Book of Abstracts

This paper presents four Csound opcodes for treating a Markov transition matrix as an explicit compositional object: BuildBiasedMatrix generates a matrix from intuitive affinity scores rather than typed probabilities, MatRotate cyclically shifts which state a hierarchy privileges, MatInterp interpolates between two matrices, and MatEntropy relaxes a matrix toward the maximum-entropy uniform matrix. These operations formalize procedures developed by hand across earlier textural compositions, now reusable across symbolic and audio-based work alike, regardless of the specific chain or material involved. Each operation is demonstrated through a short musical example built on a shared nine-state chain, with transition-matrix heatmaps, state timelines, and a Shannon-entropy curve offered as artifacts that consolidate the compositional intent behind each transformation. The result is a small, composition-first toolkit for shaping musical texture, situated within a practice-based approach to computer-assisted musical composition.

The Audio Random Access (ARA) extension has significantly expanded the capabilities of modern music production by allowing plugins access to entire audio regions. While highly powerful, developing ARA-capable plugins typically requires navigating dense, low-level APIs, a substantial barrier for audio researchers and open-source creators. This paper introduces a new ARA implementation within Cabbage 3¹. It focuses on the design patterns utilised within the core processing backend to negotiate host-to-plugin asset delivery, thread-safe memory sharing, and time-stamped playback markers within plugin environments. Furthermore, it explores how Cabbage 3 abstracts these low-level ARA callbacks, exposing the raw audio data and region metadata through dedicated Csound opcodes. This gives developers full access to the session timeline and audio assets using standard Csound syntax, enabling custom analysis and processing workflows without requiring complex ARA SDK knowledge.

¹ The source code is available at https://github.com/rorywalsh/cabbage3.

Harmonic generation in Csound typically requires explicit chord specification in the score or coordination with an external process. We present the Affective Harmony Machine (AHM), a system that embeds an Open Neural Network Exchange (ONNX) format machine-learning model as a native Csound opcode, demonstrated through affect-driven chord progression generation. The system trains a Random Forest classifier on 108 annotated jazz and pop chord progressions (six emotion classes, seven scale/mode categories) and exports it to ONNX. We also present emoChord, a Csound opcode that accepts an emotion string, performs live ML inference at i-rate via the ONNX Runtime C API, and schedules emotion-conditioned chord events into a synthesis instrument. No Python interpreter or external process is required at runtime; the current release targets macOS (arm64 natively, x86_64 via Rosetta). The implementation demonstrates a general pattern for embedding ONNX-compatible classifiers as native Csound plugins.

Most text-to-music and text-to-audio systems generate a finished waveform: something to listen to, but not something a composer can directly edit, version, or reproduce. CStore instead generates the underlying Csound project file, and unlike established Csound authoring environments such as CsoundQt and Cabbage, it is built around a generative agent trained on a large, curated corpus of Csound instruments. This paper presents the CStore Web Console, a browser-based studio built on top of that generator that lets a composer draft, audition, vary, and revise Csound instruments without leaving the keyboard. The console pairs a Next.js interface with a Flask sidecar that shells out to the Csound command-line renderer, and organizes work around a library of runs, single-key commands for drafting and playback, and two ways of producing variations: deterministic, score-derived “children” and LLM-driven variants tiered by how far they may drift from a “mother” run. A separate path lets a user hand a rendered instrument to one of six language-model providers with a plain-English instruction, such as “add a plate reverb and lower the pitch by an octave,” and get back a new, separately versioned .csd rather than an overwritten one. We describe the console’s architecture and workflow, and what building a studio around an editable .csd implies for how AI-assisted Csound instruments get made and revised.

In an ever-evolving AI landscape, rising subscription costs for proprietary models like Claude and GitHub Copilot threaten to price out small-scale indie developers. To bypass these financial barriers, this paper explores how the Cabbage 3 ecosystem leverages JavaScript to accelerate frontend development. By combining Csound’s high-performance audio processing with decoupled, web-driven frontend architectures, Cabbage 3 provides a highly structured framework for AI-assisted development. Its predictable JSON layout engine, custom-widget boilerplates, and minimalist two-way JavaScript API natively lend themselves to high-accuracy code generation by open-source LLMs without triggering token-heavy hallucinations. Ultimately, this paper argues that Cabbage 3 remains deeply relevant precisely because its structural design bridges the gap between low-friction, AI-assisted web design and human-driven, high-performance DSP.

