In an Artificial Intelligence course at ITU – Modern AI for Games – the Interactive evolutionary computation (IEC) project Picbreeder, and its predecessors, inspired me to explore the feasibility of utilising the powerful properties of Compositional pattern-producing networks, evolved through NeuroEvolution of Augmenting Topologies (CPPN–NEAT), in the formation of unique waveforms to construct oscillators producing interesting timbres. Those oscillators could then be used as novel building blocks in complex synthesizers. Furthermore, I pondered the possibility of using the evolved CPPNs to construct those synthesizers.
To facilitate the beginning of this exploration, I developed a prototype called Breedesizer, for evolving populations of waveforms with the CPPN-NEAT technique, which is accessible at:
A report on the process and its results can be read at:
It showed to be an interesting and enjoyable process to guide the evolution of waveforms with this technique, and it is an exciting prospect to see how the variety of generated waveforms can be used with different sound synthesis techniques, as can be read more about in the Discussion section of the report.
Here are two quick screen recordings of the Breedesizer prototype in use:
Flúðum, janúar 2015