Automatic Synthesizer Preset Generation with PresetGen

We refer the task of finding preset(s) (i.e. set(s) of synthesizer parameters) that approximates a target sound best, as the preset generation problem. PresetGen addresses this problem regarding the real world synthesizer, OP-1. The OP-1 consists of several synthesis blocks, and it is not fully deterministic. We propose and evaluate a solution to preset generation using a multi-objective Non-dominated Sorting-Genetic-Algorithm-II. PresetGen handles the full problem complexity and returns a small set of presets that approximate the target sound best by covering the Pareto front of this multi-objective optimization problem. Moreover, we present an empirical evaluation experiment that compares the performance of three human sound designers to that of PresetGen. The results show that PresetGen is human-competitive.