NERD (Node Editor for Real-Time DSP) is an open-source, Python-based visual programming environment that organises real-time digital signal processing pipelines as networks of heterogeneous, data-linked nodes on an interactive canvas. Its central feature is a first-class integration with Csound: synthesis instruments authored as standard .csd files are incorporated into any pipeline by adding inline comment annotations that declare their input and output channels; NERD’s meta-parser automatically exposes these channels as typed node ports without requiring any modification to the Csound instrument logic. Surrounding infrastructure—biosignal acquisition, motion tracking, MIDI, OSC, and LabStreamingLayer network streaming—is handled by a library of plugin nodes backed by a daemon-thread worker model. NERD originated in a Volkswagen-Foundation funded research project on musical improvisation and joint action. It has since been used in several scientific and artistic events, including the Science Festival Frankfurt 2024 and the neuroscience outreach event Frankfurt hat Hirn 2026, and a live heart rate sonification installation during an artist-in-residence program in 2025. The software is released under the GNU General Public License (v3) and is available at https://gitlab.gwdg.de/lovre.bogdanic/nerd.

Dr.C is a domain-specialized AI system for Csound. It exists as two independent products built around the same philosophy. Dr.C CLI is a terminal application, a two-column TUI with chat and version history on one side and a live CSD, waveform, and signal-flow view on the other, built for production: scripting, batch rendering, and shipping advanced audio-reactive web applications that open in a browser. Dr.C Standalone is a desktop application built to teach: it ships with a free-tier model out of the box, keeps sound design and code preview inside one embedded window, and adds three things the CLI does not have. A thumbs up/down feedback loop and a store of durable, user-stated “lessons” quietly steer future generations toward a student’s own taste, without ever modeling how skilled that student is. A force-directed Knowledge Graph page renders 34 people, 15 institutions, 25 techniques, and 12 landmark works of computer music history as an explorable, pannable, zoomable map. A Player page turns a finished .csd that exposes its parameters as channels into a MIDI instrument, with MIDI Learn binding a hardware knob to a Csound channel automatically, without manual mapping. Both tools share the same curated retrieval over four decades of Csound community publications, journals, and instrument collections; both propose alternatives rather than finished code; both narrate the history behind a technique as they teach it; and both export working Cabbage plugins. This paper introduces Dr.C as a new kind of collaborator for the Csound community: an AI tutor that draws on the same books, catalogs, and instrument collections its authors have taught from for over forty years, and that is already being integrated into how they teach and how their students explore, design, and compose.

The Csound Brain Lab (CBL) is a web-based application that lets anyone get started performing and researching Brain-Computer Music Interfaces (BCMI). It combines the flexible audio-synthesis building blocks of browser-based Csound with one-click connectivity to Muse and OpenBCI headsets through a local bridge, behind a beginner-friendly interface that lets users visualize and listen to their own live brain data, or experiment with a built-in EEG simulator when no sensor is at hand. The system also embeds both remote and on-device large-language models that perform multimodal analysis of EEG and musical material, helping novices find meaningful correlations between brain activity and sound. We describe the mapping layer, in which a chosen target state, a continuously measured affinity to it, and musical parameters expressed as interpolated endpoint pairs together decide what an instrument plays; and we report four preliminary research studies and two EEG composition and performance systems now being prepared with it, each packaged as an application of its own.

