Session 1 — Paper Presentations

Musitopia: Designing Interactive Musical Environments for Health and Wellbeing

Raquel Lucena, Mario Ortega, Ariana Pereira and Rafael Ramírez

Keywords: mental health, soundscapes, streaming, music therapy, midi

Abstract

Digital music-based interventions for wellbeing are widely adopted, yet most systems remain unidirectional, relying on passive consumption of pre-curated audio. In this paper, we present Musitopia, an interactive web-based platform that treats music-based intervention as a interactive process between user and system. The platform integrates emotion-aware recommendation, generative soundscapes, binaural beat synthesis, guided meditations, and adaptive breathing exercises within a unified interface. We describe its architecture and design principles, emphasizing closed-loop interaction and adaptive sonic feedback. A pilot study (N=10), combining self-reports and behavioral logs, shows high usability (8.1/10), strong engagement with interactive features, and trends toward increased valence and decreased arousal. These results suggest that interactive musical environments may enhance engagement and support emotional regulation.

 

Vigil 17: Experimental Depth Sonification System

Daniel Manz and Daniel Hearn

Keywords: depth sonification, public installation, depth estimation, umwelt, embedded systems

Abstract

Vigil 17 is a site-specific installation, tracking depth signals from a camera input mapped to an emergent soundscape. Drawing on technical dialogues within Sensory Substitution of Vision by Audition (SSVA) research, this paper presents a novel application of Monocular Depth Estimation (MDE) algorithms to artistic practice. Through practice-led exploration, a framework for translating spatial distance to a procedural audio system is proposed. Here, depth is explored as material, situating the project within established practices of stochastic processes for procedural composition and site-specific sound-art installation. The implementation advances existing SSVA approaches through gradient-based sonification mapping — capturing contours and boundaries, mapped to wavetable synthesis geometry. The viability of real-time embedded MDE-based sonification is demonstrated, performing on-device depth extraction and audio processing. Through public deployment, Vigil 17 suggests a broader ecology — a co-creation process, unfolding through authored technological mediation, between our device and the outside world.

 

Machina Inquieta – Restaging Ashby’s Homeostat

Marco Doettlinger, Arthur Flexer, Patrik Lechner and Thomas Grill

Keywords: Small Data, Intra-action, Machine Listening, Machine Learning, Feedback, Multistable Musicianship

Abstract

This paper presents Machina Inquieta, a digital feedback instrument conceptually inspired by W. Ross Ashby's Homeostat and its principle of ultrastability. Machina inquieta radicalizes cybernetic regulation by replacing equilibrium-seeking behavior with persistent multistability. The instrument consists of four interconnected digital synthesis units. Each unit forms a recurrent feedback loop combining sound synthesis, timbre-based self-analysis, and two multilayer perceptrons that map between a two-dimensional control interface and a multi-parameter synthesis space. When operating simultaneously, these models establish autonomous self-listening systems that continuously reorganize their own sonic states. A global coupling layer interconnects all units, creating a multi-agent ecology in which local perturbations propagate system-wide, including Performer intra-actions. The system demonstrates how multistability can function as a new principle of contemporary musicianship. The project understands data creation as an artistic practice, Small-Data strategies, thus functioning as both artistic resistance and speculative epistemology.

 

The MTU Laptop Ensemble (MTULE): Integrating Hybrid Granular-Markov Systems, Spectral-based Sonification and Gesture Analysis Research into Creative Practice-based Inquiry and Performance

Cárthach Ó Nuanáin, Eric Browne, Brian Sheehan, Aoife Claffey and Juan Francisco Martinez

Keywords: Laptop Orchestra, Computer Music, Generative Music, Sonification, Gesture, Practice-Based Research

Abstract

The MTU Laptop Ensemble (MTULE) was formed in 2025 to put into creative practice several theoretical approaches and lines of research inquiry through the medium of laptop orchestra performance. In this paper we briefly offer an overview of laptop orchestra ensemble performance as a computer music aesthetic, highlighting key historical developments. We describe several of the research topics - including a Granular-Markov hybrid system, spectral-based data sonification and real-time gesture using Bluetooth Low Energy (BLE) Inertial Momentum Unit (IMU) sensors - currently being investigated by members of the ensemble. Through co-creative experimentation, the process of integrating these separate approaches was iterated upon to produce a composition suitable for several public performances.

 

This Way Up: Evaluating Frequency–Elevation Mapping in Audio–Language Models

Vincent Lostanlen and Mathieu Lagrange

Keywords: audio-language models, cross-modal retrieval, environmental sound classification, fundamental frequency estimation, psychoacoustics

Abstract

In psychoacoustics, frequency-elevation mapping (FEM) refers to the widespread spatial connotation of auditory pitch: e.g., ascending and descending scales. We evaluate whether FEM is an emergent property in two deep learning models which are based on contrastive language-audio pretraining (CLAP): namely, MS-CLAP and LAION-CLAP. With MS-CLAP, the answer seems negative: after multiple attempts at prompt engineering, spatial elevation does not correlate with musical pitch in the TinySOL dataset of isolated notes from 14 instruments. The situation is different with LAION-CLAP, for which the prompt "This sound source is located above my ears" yields a surprisingly good zero-shot fundamental frequency estimator (Spearman's rho=0.673). However, the same prompt yields chance-level performance in binary classification between environmental sounds which are typically placed above the listener (e.g., fireworks) versus below (e.g., pouring water into a glass). Together, these results suggest that FEM is not robustly or consistently available in the current generation of pretrained audio--language models. We attribute this shortcoming to a misalignment between the two encoders, rather than a limitation of the audio encoder on its own.

