ECHOCHIMERA: a constrained-agency instrument for human-AI performance
Year: 2026
Type: preprint
Abstract: Generative music systems often prioritize continuous output; ECHOCHIMERA investigates the opposite paradigm: the musical potential of refusal and constraint. This technical report introduces ECHOCHIMERA, a real-time MIDI instrument for human–AI performance. The system listens to a performer, evaluates whether to answer, selects one of seven musical characters, and performs a bounded response - or silence - through an allow-listed MIDI router. The core contribution of this work is a constraint architecture - encompassing decision gates, response contracts, character-specific parameters, and a note ledger -designed to make the system's agency audible, bounded, and recoverable. The implementation combines a phenomenological corpus, a Markov model, scale-aware harmony, a 16-step rhythm layer, live controllers, and an explicit performance interface, developed in Python with a PySide6/Qt front end and packaged as a Windows application. The document outlines the system architecture, the encoded musical theory, the performance theory framing the interaction, and the results of an internal evaluation. It is important to note that this report does not present a controlled listening study or empirical audience data.
Author: Rodrigues Paquete, Hugo Filipe
Language: English
Subfield: Music
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225860264
