Abstract:

The last decade has seen vast advances in commercial synthesizer developments, both in cost and features. Programming of synthesizers require patience and detailed knowledge about sound synthesis. Also, the low-cost interfaces to these units are time consuming to access compared to the sliders of analogue synthesizers. Many musicians resort to costly off-the-shelf preprogrammed sound libraries, often without the actual characteristics and textures required. Therefore, a novel approach to programming of digital synthesizers is proposed. The task of programming is defined as an optimization problem, where the optimization function is the musician's rating of the sound, and the function arguments are the actual parameters of the sound synthesis algorithm, such as ADSR, LFO's, etc. These parameters are coded as chromosomes, and a genetic algorithm is used to search for the desired sounds. The interface works by presenting an initially random palette of sounds to the musician, for him/her to evaluate. Based on these evaluations, the genetic algorithm applies standard genetic operators to the chromosomes and over time breeds the desired sounds. Genetic algorithms ensure rapid convergence. This makes the technical details of a synthesizer transparent to the musician while it still allows him/her to explore all of its advanced features.