Heretic: Modeling Anthony Braxton's Language Music

Abstract

This article presents a new system for real-time machine listening within human-machine free improvisation. Heretic uses Anthony Braxton's Language Music system as a grammatical model for contextualizing real-time audio feature data within free improvisation. Heretic hears, recognizes, and organizes unseen musical material from a human improviser into a fluid, coherent, and expressive musical language. Systems similar to Heretic often prioritize agnostic approaches to machine listening by avoiding prior musical knowledge in the system's training stage. However, prominent improvisers such as Cecil Taylor, Ornette Coleman, Joe Morris, and Anthony Braxton detail their approaches to improvisation as languages or grammatical systems. These improvisers contextualize the real-time musical materials of their band-mates by applying their formulated grammatical systems to their decision-making processes. Taylor, Coleman, Morris, and Braxton's autonomy and musical creativity are not compromised by using grammatical systems. In regards to human-machine improvisation, Heretic demonstrates that a grammatical approach to machine listening can yield idiosyncratic interactions, full machine autonomy, and novel musical output. This article details a re-imagining of Anthony Braxton's Language Music within the context of machine listening, and an implementation of Language Music within Heretic via SuperCollider's audio feature extraction functionality and Wekinator's multi-layer perceptron neural networks.