Title

Co-improvising creative musical agents: a multi-agent autonomous live looper for improvised co-creation of musical structures

Abstract

This talk introduces a multi-agent autonomous live looper that turns each loop track into an independent listening agent, transforming loopers from passive recorders into reactive co-performers. Agents analyze recently played meso-level phrases using audio-descriptor-based rules to assess compatibility with existing loops, combining metrics such as rhythmic density, timbral similarity, spectral overlap, and frequency-range interaction to either support or contrast ongoing material. Local rule-based decisions across multiple tracks yield stigmergic, emergent macro-structures that move between sparse, dense, and equilibrium textures. To avoid tedious manual tuning, the agents’ behaviors are personalized via a corpus-based genetic algorithm that evolves rule sets from a small, musician-annotated dataset. A study with expert improvisers shows that personalization improves trust, regulates the system’s ‘patience,’ and shifts the musician’s role from controller to curator in a human–machine co-improvisation where agents’ decisions are experienced as both incremental and transformative.

Bio

Stefano Fasciani is a Professor of Music Technology at the Department of Musicology, University of Oslo (UiO), Norway, where he also heads the Creative Computing Hub Oslo (C2HO) and the master’s program in Music, Communication, and Technology (MCT).
His work centers on the intersection of musical interfaces, machine learning applications for sound and music, audio effects and sound synthesis modeling, digital signal processing, and AI-assisted musical interaction. He has an academic background in electronic engineering and professional experience in both the semiconductor industry and the club scene.