An aesthetic comparison of rule-based and genetic algorithms for generating melodies

Abstract

Various algorithmic techniques are available for generating music, many of which come from the field of artificial intelligence, which is rich with potential in this regard. However, the musical appropriateness of these techniques is less clearly understood. In this paper, I will report on a study that aimed to describe the characteristics of two of these techniques, rule-based and genetic algorithms, as they apply to melody generation. The appropriateness of these characteristics in contributing to well-formed melodies was judged by aesthetic criteria. The results indicate that most combinations of rules, mutations and evolutionary selection result in poor or average melodies, but that careful combination of these techniques can generate melodies that are not simply well-formed but in many cases display some elegance and novelty. © 2004, Cambridge University Press. All rights reserved.