Adaptive Real-Time Spatialization for Live Music Improvisation

Live spatialization poses a challenge for instrumentalists, because it requires real-time control over numerous parameters. We present a novel system for automatic audio diffusion that is tailored to improvising musicians. The system enables real-time spatialization of acoustic instruments and other monophonic sound sources by dynamically decomposing incoming audio into spectral streams using Non-Negative Matrix Factorization (NMF). By mapping these spectral streams to separate channels, the system transforms single instruments into spatialized soundscapes with emergent movement. Unlike existing methods that rely on offline training or fixed frequency bands, this approach adapts continuously to the improvised material without offline training, while remaining controllable through a single spread parameter. We implemented a Max for Live device that enables performers to easily integrate the system into their improvisational practice, thereby bridging the gap between live acoustic performance and electroacoustic spatial practices.