|
generative music speculations pt. #fu
Author: astroid on September 26 2008 Viewed 1773 times. 14 people liked this blog. You can rate it below if you haven't already.
People who enjoyed reading this: yghartsyrt, Squeal, delete, sAMsKi, bla, jp, daswesen, utofbu, sohcahtoa, DrexonField, eyesnine, ignatius, fichu, bodo
-->
well, now my 1. g2 is brokeneds and my 2. laptop is similarly brokeresd, so for the time being i'm solely carrying around any shitty reactionary ideas i may have about generative music in my branes, where they may compete for mental real estate with lists of metaphors for fat poodles. "like a hoagie that's been tarred and feathered" "like a sheep with dwarfism".
so anyway yeah i had a feeble idea seed and thought i'd plunk it out here for safekeeping
1. have a target function- that could be a midi note, chord, timbre, or an input
2. try random values (notes, chords, timbres, or whatever) and then progressively store the best ones
yeah that's it
as you were | |
Comments
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
yghartsyrt
sorry to hear about laptop and the nord.
first question, that comes to my mind. who decides abiout the best values? the algorithm? what are the best values? is it just a markov chain, that gets questioned by some random function.
that part is not clear enough for me...
please elaborate?
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
Squeal
genetic algorithmish
but you need to breed the "good" little values or mutate them to keep refining to fit your criteria.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
Squeal
you need some if/thens at least.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
astroid
best values would be closest to the target i think.
i guess it is genetics. i was thinking 'a systematic way to explore similar objects around a target object'.
thinking out loud
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
dach
I want a sheep with dwarfism
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
bla
the target function is the perfect state it wants to be and it keeps picking random things and comparing them to the target and if its closer than the last one then it plays it and picks another , if its further from the target it doesnt play, if its spot on then a new target is chosen?
id love some if/then shit for my midi stuff
im thinking of a loop of midi feedback going through my mep4's which modify the numbers a bit and if it passes the filters then it reaches the synth/drum machine/etc otherwise it goes back round the loop
i want it to be a sequence generator rather than a spazzy chaos retriggering thing like ive done with audio/midi feedback
i want a ton of midi hardware and only a tiny bit of actual audio
midi buffers are quite limited though
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
cbit
"best values would be closest to the target i think."
If the system works well though, you;ll end up with something very similar to the target function, so wheres the added value? maybe im misunderstanding.
I think it was bla (?) who suggested that there could be an audio version of facemaker experiment.. there, humans are the agents of selection, you choose the 'best' of a pair of generated individuals. Directing 'monkeys with typewriters' is a long route though, to get something interesting.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
cebec
i want a dwarf with sheepism
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
bla
cbit- yeah i was thinking about that facemaker thing when i read this blog but really the most fun is when you let the machine make all the decisions
cebec- i almost wrote that
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
Roshi
You really need a cost function (or something that evaluates what is good or bad) for your system to work.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
flies
i figure the value in such an algorithm is in the getting there rather than the destination. it's like an 'explosive' dynamical system: the system is in equilibrium (close to target), you disturb it, it flails about, gradually approaching equlibrium again.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
flies
possible mutation algorithms: (target = set of parameter values)
wait for it: if the current value of the parameter isn't the target, randomly (white noise) mutate it. If the current value of the paramter is the target, don't mutate it. this would have a rather long 'relaxation time'.
keep the better ones: create new values for each parameter randomly. compare the mutant value to the old value, and the new value will be the closer of the two. this would have a much shorter 'relaxation time'
progressively damped oscillation: each parameter varies according to drunken walk or brown noise (or sinus) or something. The amplitude of the oscillation is proportional to the distance to the target, but this distance is averaged over some time so that the oscillations will be damped more or less slowly. This algorithm could operate continuously as opposed to the previous two which are essentially step-wise operations.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
sohcahtoa
Whatsamatter with your G2? USB port?
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
astroid
bla- yeah, something like that first thing you said. sadly, my friend, i think you're gonna end up with a computer somehow-cheap old mac with a midi box and supercollider? you could get a g3 imac for like 15 pounds and then some dirt cheap usb midi thing for about the same prolly.
cbit- the key is inefficiency. you wouldn't want something that would just pass the function. i want something that's trying to hit a target and failing. the target is kind of like a mass, the effort is kind of like gravity, and the evaluation becomes something like a spirograph spiral. computer facemaker would be cool because you can go through so many generations so fast. i'd like a way to kick out the human chair under the noose, and let it go through a thousand generations per second ( with something like a timbre).
roshi- yeah that'll be the trick. i can think of a few which would be appropriate on the g2- counting zero crossings, filtering and measuring the magnitude, and then the ubiquitous if/then midi stuff. (hey g2, did you play this note/set of notes/interval/chord?)
flies- YES thank you. the trick would be making this happen automatically and quickly, for me, so that the things i would throttle would be the initial settings/temporary states (like a hard reset with my own conditions), the quality of the randomness (doesn't actually need to be random at all, could be circular and rule-based), and then the speed of the cycling.
this could be another path to theme manipulation- have the target be a major chord or something, and then put the instance of a theme in, and let it flail about with some rules trying to get closer and closer to playing a major chord with the final note of each instance being one of the notes in the chord. it might take a rather long time to get there.
sohcahtoa- i don't know. no midi, no usb. power works.
09/26/08
+
PM |
QUOTE |
PERMALINK |
REPORT
Squeal
bla said: "the target function is the perfect state it wants to be and it keeps picking random things and comparing them to the target and if its closer than the last one then it plays it and picks another , if its further from the target it doesnt play, if its spot on then a new target is chosen?
id love some if/then shit for my midi stuff
im thinking of a loop of midi feedback going through my mep4's which modify the numbers a bit and if it passes the filters then it reaches the synth/drum machine/etc otherwise it goes back round the loop
i want it to be a sequence generator rather than a spazzy chaos retriggering thing like ive done with audio/midi feedback
i want a ton of midi hardware and only a tiny bit of actual audio
midi buffers are quite limited though"
I think it's time for you to go all software....
Register / login
|
^
EM411 is Copyright 2001-2008 EM411.com
All rights reserved. | Contact | RSS
|
