MASOM: A Musical Agent Architecture based on Self-Organizing Maps, Affective Computing, and Variable Markov Models

Paper in proceeding, 2017

Musical Agent based on Self-Organizing Maps (MASOM) is a machine improvisation software for live performance. MASOM plays experimental music and free improvisation. The agent perceives and generates audio signals. MASOM combines Self-Organizing Maps for sound memory, Variable Markov Models for musical structure, and Affective Computing for machine listening. The agent learns the sonic content and the musical structure to generate live performances. MASOM’s offline learning uses an audio corpus of recordings of performances or compositions. The machine listening module of MASOM extracts high-level features such as eventfulness, pleasantness, and timbre. The agent listens to itself and other performers to decide what to play next.