everything about the track in advance before you write an algorithm. amarite? 

cmon! i know we have clever ppl here... show your self! zeffolia? others?

Some cool things you could do are as follows.  I highly doubt ae do this but it could be used to synthesize abstract electronic music this way in the future

Take a max/msp patch programmed to generate a random sequence of squelches.  Make this synth sequence be the input to a generate sequencer which builds higher and higher levels of abstraction by modifying tempo-modulating structures, the melody sequence, scale, and tuning parameters.  first this machine is programmed randomly.  then the machine will iterate the sequence forwards until it starts to have a 0.1% higher manifestation as enjoyable music at some few second time snippet, and you point out where by drawing an "enjoyment" envelope.  You train this using a recurrent neural network, specifically an LSTM (a technology used at Google to implement Google translate), to generate generative sequences from noise which are mapping the language which are some small percent closer to your tastes after that training epoch.  So you run a human-guided like generative adversarial network translating from noise into your sequence style of choice which were adversarial selected for during the listening period from which you chose good envelopes.  It's a reverse GAN (Generative Adversarial Network) where instead of a generative model sending input to a discriminative model which then ranks it and chooses the shape of the next epoch, instead it is the human playing the role of the discriminative network by acting mathematically as a training data generator as you select the enjoyable sound segments.  Presumably over time it would result in an outward manifestation of that aspect of the individual's musical taste as a real song, if it was implemented properly.  It's as if instead of using an instrument to generate music, you let the neural network use you to generate music from your mind by acting as the discriminative network through your scoring

I actually vaguely remember some article mentioned here where RDJ was commissioning a programmer to generate him software implementing an evolutionary algorithm where he can select the compositions he likes and it migrates towards those.  This is the same thing, except instead of using a genetic algorithm you are using an LSTM recurrent neural network

Here's an article referencing it:

http://www.engadget.com/2014/12/29/aphex-twin-mutation-music-software/

Edited June 22, 2019 by Zeffolia

[zlemflolia]

If Aphex Twin actually uses this evolutionary algorithm for composing, I highly suggest he also hire someone to implement him a sequence2sequence model doing the exact same thing except as a continuous stream.  Instead of generating 6 samples and you pick from one of them, it could send out a continuous stream of music and you merely rate it with an envelope, in fact you could rate it with multiple different envelopes meaning different things at the same time, such as your satisfaction with various individual qualities like tempo, melody, texture, etc.  Surely there are truly abstract envelopes that can be applied to existing technology that would sound extremely strange.  In fact due to the Plancherel theorem (https://en.wikipedia.org/wiki/Plancherel_theorem) which says any continuous function can be approximated by the summation of an arbitrarily long list of harmonics, proves you can.  That means this is definitely do-able.  It all assumes you have a sufficiently configurable synthesizer hooked up to the output of a stream of a finite number of symbols

Edited June 22, 2019 by Zeffolia

[hello spiral]

29 minutes ago, Zeffolia said:

"Hecker - Acid In The Style Of David Tudor"

 

holy shit

Yeah this is my fave Hecker LP

[Freak of the week]

17 minutes ago, Zeffolia said:

Some cool things you could do are as follows.  I highly doubt ae do this but it could be used to synthesize abstract electronic music this way in the future

Take a max/msp patch programmed to generate a random sequence of squelches.  Make this synth sequence be the input to a generate sequencer which builds higher and higher levels of abstraction by modifying tempo-modulating structures, the melody sequence, scale, and tuning parameters.  first this machine is programmed randomly.  then the machine will iterate the sequence forwards until it starts to have a 0.1% higher manifestation as enjoyable music at some few second time snippet, and you point out where by drawing an "enjoyment" envelope.  You train this using a recurrent neural network, specifically an LSTM (a technology used at Google to implement Google translate), to generate generative sequences from noise which are mapping the language which are some small percent closer to your tastes after that training epoch.  So you run a human-guided like generative adversarial network translating from noise into your sequence style of choice which were adversarial selected for during the listening period from which you chose good envelopes.  It's a reverse GAN (Generative Adversarial Network) where instead of a generative model sending input to a discriminative model which then ranks it and chooses the shape of the next epoch, instead it is the human playing the role of the discriminative network by acting mathematically as a training data generator as you select the enjoyable sound segments.  Presumably over time it would result in an outward manifestation of that aspect of the individual's musical taste as a real song, if it was implemented properly.  It's as if instead of using an instrument to generate music, you let the neural network use you to generate music from your mind by acting as the discriminative network through your scoring