 

MuSA Voice: An Intelligent Platform for Automated, Reflective Vocal Technique Analysis

Suvi Haeaerae, Isabelle Oktay and Rafael Ramírez-Meléndez

Keywords: Music Performance Analysis, Automated Feedback Systems, Vocal Technique Classification, Deep Learning, Transfer Learning

Abstract

We present MuSA Voice, an open, web-based platform for automated, reflective feedback on vocal technique. The system deploys specialised CLAP- and Whisper-based Vocal Technique Classification (VTC) models, trained separately by singer gender and modulation type (pitch and timbre), within a pedagogically motivated analysis pipeline that visualises technique usage across a recording. Applying these large pretrained models to VTC for the first time, we find that Whisper achieves state-of-the-art accuracy and CLAP offers a competitive lightweight alternative, consistent with their strong generalisation to downstream audio tasks. Cross-dataset evaluation on GTSinger reveals that generalisation to real-world singing remains an open challenge, motivating MuSA's descriptive rather than corrective feedback design. MuSA Voice establishes a foundation for a new form of human-machine interplay for singers: one in which deep audio models externalise aspects of vocal technique, supporting reflective practice.

 

BioKybérnēsis: Case Studies on Biomusic and Bioaesthetics in Contemporary Sound Performance

Matt Matt Rogerson, Penelope Penelope Bekiari, Lou Barnell and Courtney Askey

Keywords: Biomusic; Bioaesthetics; Sound and Music Computing; Biofeedback; Adaptive Musical Systems; Cybernetics; Embodied Interaction

Abstract

This paper examines biomusic as a framework for Sound and Music Computing, focusing on the integration of physiological signals such as EEG, heart rate, ECG,
 EMG, breath, and GSR into interactive musical systems. It argues that biosignals should not be understood as direct representations of internal states or as simple control streams, but as mediated inputs shaped by signal conditioning, feature extraction, adaptive mapping, and rule-based or inferential processes. Drawing on cybernetics, neuroaesthetics, predictive processing, and embodied cognition, the paper conceptualises biomusic systems as real-time feedback ecologies in which performer, physiology, and computation co-produce musical behaviour. Through contemporary artistic case studies, it analyses how adaptive system design, instability, and physiological variability shape musical time, texture, and form. The
 paper contributes to SMC discourse by linking embodied interaction, biosignal processing, and aesthetic design, and by proposing biomusic as a model for understanding how musical form can emerge through continuous negotiation between human and computational agency.

 

 

Algorithm-bending as hacktivist musicking: dsp fault injection and augmented saxophone in the ERRO. Project

Alberto Barberis and Bera Romairone

Keywords: algorithm-bending, fault-injection, digital signal processing, augmented saxophone, critical creative coding, mixed-music performance

Abstract

The ERRO. project follows a practice-based approach to develop a situated typology of algorithm-bending strategies, forms of critical creative coding realised through instrumental augmentation and mixed-music performance. We introduce algorithm-bending as a real-time fault injection practice that intervenes audio Digital Signal Processing (DSP) algorithms to produce a sonic ‘error aesthetic’ through the critical interplay between digital objects, augmented instrumentalities, and human performers. This interaction constitutes a form of hacktivist musicking, a practice of code sabotage expressed musically, which contests the algorithmic governance structures in the current technocratic realm. Revisiting Raley’s definition of media tactic as a "disruption of a dominant semiotic regime”, ERRO. re-examines the relationship between (post-)human agency and algorithmic systems through a reconceptualisation of error, and of erring, as potential tools of subversive artistic practice. Errors are systematically considered as latent or manifest events that, whether deliberate or inadvertent, signal a deviation from technical, moral, or semantic systems that prescribe normative functionality. Within this framework, algorithms are not approached ontologically as neutral rule-inscription and execution mechanisms, but as actionable and thus epistemologically situated socio-technical structures that exceed their formal specification. It is precisely within this exceedance that ERRO. finds a threshold for action.

 

Rendering alternative recorded histories of music

Nick Collins and Andrew Brown

Keywords: alternative history, counterfactual, music AI, voice conversion, historical reconstruction

Abstract

Over 150 years or so, a rich and massive recorded history of music has been created. At many points, the musical audio created in studios and live recordings was highly contingent on particular sociocultural circumstances, and could very well have turned out differently. Recent AI technologies such as voice substitution and text-to-audio engines can help facilitate exploring alternative histories of recorded sound. We outline various approaches from reconstruction of early demoes to overt substitution of vocalists, render and discuss case study examples to accompany the paper from The Beatles, Radiohead, Depeche Mode and New Order, and situate the whole study with respect to ethical, legal and security concerns.

 

 

Towards minimal loudness sound masking to enhance sound comfort in offices: an exploratory experiment

Clara Boukhemia, Mathieu Lagrange and Nicolas Misdariis

Keywords: Indoor sound environments, Sound masking, Sound comfort, Indoor quality of life

Abstract

Noise is a persistent issue in indoor environments like shared offices, where acoustic treatments may not provide sufficient auditory comfort. This paper explores two types of masking to enhance the perception of specific sound sources in the office through an exploratory experiment. The masker approach represents a traditional way to do sound masking; superimposing an existing sound to a noise to create a more pleasant mixture. Such an approach can be effective but will increase overall sound levels. We propose an alternative approach, termed concealer, where we seek to adapt the spectro-temporal structure and content of the added sound to that of the noise, in order to reduce annoyance while minimizing overall loudness. Sound mixtures were generated using two ventilation noises and five water sounds across the two approaches and perceptually evaluated for their pleasantness. While no significant difference was found between proposed approaches, we find that more important variations in pleasantness are linked to the type of water sound.