Genetic Programming for Sound Synthesizers
< prev next >

Design of synthesis algorithms has traditionally been a handcrafted process. Given a handful of standard synthesis algorithms and a target sound to imitate, an expert human/programmer uses different parameter estimation techniques to find a suitable set of parameters to produce a sound similar to the target sound.

Base 10 Labs' sound synthesis design technique is different. Our approach fully automates the process of design and it is not limited to the handful of standard synthesis techniques.

It uses the concepts of evolutionary computation, in particular Genetic Programming, to explore the universe of possible solutions for a given target sound. The building blocks of the solutions are specified (usually in the form of simple operations such as addition, oscillator frequency, multiplication, etc), and a perceptual distance metric is defined by means of psychoacoustic models. The process then iterates and explores possible combinations of the building blocks, guided by the perceptual distance metric. New topologies (synthesis techniques) which could or could not resemble standard synthesis techniques are commonplace findings with this technique.