Skip navigation
PDF 949 KB
DOI: http://dx.doi.org/10.7551/978-0-262-33027-5-ch102
Pages 587–594
First published 20 July 2015

Corpus-taught Evolutionary Music Composition

Csaba Sulyok, Andrew McPherson, and Christopher Harte

Abstract

In this paper we present a music composition system that uses a corpus-based multi-objective evolutionary algorithm. We model the composition process using a Turing-complete virtual register machine to render musical models. These are evaluated using a series of fitness tests, which judge the statistical similarity of the model against a corpus of real music. We demonstrate that the methodology succeeds in creating pieces of music that converge towards the properties of the chosen corpus. These pieces exhibit certain musical qualities (repetition and variation) not specifically targeted by our fitness tests; they emerge solely based on the similarities.