Marco Fiorini, PhD candidate at EDITE (ED 130), conducted his research entitled "Co-Creative Istruments for Improvised Musical Interaction" within the Music Representation Team at STMS (IRCAM, Sorbonne Université, CNRS, et Ministère de la Culture), supervised by Gérard Assayag and co-supervised by Hugues Genevois.

The main object of study of this thesis is the design and implementation of co-creative instruments for improvised musical interaction. This thesis focuses on the introduction of shared agency and autonomous behaviors in human-computer music improvisation, shifting the conception of digital technology from a passive, reactive tool to a comprehensive, interactive music system. Operating within the theoretical framework of the ERC REACH project (https://reach.ircam.fr/), this research investigates how artificial agents, such as the REACH Somax2 agents, can be endowed with sufficient cognitive and perceptual capacities to sustain compelling, spontaneous, and symbiotic musical dialogues with human performers.

The jury will be composed of:
• Gérard ASSAYAG – IRCAM - STMS – Supervisor
• Alain BONARDI – Université Paris 8 – Reviewer
• Cheng-Zhi Anna HUANG – Massachusetts Institute of Technology – Reviewer
• George E. LEWIS – Columbia University New York – Examiner
• Nicolas OBIN – Sorbonne Université - IRCAM - STMS – Examiner
• Stefania SERAFIN – Technical University of Denmark – Examiner

The main object of study of this thesis is the design and implementation of co-creative instruments for improvised musical interaction. This thesis focuses on the introduction of shared agency and autonomous behaviors in human-computer music improvisation, shifting the conception of digital technology from a passive, reactive tool to a comprehensive, interactive music system. Operating within the theoretical framework of the ERC REACH project (reach.ircam.fr), this research investigates how artificial agents, such as the REACH Somax2 agents, can be endowed with sufficient cognitive and perceptual capacities to sustain compelling, spontaneous, and symbiotic musical dialogues with human performers.

To achieve this, the dissertation addresses three critical shortcomings in current interactive music systems. First, on the technical side, we contribute to closing the "perceptual gap" by developing a real-time machine listening workflow dedicated to Instrumental Playing Techniques (IPTs). Implemented via the ipt~ Max/MSP external and trained on the open-source EG-IPT dataset thanks to an end-to-end pipeline, this grants artificial agents a deeper gestural awareness of their human partners, moving beyond traditional pitch and amplitude tracking. Second, we address the "autonomy gap" by introducing SoVo, a novel hybrid machinic assemblage that fluidly navigates the dialectic between a purely reactive artificial instrument and an independent artificial player. Third, we tackle the "imagination gap" by expanding the Somax2 corpus-based environment with predictive deep learning models, providing a first prototype built on the Anticipatory Music Transformer mechanism to augment in real-time the corpus of an artificial agent. This architectural coupling introduces mechanisms of anticipation in the generation process, enabling the system to project stylistically coherent musical futures that extend outside the fixed boundaries of its pre-existing corpus.

Our work employs an interdisciplinary and practice-based design approach to investigate the potentials and affordances provided by such systems in the ecosystem of real-world musical performance and creative practices. Rather than evaluating these models in isolation, we considered the use of such co-creative systems by expert users, such as professional improvisers and composers. Finally, this work fully integrated the results of frequent interactions with expert musicians into the iterative design of the models and architectures. Through extensive rehearsals, workshops, and real-world stage performances, we gathered qualitative evaluations and phenomenological feedback to validate and refine our scientific and technological choices. This situated evaluation ultimately demonstrates the emergence of forms of distributed cognition, co-creative dynamics and shared musical agency, and opens to a wider philosophical and aesthetic reflexion on the position and structure of the musical subject in these radically new cyber-human assemblages.