Musically meaningful fitness and mutation for autonomous evolution of rhythm accompaniment

Abstract

Autonomous evaluation of computer generated musical pieces remains one of the most challenging open problems in evolutionary music composition. This paper introduces the design of novel mutation and fitness operators for autonomous evolution of human-competitive rhythm accompaniment using genetic algorithm. We propose several fitness operators that autonomously evaluate the quality of generated rhythm phrases according to patterns of another accompanying instrument, such as bass or rhythm piano, and according to user's specified criteria such as a density of rhythm accompaniment. In the next, we introduce four musically meaningful mutation operators that alter rhythm consonance, dynamic phrasing, drum stroke types or perform syncopation of rhythm phrases. We also provide a complete example of mutation and fitness settings used for evolution of a rhythm accompaniment to Dizzy Gillespie's "Night In Tunisia" jazz tune.