Recursive speculation

Recursive speculation Vincenzo Madaghiele Department of Musicology University of Oslo [email protected] Kelsey Cotton Chalmers University of Technology and University of Gothenburg [email protected] Abstract Recursive Speculation is an improvised vocal performance that features a vocalist and the Multi-Agent Autonomous Looper (MAAL), a co-creative sampler/looper based on a multi-agent logic algorithm and machine listening. The MAAL is composed of several agents, each controlling a loop track, which can autonomously decide to sample and play back segments of a live vocal performance by listening to each other. The performance evolves as a dialogue between the musician and the system, in which the machine acts as speculative mirror able to autonomously select and repeat segments of the performer’s voice, layering them to construct evolving musical structures. The system appropriates vocal segments, it disembodies them through repetition, reimagining their meaning and their musical function by recombination and association. The musician and the system are in a co-creative feedback where the musician generates the material the system selects, and at the same time she must respond to the system’s autonomous choices throughout the performance. 1 Project description Recursive Speculation is an improvised vocal performance that features a vocalist and the MultiAgent Autonomous Looper (MAAL), a co-creative sampler and looper based on a multi-agent logic algorithm and machine listening. This performance imagines the autonomous machine as a tool for exploring and understanding oneself, imagining new possibilities. Throughout the performance, the musician improvises employing a wide range of extended vocal techniques, which are sampled and layered by the MAAL as the piece proceeds, forming interlocking, continuously evolving musical structures. The musician adapts to the structures provided by the MAAL, at the same time improvising new materials that are later used as elements of the following structures. This peculiar co-creative relationship, together with the multi-agent nature of the MAAL, generates a complex ever-evolving sound environment in which moments of emergent equilibrium alternate to apparent disorder1. The MAAL is a responsive interactive music system inspired by a live looper, in which several agents, each controlling a loop track, can autonomously decide to sample and play back segments of a vocal performance. Each agent listens to the concurrently looping segments in the other loop tracks, and takes decisions based on global musical properties and its individual objective, previously set by the musician. In this way, the agents are mutually interdependent because they listen to each other, and their actions are bound to the musical context they are immersed in. The interdependence of the agents makes the MAAL capable of highly complex behavioral responses stemming from simple instructions, while maintaining stylistic coherence to the performer’s input. Each agent’s behavior is governed by a rule-based algorithm, employing rules based on machine listening. The agents evaluate the compatibility of a candidate segment from the live performance with other concurrently looping segments based on rhythmical, harmonic and melodic relations computed through machine listening descriptors. The design and implementation of the MAAL are presented in the paper MAAL: a multi-agent autonomous live looper for improvised co-creation of musical structures, presented at AI and Music Creativity conference 2025 (Madaghiele et al., 2025). 1A teaser video of the performance is available at https://www.youtube.com/watch?v=4TMo1WsA-JI. Proceedings of the 6th Conference on AI Music Creativity (AIMC 2025), Brussels, Belgium, September 10th-12th Figure 1: Kelsey Cotton performing Recursive Speculation. The compositional approach of this performance is based on the textural and uncanny qualities of extended vocal techniques, and the use of repetition. Repetition is a fundamental aspect of music (Margulis, 2013), and it is perceived differently at different time scales. This piece explores the repetition and manipulation of mesostructural elements such as melodies and rhythms (Roads, 2015) in the form of loops, to create harmonic, rhythmic and textural relationships (Wishart, 1994). Repeating extended vocal samples has the effect of emphasizing them—promoting reduced listening of their sonic properties (Chion, 2012)—and at the same time disembodying them, estranging from their embodied origin to be perceived as sound objects with a musical function within their context. The MAAL interlocks segments autonomously based on predefined criteria, generating harmonic and rhythmic relations that would be unachievable by a human performer controlling a sampler/looper. The macro-structure of the performance is constantly evolving, however the characteristic behavior of the multi-agent system tends to alternate moments of release (when the system is in a state of equilibrium) and tension (when the equilibrium is broken by the introduction of a new vocal segment in the interlocking loops). Conceptually, the piece imagines this technology as a twisted mirror (Karpashevich et al., 2022), a machine for speculation. “Speculating” can mean taking risks, but it can also signify imagining different versions of oneself, or the future. The word comes from the Latin speculum, which means mirror—an external object that helps us investigate and understand ourselves. In this project, the machine acts as an imaginative mirror, speculating on possible future sonic pathways, revealing alternative potentials of the performer’s intentions, possibly opening up unimaginable paths. By coherently re-combining and layering segments of a vocalist’s improvised performance, the machine acts as a tool for speculation on the performer’s vocal intentions and her future, building a musical structure and trajectory composed of fragments from her past. The interaction between human and the MAAL is co-creative: the agent re-combines segments of the singer’s improvisation, whilst the the singer adapts and responds to the agent—providing further materials for potential evolution to the improvisation. Acknowledgments and Disclosure of Funding This work is funded by the Department of Musicology at the University of Oslo as part of the first author’s project for his doctoral research fellowship. This project relies on open-source coding frameworks, and it would not have been possible without the Pure Data and Python open-source software and online communities. This work was supported by the Wallenberg AI, Autonomous Systems and Software Program – Humanities and Society (WASP-HS) funded by the Marianne and Marcus Wallenberg Foundation and the Marcus and Amalia Wallenberg Foundation. The authors report no conflicts of interest. 2 References Chion, M. (2012). The three listening modes. The sound studies reader, 1. Karpashevich, P., Sanches, P., Garrett, R., Luft, Y., Cotton, K., Tsaknaki, V., and Höök, K. (2022). Touching our breathing through shape-change: Monster, organic other, or twisted mirror. ACM Transactions on Computer-Human Interaction (TOCHI), 29(3):1–40. Madaghiele, V., Fasciani, S., Kelkar, T., and Erdem, Ç. (2025). MAAL: a multi-agent autonomous live looper for improvised co-creation of musical structures. In Proceedings of AI and Music Creativity Conference (AIMC), Bruxellses (BE). Margulis, E. H. (2013). On Repeat: How Music Plays the Mind. Oxford University Press. Roads, C. (2015). Composing electronic music: a new aesthetic. Oxford University Press. Wishart, T. (1994). Audible design. Orpheus the Pantomime York. 3