CsoundMeta is an immersive multiplayer environment, built in Unity for Meta Quest headsets, that lets performers touch, shape, and share Csound instruments as physical objects in a shared virtual or physical space. Since the system’s Orbs, Wanderers, and AI-generated Worlds were first introduced, our central goal has shifted from handing performers a curated palette to play with, toward giving every Csounder the tools to build that palette themselves. This paper documents three developments that carry that goal forward. First, performers can now bring their own Csound work into CsoundMeta from further away than the headset itself: any .csd file becomes an importable sound for an Orb or Wanderer, Cabbage-authored .snaps files import as ready-to-use presets, and a companion web application now lets a performer upload an instrument, complete with its samples, from a desktop browser, with a Dr.C-based assistant on hand to help design, modify, and debug a Csound instrument that will not yet compile. Second, the AI-generated Worlds that once came from a fixed, curated set can now be prompted, remixed, and renamed from inside the headset itself, and repositioned and rotated so a stage arrangement rehearsed at home lines up exactly once the same World is loaded on the performance stage, and every performer, whether standing shoulder to shoulder or joining from across the world, lands in the same generated room and can hand instruments to one another there. Third, and perhaps most consequential for ensemble work, we rebuilt the live audio pipeline around a native engine that carries the microphone-to-effect-to-speaker signal path outside Unity’s own audio thread, fast enough that we now rely on it for real-time capture, playback, and processing shared live across every client in a session rather than kept local to one, and judge it, as musicians do, by ear. Around that engine we added a dedicated, proximity-gated Mic instrument that can also serve the whole session as one shared microphone with automatic handover between performers, an overdub-capable sampler that records from any source in the patch and shares every take across the room instead of keeping it local, and a separate spatialized voice channel with hands-free push-to-talk for talking through a session without breaking it. We also describe the session tools and in-headset tutorials that keep a room full of personalized instruments and new performers from tipping into chaos. Together, these features move CsoundMeta from a system performers play inside toward a system they compose, build, and record in, one shaped as much by the Csound community as by its designers.

With Csound 7, support for bare-metal microcontrollers with ARM Cortex CPUs was introduced. This allowed Csound applications to target these embedded platforms without OS dependencies for the first time, expanding the possibilities in areas such as custom instrument design, interactive art installations, and DIY performance systems. This paper extends this work to two more embedded architectures, RISC-V and Xtensa, through the introduction of Csound for the ESP32 series of microcontrollers. These chips are widely available, very inexpensive, low-power, usually dual-core, and support extensive wireless networking capabilities. They also run FreeRTOS, providing an unexplored middle ground between bare-metal systems without schedulers and complete operating systems. This balance between extensive features and accessibility makes them an ideal target for expanding Csound’s range of supported platforms. In this paper, we describe the process of porting Csound to these platforms, highlight their benefits and limitations in relation to previously supported platforms, and discuss potential use cases and applications of Csound on these platforms.

Many digital sound tools are powerful, but the technical barrier to entry in DAWs can make them hard to approach. This paper presents OnLive 音, a browser-controlled real-time audio effects console for live input and uploaded samples, built on Csound with a React performance interface. It treats real-time audio processing like a pedalboard rather than a production suite: the user picks an audio source, toggles effects, adjusts controls, and hears the result immediately, without having to build a full digital audio workstation (DAW) session. And every effect can be opened to reveal the Csound .csd file behind it, so users can see exactly how each sound is made and start learning Csound itself. This paper explores how a browser interface can make the depth of Csound directly playable in live performance, through the design and implementation of OnLive 音.

The Tupipa is a traditional three-string plucked lute from southwest China whose sound is preserved mainly through recordings and documentation. Its three strings form two courses: a double course tuned in unison and a single-string course, the latter providing a controlled first case for playable modeling. Such materials document performance practice but do not provide playable controls for physical-model synthesis. This paper presents an analysis-guided workflow for constructing playable Csound physical models from recordings, supplemented by limited physical measurements. The workflow combines ATS and time-domain analysis of recordings with rule-based parameter initialization, diagnostic acoustic-descriptor comparison, bounded local parameter search, and listening-panel calibration. The case study therefore focuses on the single-string course and links measured plucking position, ATS F0, partial balance, decay, and residual/noise cues to Csound controls. In the model-construction stage, initialized controls are instantiated as a four-layer Csound instrument: a wgpluck2 core string-vibration layer, a plectrum-contact layer, a friction/noise layer, and a body-response layer. Rather than introducing another instrument-specific synthesizer, the study shows how recordings can become playable, physically interpretable Csound instruments while retaining listening judgment in the final calibration stage.

