I am primarily interested in exploring the use of adaptive systems as mechanisms for live performance and installation. Aesthetically, I am interested, not so much in capturing existing styles of music, but more in exploring other 'possible' musics (in an Alife sense). The development of these systems can be practically and conceptually seperated into seperate areas:
:: interactive music systems [2004 ...]
: exploring different ways of utilising the responsive nature for aesthetic interactions.
: mapping the output of formal systems into sound
: finding systems that produce dynamics that are useful musically when mapped into sound
[click on pictures for more]
Self karaoke pond
Ashby's Grandmother's footsteps
Musicians or artists using mathematical models for artistic means generally assume that the dynamics or structures which have formal appeal also have aesthetic appeal. This in turn assumes that they can be appreciated when mapped onto auditory parameters. The aim of this project was to test this assumption by investigating whether people could classify the qualitatively different states produced by different cellular automata rule sets.
Cellular automata have been used extensively in algorithmic composition. Rule sets can be categorised mathematically (according to a measure of entropy variance in the look up table (Wuenshce)) into one of three classes: ordered, complex or chaotic. These classes can be readily distinguished from a graphical representation of the global CA states which provides a convenient means for testing whether similar classifications can be made from an auditory representation.
Audio representations are particularly useful for comprehending high-dimensional continous data such as neural network outputs. I first started using audio as a debugging technique whilst developing an implementation of Ashby's homeostat. The following examples demonstrate how easily we can hear the global state of a network.
In many situations this is true, and is exploited in auditory display or sonfication, a sonic alternative to graphical visualisation techniques which should be used more often - especially in computational modelling!.
In an early project, CAs were combined with a modified homeostatic network. Outputs from the homeostat are used to define pitch changes; the CA states are used to define when notes play. The following extracts give examples from early in development; some include a 'melody' line, the timings of which were determined by a simple stochastic method
For details please see Adaptive Systems Music MSc Thesis 2002
samples of loops recorded during development of homeostat and CA system:
... sines (commissioned for LUX:open festival of film april 2003):
generated using homeostat to control amplitudes of 10 sine tones in AudioMulch.
... Example output from a final version of this system (8.75M)
Neural oscillators are models of toy neurons. Two 'neurons' are arranged in mutual inhibition such that they oscillate. For a certain (fairly large range) of parameter settings, the nodes will entrain the frequency of an input signal. This entrainment property has proved useful in beat induction tasks, and in rhythmic robotics tasks. Musically this provides a mechanism for creating a sense of ensemble in a network of oscillators.
snippets from early development (apologies - dodgy MIDI implementation):
... Output of 2 nodes with similar natural frequencies out of phase
... Output of 4 nodes with similar natural frequencies
... changing weights between 2 nodes
... output of one node with greater low natural frequency relative to rest of network