Autonomous Evolution of Complete Piano Pieces and Performances

Evolutionary algorithms are used to evolve musical score material and corresponding performance data, in an autonomous process. In this way complete piano compositions are created and subsequently performed on a computer-controlled grand piano. The efficiency of the creative evolution depends to a large extent on the representation used, which in this case is based on recursively described binary trees. They can represent a wide variety of musical material and corresponding performance data in a compact form, with an inherent potential for musically meaningful variations and archetypal musical gestures. This is combined with a set of automated formalized selection criteria based on experiences from human selection processes in a previous, interactive version of the same system, leading to surprisingly musical output and convincing performances. The system is also capable of rudimentary learning, through recycling of its own musical output, and an accumulated database of human musical input