graphgrow feedback delay network

[claude]

Experimenting with GraphGrow (my thingy for exploring graph-directed iterated function systems) sonification using a feedback delay network containing pitchshifting elements:

https://mathr.co.uk/misc/2016-12-13_graphgrow_feedback_delay_network.mkv (60MB)

(screen filmed with camera as I couldn't figure out how to get audio sync working in a pure digital desktop capture)

The topology of the feedback network matches the topology of the fractal generators, and scaling of edges is mapped to pitch shifting amount (longer edges transpose down, shorter edges transpose up).

Haskell and Pure-data source code at:
https://code.mathr.co.uk/graphgrow/tree/HEAD:/graphgrow3

Since recording that video I've tweaked the audio algorithms, making some delay lines shorter, which gives a more pitched / less noisy sound.  If you don't have a Haskell development environment, you can click the randomize button at the top of the Pure-data patch graphgrow3.pd to hear what it might sound like.

(Earlier experiments modelling the fractal shape itself as a resonator weren't successful - trying to model the frequency response (in particular the phase response) of something infinitely long is tricky (if not impossible), and the sounds were pretty painful to listen to.)

[Chillheimer]

as composer I can say that this would be a nice sounddesign tool. the rumble noise at 2:30 sounds very natural!

I have no idea how the environment you are doing this in works, but an interesting approach to make it all a bit more musical woud be:
-when there are tonal elements, add a notch-filter that only lets through the nice sounding harmics/overtones of a base note that is triggered via midi.

If something like this is possible, this might be a first step towards a fractal-synthesis-instrument..

[claude]

The problem is that the tonal elements with pitches are all emergent from the possibly chaotic dynamics of the network.  So it's impossible to control them in a simple way.  What could be done instead is fix a topology (for example a Koch snowflake hybridized with a tree) and perturb the node positions many times, analysing the audio at each set of positions to form a parameter space map, which you could use to guide live performance.  If you were really lucky you might even get a musical scale, which you could play with a keyboard...