ï~~Genetic Algorithms and
Computer-Assisted Music Composition
Andrew Homer and David E. Goldberg
University of Illinois at Urbana-Champaign
Urbana, IL 61801
Abstract
Genetic algorithms (GAs) have been used with increasing frequency and effectiveness in
a variety of problems. This paper investigates the application of genetic algorithms to
music composition. A technique of thematic bridging is presented that allows for the
specification of thematic material and delegates its development to the genetic algorithm.
Some preliminary results are then discussed with an eye toward future work in GAassisted composition.
1 Introduction
Genetic algorithms (Goldberg, 1989; Holland, 1975) have been applied to a wide array of problem domains
from medical image registration (Fitzpatrick, Grefenstette, & Van Gucht, 1984) to stack filter design (Chu,
1989). The accessibility of their interface has contributed to experimentation on problems from many
diverse disciplines. Unlike most traditional optimization techniques, GAs don't rely on a particular problem
structure or problem-specific knowledge. The general effectiveness of GAs in blindly optimizing problems
in no small part accounts for their growing popularity.
It is this accessibility and effectiveness that suggests that GAs might provide a powerful tool in the
computer-assisted composition of music. Various approaches to computer-assisted composition have been
employed in generating music. These range from the relatively straightforward mapping of functions to
compositional parameters (Dodge, 1988) through elaborate stochastic feedback models (Tipei, 1989;
Xenakis, 1971). Recent work using neural network composition methods to learn and generalize structures
from existing musical examples is also of particular interest (Todd, 1989).
The approach adopted in this paper is one of thematic bridging (Horner and Goldberg, 1991). Thematic
bridging is the transformation of an initial musical pattern to some final pattern over a specified duration.
This method is often characteristic of phase or minimalist music and was chosen because of the fact that it
lends itself so naturally to GA encoding.
The bridging is brought about through modifying or reordering elements of intervening patterns through a
sequence of operations. The elements themselves may be discrete pitch, amplitude, or duration values. The
resulting musical output consists of the concatenation of each partially developed pattern (or some variation
of it) through the final pattern. The initial pattern itself is not part of the output, but could be a part of a
preceding section as the final pattern.
As a simple example, with an initial note pattern (Gb,Bb,F,Ab,Db), a final pattern (F,Ab,Eb), and a note
duration of 17, a bridge could be as follows:
operation description resulting pattern
delete the last note of the initial pattern Gb Bb F Ab
rotate the pattern Bb F Ab Gb
delete the last note B b F Ab
mutate the first note Eb F Ab
rotate the pattern F Ab Eb
ICMC 479
