Short Bio: Stefano Fasciani has an academic background in electronic engineering and professional experience in the semiconductor industry and in the club scene. His research and personal interest are focused on technologies for sonic arts, including sound and music computing, sound synthesis and analysis, applied machine learning, human computer interaction, networked music performances, digital signal processing, and real-time embedded systems.
Madaghiele, Vincenzo; Fasciani, Stefano; Kelkar, Tejaswinee & Erdem, Cagri
(2026).
Modeling Relations Between Musical Events in Continuous Time with Transformer Models for Live Co-Improvisational Interactions | Zenodo.
In Coelho, Guilherme (Eds.),
Proceedings of International Conference on AI and Musical Creativity.
AI Music Creativity (AIMC).
doi: 10.5281/zenodo.22276042.
Full text in Research ArchiveShow summary
This paper presents a live semi-autonomous system that co-improvises with a live musician using a model learned from the relationship observed in paired recordings of an improvising duo. This responsive system is based on a deep learning approach that models the relations between sequences of events produced by co-playing musicians in continuous time, using transformer models. We present the architecture for simultaneous sequences of sonic events and provide a quantitative evaluation on several representative tasks. Results are compared against a canonical transformer and a multichannel Factor Oracle, the latter being a widely used model for live sequence-based symbolic music generation. We detail a live implementation employing concatenative synthesis and introduce a customization procedure in which the generative model is iteratively retrained on curated, satisfactory sections from sound recordings of actual human-system co-improvisational interactions. This iterative fine-tuning enables the model’s stylistic output to diverge from the original training corpus. Two musical use cases demonstrate the application of this technique.
Madaghiele, Vincenzo; Fasciani, Stefano & Erdem, Cagri
(2026).
Personalising behaviours of the multi-agent autonomous looper with a corpus of annotated co-performances.
Organised Sound.
ISSN 1355-7718.p. 1–12.
doi: 10.1017/s1355771826101484.
Full text in Research ArchiveShow summary
This article introduces a method to personalise the behaviour of a music co-improvisation system using a genetic algorithm that iteratively identifies optimal configurations governing the system’s behaviour according to a musician’s aesthetic objectives, using a corpus of annotated co-performances. The corpus annotations represent the decisions the semi-autonomous system should learn to make in different musical contexts, reflecting the musicians’ styles, ideas and aesthetic preferences. We apply this approach to personalise the multi-agent autonomous looper (MAAL), a co-improvisation system that autonomously samples and loops segments of an improvised performance by integrating machine listening with a rule-based framework. After detailing the design of the evolutionary search algorithm, we present the results of a study involving five expert improvising musicians. These musicians applied the proposed method to personalise the MAAL and co-improvised with it. We assessed the extent to which the MAAL’s behaviour, adapted to their personalised data, aligned with their individual aesthetic preferences. Findings show that automating the functions of sampling and layering loops seems to invite explorative and responsive creativity and that personalisation can facilitate co-adaptation between musicians and artificial agents by establishing a defined aesthetic vocabulary of co-creative behaviours.
Afsari, Kiyan; Barachi, May El; Ritz, Christian; Copiaco, Abigail & Fasciani, Stefano
(2026).
Low-latency six-channel Bi-LSTM for near-real-time epileptic seizure detection.
Discover Artificial Intelligence.
6(1).
doi: 10.1007/s44163-026-01340-x.
