Adaptive Hyperinstruments: Applying Evolutionary Techniques to Sound Synthesis and Performance

DESCRIPTION This paper describes the Genophone [2], a hyperinstru-ment developed for Sound-Performance-Design using the evolutionary paradigm of selective breeding as the driving process. Sound design, and control assignments (performance mappings), on most current systems rely heavily on an intimate knowledge of the Sound Synthesis Techniques (SSTs) employed by the sound generator (hardware or software based). This intimate knowledge can only be achieved by investing long periods of time playing around with sounds and experimenting with how parameters change the nature of the sounds produced. This experience is also needed when control mappings are defined for performance purposes, so external stimuli can effect changes in SST parameters. Often such experience can be gained after years of interaction with one particular SST. The system presented here attempts to aid the user in designing performance sounds and map-pings without the necessity for deep knowledge of the SSTs involved. This is achieved by a selective breeding process on populations of individual sounds and their mapping. The initial populations are made up of individuals of existing hand-coded sounds and their mapping. Initial populations never have randomly derived individuals (this is not an issue as man's best friend was also not selectively bred from protozoa). The user previews the population then expresses how much individuals are liked by their relative repositioning on the screen (fitness). Some individuals are selected as parents to create a new population of offspring, through variable mutation and genetic recombination. These operators use the fitness as a bias for their function, and they were also successfully used in MutaSynth [1]. The offspring are then evaluated (as their parents were) and selected for breeding. This cycle continues until satisfactory sounds and their mapping are reached. Individuals can also be saved to disk for future " strain " development. The aim of the system is to encourage the creation of novel performance mappings and sounds with emphasis on exploration , rather than designs that satisfy specific a priori criteria. By using "locked" parameter sets (a definable set of parameters who's values are not allowed to change), variable control can be exercised on the non-deterministic effect of the evolutionary process. This feature, on the one hand, exercises some control on the shape evolution takes and on the other, allows a gradual familiarisation with the SST involved (if desired). Manual editing of individual parameters for the particular SST is also provided , therefore allowing for precise …