This paper presents an approach to navigating the hyperspace of a wavetable, interpreted as a vector in ℝ^N, based on Givens rotations. A discrete audio signal of N samples is treated as a point in N-dimensional Euclidean space; Givens rotations, acting on pairs of coordinates through the sine and cosine of an angle, displace this point while preserving the L² norm. Four strategies for selecting rotation pairs and angles are proposed and compared: a local adjacent chain, a non-local prime step, Sobol quasi-random sequences, and a Kuramoto coupled-oscillator model. Each strategy is realized both as an interactive Python prototype and as a Csound plugin opcode operating on 1024-sample wavetables. A preliminary, qualitative audio observation of the four strategies under controlled parameter sweeps is also reported, with a supplementary rendering dataset.

This paper presents a real-time vision-to-music system designed to explore an accessible and interactive auditory experience for people with visual impairments. A webcam feeds a Python/OpenCV application that identifies up to five prominent colored regions as object candidates and extracts each one’s position, relative size, and dominant color. These features travel to Csound over Open Sound Control (OSC), where five continuously running synthesis instruments turn them into an evolving soundscape: horizontal position sets stereo panning; vertical position controls pitch, chord voicing, or rhythmic pattern depending on the voice; relative size shapes amplitude, filter cutoff, reverberation, and delay; and the scene’s dominant color selects one of eight musical modes shared by all five voices, so color becomes harmony. The soundscape continues to evolve through timed chord changes and slow modulation even when the visual scene remains still. The system demonstrates how computer vision and real-time synthesis in Csound can support a musical, nonverbal form of environmental interaction alongside existing assistive approaches.

This paper presents Binaural Explorer: Theta, a cross-platform Cabbage plugin that embeds precise theta-frequency binaural beats directly into existing music, generative soundscapes, and live DAW environments. The plugin uses the Csound synthesis engine to apply a real-time spectral phase vocoder, superimposing a neurologically-active theta differential (4.0–7.2 Hz) onto arbitrary audio signals without degrading the underlying sonic structure. Users can substitute source tracks, activate a fully binauralized generative composition mode, and fine-tune target beat frequencies via granular controls. The system includes dynamic visual feedback (a pulsing “beating cabbage” graphic scaled to the live theta rate), an interactive educational mode, and a linked informational pane for psychoacoustic study. A small early-adopter cohort reported reduced sleep latency, enhanced study focus, and a striking expansion of the perceived soundstage. This readily-customizable, open-source tool functions as both a research utility and a pedagogical toolkit, supporting informal clinical research, interactive sonification, and meditation practice.

Implementing a custom physical modelling opcode in Csound requires fluency in C programming, a skill that remains a significant barrier for many musicians and researchers in music acoustics, despite the strong potential of physics-based sound synthesis. This paper presents a development workflow for creating a Csound opcode, illustrated through the case study of the Đàn Bầu, a Vietnamese monochord instrument plucked by the right hand and pitch-bended by the left hand through a stem that modulates the tension of the string. Starting from a generic string model drawn from the acoustics literature, we first prototyped it in Python, allowing for straightforward implementation and assessment of the energy balance. We then translated the Python code into C with the help of Large Language Models (LLMs), which produced a functionally correct translation in a single pass. Finally, the resulting C code was manually adapted into an opcode plugin, enabling its integration into Csound. This case study suggests that the complete process from mathematical model to performable instrument through Csound can be achieved without deep C programming expertise. The workflow presented here offers an illustrative example that musicians and researchers in music acoustics may draw on when developing other physical modelling opcodes for real-time implementation in Csound.