Full text in Research ArchiveShow summary
This paper proposes a deep learning framework for near-real-time epileptic seizure detection using EEG data, designed to address two critical limitations of existing approaches: high detection latency and full-channel EEG dependence. The model employs a Bi-directional Long Short-Term Memory architecture to capture temporal dependencies in EEG signals, processing overlapping 0.5 s windows to enable low-latency sequence-to-sequence classification. Experiments are conducted on the publicly available CHB-MIT Scalp EEG dataset, evaluating different feature inputs including raw signals, channel means, and continuous wavelet transform representations. To address the well-known class imbalance problem in seizure detection, a dataset balancing strategy is applied to improve recall and minimize undetected seizures. A composite scoring metric is introduced to evaluate models jointly on accuracy, detection delay, and false negatives under clinically motivated weightings, providing a more meaningful basis for model selection than accuracy alone. Through systematic architectural search and correlation-based channel selection, the final model operates using only six EEG channels, reducing electrode requirements by 74% compared to the standard 23-channel setup. The system achieves an accuracy of 87.75% with a recall of 77.3% and an average detection delay of 1.33 s, representing a clinically meaningful improvement over prior studies reporting delays of 5–10 s on the same dataset. Performance is further reported in terms of specificity, F1-score, and false alarm rate per hour to provide a complete characterization under class imbalance. The reduced six-channel configuration is validated across all patients using Monte Carlo cross-validation and demonstrates strong suitability for wearable devices and edge computing platforms, where computational efficiency, patient comfort, and real-time responsiveness are critical design constraints.
Fasciani, Stefano
(2026).
Teaching Interactive Music Systems: a Research-Oriented, Project-Based Graduate Course in a Multidisciplinary Master’s Program.
Proceedings of the International Conference on New Interfaces for Musical Expression.
ISSN 2220-4792.
2026,
p. 709–721.
doi: 10.5281/zenodo.20784254.
Full text in Research ArchiveShow summary
This paper presents a project-based, research-oriented graduate course in Interactive Music Systems delivered to a multidisciplinary cohort at the University of Oslo, grounded in a strong NIME perspective and constructive alignment among intended learning outcomes, teaching and learning activities, and assessment tasks. A flipped-classroom model and a Bela + Pure Data toolchain scaffold eleven workshops, progressing from foundational electronics and mapping to a standalone, performable digital musical instrument prototype showcased in a final course concert. Portfolio-based assessment combines open documentation with a short NIME-style academic paper. Analysis of student data and feedback from 2022 to 2024 indicates high engagement and strong outcomes, successful research integration, several projects accepted in the NIME paper track, and robust deliverables. Student feedback highlights hands-on strengths and high satisfaction, while persistent challenges remain around fabrication and time constraints, which in turn inform actionable guidance for NIME pedagogy.
Madaghiele, Vincenzo; Fasciani, Stefano; Kelkar, Tejaswinee & Erdem, Cagri
(2026).
Autonomous Listening-Based Synthesizer Control with Reinforcement Learning for Live Sound Matching.
Proceedings of the International Conference on New Interfaces for Musical Expression.
ISSN 2220-4792.p. 751–758.
doi: 10.5281/zenodo.20784273.
Full text in Research ArchiveShow summary
In this paper, we build on a previously proposed method for training an artificial agent that performs live control of sound synthesizers, using reinforcement learning and machine-listening to match the sound of a live musician. The agent listens to an incoming live audio stream and attempts to match its sound according to specific audio descriptors by controlling the parameters of an arbitrary synthesizer. We introduce a reinforcement‑learning model with a continuous action space that integrates sound synthesis during training and live performance. We propose five complementary reward functions for matching target sounds, and evaluate their effectiveness in three scenarios: in‑domain matching of synthesized target sounds, out‑of‑domain matching of synthesized targets, and out‑of‑domain matching using a corpus of musician‑recorded audio. Finally, we compare the proposed reward functions across these scenarios, we discuss a live implementation and we present the results of quantitative experiments evaluating the matching task.
This paper introduces HS-ims, an integrated virtual environment designed for developing real-time musical multiagent systems. It addresses the need for technological platforms that emphasize sound applications, flexible prototyping, and comprehensive data collection from these multi-agent systems. HS-ims offers a 3D virtual environment with built-in sound synthesis via binaural audio, an Open Sound Control (OSC) API implementation that is language-agnostic, and a data collection module. The platform operates with an External Controller managing multi-agent logic. We present relevant performance metrics and provide examples highlighting how HS-ims has proven valuable in research and education. This encourages further exploration and establishes it as a useful tool for music performance.