I actually vaguely remember some article mentioned here where RDJ was commissioning a programmer to generate him software implementing an evolutionary algorithm where he can select the compositions he likes and it migrates towards those.  This is the same thing, except instead of using a genetic algorithm you are using an LSTM recurrent neural network

Here's an article referencing it:

http://www.engadget.com/2014/12/29/aphex-twin-mutation-music-software/

You may want to check out this one:

http://editionsmego.com/release/EMEGO-241

Neural networks aside, this  "injecting order into randomness" idea that you mention is what personally interests me a lot. Sound synthesis based on this approach was introduced by Xenakis decades ago. In his own words, the idea is  "to start from a disorder concept and then introduce the means that would increase it or reduce it". Basically, what he says is that in order to be able to discern the pitch of a given sound, the waveform of that sound needs to be rich in symmetries and periodicities. So, more regularities - more stable tone (more instrument-like), less regularities - less stable tone (more white noise-like). Of course, as far as listening is concerned, the most interesting regime is probably the intermediate one, between these two extremes. The synthesis itself works by stochastically distorting the previous waveforms in order to create the new ones, with certain constraints imposed. If you are more interested, this synthesis is described in detail in this article: http://sites.music.columbia.edu/cmc/courses/g6610/fall2012/week4/Gendy3.pdf. And this is how it sounds:

Other composers have also done similar things. If you really wanna go deep, I recommend going through this dissertation: https://pearl.plymouth.ac.uk/bitstream/handle/10026.1/2841/2013Valsamakis341008phd.pdf?sequence=1&isAllowed=y

Eduardo Miranda who supervised this dissertation is a very interesting character. He wrote/edited many books on computer music, of which you may want to check out this one: https://www.amazon.com/Evolutionary-Computer-Music-Eduardo-Miranda/dp/1846285992

[Silent Member]

You guys are overthinking this. It's 90% weed, 1% pressing randomise, .5% pressing play and 8.5% editing the results.

 

I can't tell you how I know this for a fact, but I do.

[xox]

4 hours ago, Freak of the week said:

You may want to check out this one:

http://editionsmego.com/release/EMEGO-241

Neural networks aside, this  "injecting order into randomness" idea that you mention is what personally interests me a lot. Sound synthesis based on this approach was introduced by Xenakis decades ago. In his own words, the idea is  "to start from a disorder concept and then introduce the means that would increase it or reduce it". Basically, what he says is that in order to be able to discern the pitch of a given sound, the waveform of that sound needs to be rich in symmetries and periodicities. So, more regularities - more stable tone (more instrument-like), less regularities - less stable tone (more white noise-like). Of course, as far as listening is concerned, the most interesting regime is probably the intermediate one, between these two extremes. The synthesis itself works by stochastically distorting the previous waveforms in order to create the new ones, with certain constraints imposed. If you are more interested, this synthesis is described in detail in this article: http://sites.music.columbia.edu/cmc/courses/g6610/fall2012/week4/Gendy3.pdf. And this is how it sounds:

Other composers have also done similar things. If you really wanna go deep, I recommend going through this dissertation: https://pearl.plymouth.ac.uk/bitstream/handle/10026.1/2841/2013Valsamakis341008phd.pdf?sequence=1&isAllowed=y

Eduardo Miranda who supervised this dissertation is a very interesting character. He wrote/edited many books on computer music, of which you may want to check out this one: https://www.amazon.com/Evolutionary-Computer-Music-Eduardo-Miranda/dp/1846285992

exactly what id call a 'nonsense music' 

like, ''i can't make anything musically deep or interesting, but i want ppl to call me genius, so im gonna go full retard like nobody before me and if they can understand it or if they say it's emotionless im just gonna call them out''

Edited June 22, 2019 by xox

[xox]

@ Zeffolia and others

thnx for sharing your knowledge 

 

ppl continue and especially if you have specific solutions in max pls just bring it on!

Edited June 22, 2019 by xox

[xox]

another thing that i dont understand is why would ae use max for all this algorithmic craziness?! wouldnt it be better/easier/more efficient/more powerful to use something like haskell+sc as pointed out by Nil?

[Nil]

Because they're so used to Max and have been using it for decades ?

From what I've understood, both environnement are quite similar in many, many, many aspects. Besides, if you re-read Ae's RA interview, they mention live-coding as a part of their process.

xox, just give TC a try for a week or two and I'm sure you'll see what I mean(t).

Anyway my previous post was kind of a tribute to their talent and vision more than anything else really.