Automatic Generation of Sound Synthesis Techniques (Garcia 2001)

Master’s thesis at MIT Media Lab

Summary

This thesis proposes a framework for the automatic construction of so-called Sound Synthesis Techniques (SST) based on Genetic Programming. An SST is defined by its “functional form” and its “internal parameters”. The functionl form is the arrangement of the circuit of synthesis blocks used, the internal parameters are relative to each synthesis block.

> Design of SSTs is usually done by selecting a fixed functional form from a
handful of commonly used SSTs, and performing a parameter estimation
technique to find a set of internal parameters that will best emulate the
target sound.

The thesis proposes a way to explore the space of functional forms using Genetic Programming, given a set of inputs (e.g. $A(t), f(t)$) and an expected target output. The internal parameters are optimized within a sub-loop using e.g. gradient descent on the selected functional forms (“Lamarckian evolution”).

A system doing this is developed, based on a custom-designed sound engine.

Some fitness functions are proposed.

Outcomes

The AGeSS system.

Limitations

• The computation time for a given pair of inputs and target is extremely long (on order of 10 hours)