Under Construction @ Keele 2016 Volume 1 Issue 1 | Page 51

41
!

A number of systems have demonstrated the effectiveness of applying
interactive genetic algorithms to sound design by providing a user with a simpler,
more accessible interface to a range of sound synthesis and processing techniques
which does not require any knowledge of how the parameters will affect the sound.
The user can simply listen to a collection of randomly generated sounds, then
generate variations on those they consider most suitable for their purposes.
These systems can be loosely placed into two categories: an open ended
approach and a domain specific approach. Both approaches have a number of
advantages and weaknesses. The domain specific approach tailors the genetic
algorithm to a specific process, allowing for the development of systems which are
very easy to use and require no customisation from the user. However, they also
tend to support very few sound syntheses and processing techniques meaning it is
unlikely they could be used to create multiple, varied compositions. The open ended
approach allows an interactive genetic algorithm to be customised for a vast range of
sound synthesis and processing techniques providing endless possibilities. The
drawback of these systems is the complex and time consuming nature of the
customisation process. This is arguably self-defeating as the primary benefits of an
interactive genetic algorithm are ease of use and quicker search space navigation.
The objective of ESDE (Evolutionary Sound Design Environment) was to find
a middle ground between these two approaches where the system retains the ease
of use of the domain specific approach while allowing some of the versatility present
in the open ended approach.
2. Background
2.1. Interactive Genetic Algorithms
An interactive genetic algorithm is a biologically inspired search procedure based on
artificial selection, similar to the ways in which natural selection affects the evolution
of animal and plant species. It maintains a population of solutions which are referred
to as genotypes. Genotypes are typically represented as a binary string which is
converted to a decimal number. This number is then scaled and mapped onto a
parameter. Offspring populations are created which inherit the characteristics of their
parents with a level of random variation, this process continues until a solution has
been found. Offspring populations may be created through one or two parent