Spatial Agents is a live-electronics instrument built on a single premise: that the place of a live sound should follow its timbre. Five piezoelectric contact microphones on a hand-built resonating box are analysed in real time into five audio descriptors, and those descriptors move five autonomous sources, or agents, through an Ambisonic field decoded to a horizontal loudspeaker ring or to headphones. The paper is as much about a compositional stance as about a system. Following a tradition of spectral and timbre spatialisation, we treat space not as a container for sound but as a dimension the performer shapes directly: a still, pure resonance settles near the centre of the field, a bright gesture rises, a noisy evolving texture is thrown to the periphery and set spinning. We describe how the descriptors map to motion, focusing on a mapping in which spectral stillness and purity pull a source inward while spectral change pushes it outward, and we explain why we run Csound as the engine inside Max rather than Csound alone: the pairing joins Csound’s analysis and synthesis vocabulary and its portable, sample-accurate processing to the live routing, control and visual feedback that a performance patch needs. The Ambisonic decoder uses the ACN/SN3D convention and is rebuilt at run time from a description of the loudspeaker layout. The approach adapts a descriptor-driven spatialization model previously introduced for corpus-based improvisation; what is new here is a self-contained Csound engine, built from scratch and driven live by a physical contact-microphone instrument rather than a pre-analysed corpus. The instrument is functional and has been validated; we close with the compositional questions it opens and with work still in progress.

Live performance leaves little room for a sensor pipeline that misfires. When a violinist’s gesture is intended to trigger one sound but produces two, the error is immediate and audible, and mapping design alone cannot fully compensate for it. This paper presents L-Bow, a wrist-worn six-degree-of-freedom (6-DoF) inertial measurement unit (IMU) controller for augmented violin performance, designed and built by the authors, and asks whether the sensor-to-sound pipeline can be made more reliable by removing an intermediate software layer. L-Bow integrates gesture sensing, six performance modes, and a shared effects chain within a single self-contained Csound file, using Csound’s native arduinoRead opcode for Arduino communication rather than a Python serial parser and OSC bridge. A controlled trigger test comparing the two workflows showed a modest latency advantage for the native path. The paper discusses what this means for designing dependable, low-maintenance performance systems, and where the native approach still falls short.

While the Csound ecosystem provides a vast library of tools for synthesis and signal processing, practitioners in real-time algorithmic composition and experimental sound design often encounter a gap in native, instrument-level opcodes for complex generative processes. This limitation frequently forces composers into hybrid workflows with external environments, introducing latency and synchronization overhead. This paper introduces csonus, an open-source collection of C++ opcodes designed to bridge this gap by integrating algorithmic composition and sound generation techniques directly into the Csound instrument block. Built using the Csound Plugin Opcode Framework (CPOF), the package provides efficient implementations of diverse tools, including chaotic systems, cellular automata, Lindenmayer systems, and neural network-driven synthesis. By enabling tightly coupled, multi-level algorithmic decision-making and experimental sound manipulation without external dependencies, csonus fosters a more integrated workflow for both studio and live electronics contexts. The library is written in modern C++ and is publicly available under the MIT license.

We present an exploratory audiovisual biofeedback system that integrates real-time electroencephalogram (EEG) signals, large language model (LLM)-based assistance, and Csound within Unity, a real-time game engine, transforming a listener’s brain activity into continuously evolving sound and a synchronized visual scene. NeurOSC, a Python-based EEG acquisition and signal-processing layer, estimates power within five predefined EEG frequency bands, calculates their cross-channel mean relative composition, and sends these features to Unity via Open Sound Control (OSC). Unity converts the features into Z-scores using a calibration reference established for the current session and assigns one of five system-defined state labels: Focused, Relaxed, Aroused, Drowsy, or Neutral. The system uses an LLM in two roles. Before a session, a mapping assistant combines the creator’s natural-language intent with a catalog of Csound parameters parsed from the instrument’s Cabbage header and proposes EEG-to-Csound bindings that the creator can audition and revise. At run time, a scene agent responds to confirmed state transitions and other configured events. It sets larger-scale base targets for selected Csound parameters and issues coordinated visual actions, while a continuous path maps changing EEG features to smaller-scale modulation around those targets. The prototype runs in real time with a Muse S Athena headset and includes a simulator for development without hardware. Its contributions are a human-in-the-loop mapping workflow that combines parameter auto-discovery with LLM-generated suggestions, and a two-path architecture that combines continuous physiological modulation with discrete, context-informed audiovisual changes.

