An Application of Evolutionary Algorithms to Music: - Co-Evolving Melodies and Harmonization
Abstract
Algorithmic composition is a field that is close to 60 years old, and has seenmuch research. Systems today are able to do a wide range of compositionaltasks, ranging from simple melody generation to fully automated orchestralcomposition. Systems for computer aided composition are becomingmore and more common, either to evaluate music created by humans, oras generators of raw material to be used by composers.This Master s Thesis describes a novel implementation of a multi-objectiveevolutionary algorithm, that is capable of generating short musical ideasconsisting of a melody and abstract harmonization. The implementationis capable of creating these ideas based on provided material, or autonomously.Three automated fitness features were adapted to the model toevaluate the generated music during evolution, and a fourth was developedfor ensuring harmonic progression. Four rhythmical pattern matching featureswere also developed.The implementation produced 21 pieces of music, under various configurations,that were evaluated in a study. The results of this study indicatesthat the system is capable of composing ideas that are subjectivelyinteresting and pleasant, but not consistently.