Leveraging diversity in computer-aided musical orchestration with an artificial immune system for multi-modal optimization

Abstract

The aim of computer-aided musical orchestration (CAMO) is to find a combination of musical instrument sounds that perceptually approximates a reference sound when played together. The complexity of timbre perception and the combinatorial explosion of all possible musical instrument sound combinations make it very challenging to find even one orchestration for a reference sound. However, finding only one orchestration is seldom enough given the creative nature of the compositional process. Compositional applications of computer-aided musical orchestration can greatly benefit from multiple orchestrations with diversity. In this work, we use an artificial immune system (AIS) called opt-aiNet to search for combinations of musical instrument sounds that minimize the distance to a reference sound encoded in a fitness function. Opt-aiNet was developed to maximize diversity in the solution set of multi-modal optimization problems, which results in multiple alternative orchestrations for the same reference sound that are different among themselves. We compared the diversity and the similarity of the orchestrations proposed by opt-aiNet (CAMO-AIS) against a standard genetic algorithm (CAMO-GA) and Orchids, which is considered the state of the art for CAMO, for 13 reference sounds. In general, CAMO-AIS outperformed CAMO-GA and Orchids for several measures of objective diversity. We performed a listening test to evaluate and compare the perceptual similarity of the orchestrations by CAMO-AIS and Orchids. CAMO-AIS generated orchestrations that were perceived to be as similar to the reference sounds as those returned by Orchids. Therefore, CAMO-AIS has higher diversity of orchestrations than Orchids without loss of perceptual similarity.