Jónsson, Björn Thór; Erdem, Cagri; Fasciani, Stefano & Glette, Kyrre
(2025).
Live Coding the Lineage.
In Sedo, Anna Xambo & Buffa, Michel (Ed.),
Proceedings of the International Web Audio Conference (WAC).
IRCAM.
doi: 10.5281/zenodo.17642435.
Full text in Research ArchiveShow summary
This paper presents a web-based system for exploring and performing with sounds discovered through Quality Diver- sity evolutionary algorithms. Our system renders interactive phylogenetic trees of evolutionary sound discovery processes, transforming genealogical relationships into a performable interface—the “Harpsitree”—using trajectory-based interaction and automatic code generation. The core concept we explore is “algorithmic sketching,” where trajectory-based discovery through evolutionary structures progressively formalises into executable code, bridging visual exploration and textual live coding paradigms. This work contributes a novel performance interface as part of a larger research project investigating evolutionary computation for sound discovery, demonstrating how phylogenetic structures can reveal com- positional affordances through appropriate interaction design.
This paper evaluates and benchmarks the Teensy 4.1 Ecosystem as a low-cost platform for Networked Music Performances. The study focuses on bidirectional streaming of uncompressed audio and assesses various latency components across the system, including I2S passthrough, buffer passthrough, and Ethernet transmission in unidirectional and roundtrip modes, as well as "My Mouth to My Ear" latency over Local and Wide Area Networks. UDP bandwidth utilisation (both unidirectional and bidirectional) and compatibility with JackTrip software are also examined. Results show that a Teensy 4.1-based audio streaming system achieves one-way source-to-ear latencies as low as 0.51 ms at 96 kHz with an 8-sample buffer and 2.50 ms at 48 kHz with a 32-sample buffer, well below the Ensemble Performance Threshold of 25 ms, which marks the limit of perceivable delay. This low latency accommodates additional network propagation delays. Furthermore, the Teensy 4.1 efficiently utilises Ether-net bandwidth, minimising impacts during simultaneous data transmission. Overall, the findings indicate that microcontroller-based boards like the Teensy 4.1 are ideal for low-latency audio streaming, offering better performance and lower costs than general-purpose and some single-board computers.
Madaghiele, Vincenzo; Fasciani, Stefano; Kelkar, Tejaswinee & Erdem, Cagri
(2025).
MAAL: a multi-agent autonomous live looper for improvised co-creation of musical structures.
In Bown, Oliver (Eds.),
Proceedings of International Conference on AI and Musical Creativity.
AI Music Creativity (AIMC).
doi: 10.5281/ZENODO.16946080.
Full text in Research ArchiveShow summary
Live looping is a musical practice based on recording, repetition and layering segments of a live music performance. Layering loops is a technique used to create musical structures that evolve over time throughout a performance, and it is particularly popular among solo improvising musicians. This paper presents the Multi-Agent Autonomous Looper -- a system that autonomously selects musical segments to be sampled and looped during a live performance, using a set of rule-based agents. Each loop track is managed by an agent that listens to the musician's performance and to the concurrently looping segments on other tracks. The decision on which segments of the performance to loop depends on user-selected rules based on comparison of musical properties, expressing varying degrees of sonic compatibility. This approach characterizes the entire system as a multi-agent network capable of autonomously generating emergent musical structures throughout a performance. The Multi-Agent Autonomous Looper aims to expand the possibilities for indirect control and co-creative interactions in live looping for improvising musicians. In the paper, we detail the system's objectives and design, its offline and online implementations, and discuss specific settings and behaviors across three representative use cases.
This roundtable discussion focused on the AI infrastructure that Norwegian higher education, students, teachers, freelancers, and the public and private sectors have access to today, and how MishMash can help improve the situation.
