Abstract

This paper presents the Computoser (http://computoser.com) hybrid probability/rule based algorithm for music composition. It addresses the issues of unpleasantness and lack of variation exhibited by many existing approaches. The paper provides a reference implementation in order to bridge theory and practice.

А sample of 500 popular songs was analyzed to derive probabilities for basic musical characteristics and events (e.g. scale, tempo, going from one note to another). The algorithm was constructed to produce short musical pieces using the derived probabilities combined with a large set of composition rules. The rules were obtained and structured after carefully studying established composition practices. The resulting hybrid algorithm produces pieces that are significantly different from each other and are rhythmic and harmonic. Generated musical pieces were published on the Computoser website where evaluation was performed by human listeners. The overall feedback was positive (3500 likes out of 6000 evaluations), asserting the merits of the undertaken approach.