OTIAC: Co-Improvising With a Musical Agent in a Feedback-Based Guitar Performance

Claudio Panariello; Ken Déguernel

OTIAC: Co-Improvising With a Musical Agent in a Feedback-Based Guitar Performance
Image credit: Claudio Panariello; Ken Déguernel

Abstract:

OTIAC —``O totaro int’a chitarra’’ (the octopus in the guitar), from Neapolitan folklore — explores co-improvisation between a guitarist playing on a feedback-augmented guitar, an electronics performer, and an artificial musical agent. The system combines online machine learning with algorithmic sequence generation: a Self-Organizing Map clusters live feedback-based audio in real- time, creating a learned symbolic vocabulary that feeds a Factor Oracle automaton. The oracle then generates and re-injects sonic material into the guitar’s body via transducer, creating a self- referential feedback ecosystem where all the agents shape the emerging performance.

This paper presents an open-source SuperCollider implementation integrating real-time feature extraction, unsupervised clustering (SOM), and context-aware sequence generation (Factor Oracle) within a unified performance system, and reports on preliminary observations from practice-led research with performers. These include the emergence of asymmetric awareness between performers with and without visual access to system state, early performance strategies for balancing manual intervention with autonomous behavior, and empirical insights into the performative and collaborative dimensions of human–AI musical interaction.