This paper evaluates Csound7’s Orchestra language as a host for complex symbolic programs. As a case study, I implemented a Clojure-inspired Lisp interpreter entirely in Orchestra using the Make-A-Lisp process. The interpreter uses recursive structs, arrays, and typed user-defined opcodes to implement lexical environments, closures, macros, tail calls, and exceptions. A musical example uses first-class graph values to build instruments and schedule musical patterns, keeping the symbolic program and audio engine in one process. The work exposed defects in type propagation, nested struct arrays, overload selection, pass-by-reference semantics, recursion safety, and default initialization, leading to fixes merged into Csound7’s development branch. The main limitation in the version studied here was memory lifetime: temporary recursive values persisted until the enclosing instrument ended, making even a small self-hosting test costly in memory. The result shows that Csound7 can support substantial symbolic programs close to its music engine, and that the lifetime of structured values matters for programs that allocate heavily.

Empirical impulse responses can be obtained by measuring physical systems. Csound has built-in means for utilising these in musical applications through its various types of convolution opcodes. Until now, the system lacked out-of-the-box means for preparing empirical impulse responses, although this can actually be implemented from first principles in Csound code. This paper introduces a native toolkit developed for this purpose, in three forms: a utility, a GEN, and opcodes.

This paper addresses the high cost and long development cycles of custom audio plugins by proposing an agile workflow that couples Csound with large language models (LLMs). Building on this workflow, we developed Wildtrace, a quadraphonic spatial audio plugin suite driven by a family of generative algorithms. The core idea of the suite is to invert the usual premise of spatial audio design: rather than moving sound objects through space as an end in itself, it treats the spatial motion of sound as the engine that drives sonic effects such as Doppler shift, panning, granulation, and chorus, so that motion shapes the sound instead of merely positioning it. The paper systematically describes the development logic, the human–AI collaboration model, the layered technical architecture built around a replaceable global renderer, and the self-designed front-end interaction. We assess the approach against conventional C++ plugin development: although no systematic benchmarks were run, the qualitative evidence is encouraging—eight plugins were completed and tested in under one month at zero direct API cost, demonstrating a viable paradigm for agile audio prototyping under resource constraints.

Csound’s native virtual analog opcodes rely on fixed algorithmic structures that cannot learn transfer functions from empirical data or capture physical analog hardware. We present a real-time neural inference pipeline integrating Csound’s C++ Plugin Opcode Framework (CPOF) with the RTNeural engine, executing stateful recurrent networks directly within Csound orchestra syntax. We evaluate the pipeline on two targets: a parameterized 4-pole Moog ladder filter trained on oversampled numerical simulations, and a fixed-setting analog distortion pedal trained on physical loopback recordings. For the filter, shortening the loss-masked training warmup to 256 samples forces rapid convergence from a cold hidden state, enabling click-free note-on with only a brief output ramp; reducing the warmup further to 128 samples reduces static ESR at the 100 Hz cutoff by 7.6 dB. A multi-rate LFO training regime addresses dynamic tracking degradation: a static-trained 128-unit baseline degrades by 17.3 dB under a 5 Hz cutoff sweep, while multi-rate LFO training compresses rate-dependent variance to 3.6 dB across 1–20 Hz on held-out audio. The same recurrent architecture, without the cutoff input, achieves −29.3 dB held-out ESR and −37.9 dB aliasing-to-signal ratio at 8 kHz on the hardware distortion pedal when trained via pre-emphasized ESR on loopback recordings. Deployed 128-unit Gated Recurrent Unit (GRU) opcodes consume under 17% CPU per voice, sustaining 6-voice polyphony at 48 kHz.

This case study presents RIFFIT, a commercial text-to-song platform that utilizes the Csound audio programming language. This platform enables users to turn any text into a fully-produced song in near real-time, with a variety of genres and voice models to select from. Csound serves as the platform’s primary synthesis engine, performing a custom CSD file that is dynamically generated based on both the inputted text and selected musical preferences.

This paper begins with a brief introduction to the field of automatic music generation, which is followed by an overview of the RIFFIT platform and key developments that have occurred to date. The software’s architecture is then outlined, with topics including the dynamic generation of Csound code, the manual performance of Csound files via the Csound API, and strategies for the web deployment of Csound-enabled software. Lastly, an analysis of Csound’s suitability for such commercial applications is provided.

