Two Approaches for Interaction Management in Timbre-Aware Improvisation Systems

This paper describes two recent improvisation systems that incorporate timbre as an integral element in performance characterization and decision-making. The first London system uses high-level performance descriptors, adapted from recent work in mood detection in music, to coordinate low-level performance events. It was used in performances at LAM 2006 and NIME 2007. After careful evaluation of the results, we addressed the perceived shortcomings in the design of the ARHS system, which manages performance statistics hierarchically; this improved short-term responsiveness, as well as the ability to monitor and adapt to long-term performance variations.