This paper describes the creative and technical process of the performance Postcard from Sertões. The work details the transposition of Almeida Prado’s Sertões (1978) for solo viola [1] into the interactive electroacoustic domain, inspired by the compositional economy of James Tenney’s Postal Pieces (1971) [2]. The research was developed within an Interpretive Experimentation Laboratory, which consisted of distilling Prado’s “transtonal system” into four essential sound blocks, interpreted as “clangs”, primary perceptual units based on Tenney’s theory. The performance establishes a bridge between written tradition and orality by distributing musical materials between two distinct performers. An acoustic interpreter plays the Brazilian rabeca and the viola, while an algorithm interpreter manipulates the audio input in real-time. The implementation, done via Csound and Cabbage, involves a digital signal processing (DSP) system operated by a postcard-inspired visual interface. Through four looper tracks enhanced by effects, the algorithm interpreter reshapes the clangs’ identities. The laboratory explored and refined the interaction protocol between the musicians through remote testing and in-person rehearsals. The artistic practice suggests that technological mediation promotes a transformation of the work, resulting in a new hybrid composition that functions as a “virtual postcard”, reactivating the symbolic resonances of the sertão through a phenomenological and immersive listening zone.

Somatic Continuum is an interactive installation in which a visitor’s face, hands, heartbeat, and presence become the instrument. There are no buttons and nothing to hold. A camera reads the face through MediaPipe FaceMesh, a hand tracker reads the hands and the movement of the arms, and a worn heart monitor reads the pulse. All of it drives a live Csound orchestra of more than twenty instruments over Open Sound Control. The instruments are spread across nine worlds, and each world puts a different instrument on each hand. A shared key keeps every combination in tune. The visitor’s heartbeat pulses the image and sets the timing of the arpeggios, so image and sound move with the body. Each person who steps in and connects their heartbeat opens a new sonic world. Nothing is recorded in advance. The music is generated live by rules that the body shapes as it moves. Koenig and Tenney composed by setting rules and letting them generate the music; Somatic Continuum does the same, except the rules run in real time and the body sets them in motion. This installation gives Csound a new kind of interface, one played with the body, that lets an untrained visitor compose.

The pch2csd project, initiated in 2015, preserves the sonic heritage of the Clavia Nord Modular G2 synthesizer by converting its patches into Csound code. This paper presents the evolution of pch2csd into a web-based modular synthesis environment that combines the proven Python conversion engine with a real-time visual patching interface. We detail the system architecture, the reuse of the existing codebase through a Flask API, and the design decisions enabling seamless interaction between the visual patch editor and the Csound audio engine. The result is a platform that not only preserves the Nord Modular heritage but extends it into browser-based sound design, allowing users to load, visually edit, and hear patches directly in a web browser without proprietary software.

I show how to evoke the pitch-class content of a chord from an arbitrary source by convolving it with a short impulse response whose grains lie on those pitch-classes in every octave. A Csound implementation (chord convolver) builds one joint IR with frequency-dependent grain length, equal-energy normalization across octaves, inverse A-weighting, a dry Dirac, and a single ftconv. I contrast the effect with a linear comb and a generic constant-Q bank, and demonstrate it on a twilight field recording alongside Château de Lagarde.

CsoundUnity is an open-source Unity package embedding the Csound audio engine, exposing its synthesis and signal-processing capabilities through a MonoBehaviour component and its custom Inspector. This paper presents version 4.0’s changes, marking a shift from a channel-driven wrapper toward an integrated, interactive audio environment.

Version 4.0 upgrades the embedded engine to Csound 7 (beta) across all platforms (macOS, Windows, Android, iOS, visionOS, WebGL) and adds new functionality in several areas. Native low-latency input backends (CoreAudio, WASAPI, AAudio, RemoteIO) bypass Unity’s buffered Microphone API through a native callback it does not provide; the AAudio backend reports approximately 4 ms of input latency on a Pixel 4a. An audio-routing layer lets CsoundUnity instances feed one another’s input, with cycle detection and an interactive node-graph editor; two new AudioPath options, IAudioGenerator and RootOutput (Unity 6.3+), join the existing OnAudioFilterRead callback – the former feeds an AudioSource directly, the latter writes straight to Unity’s main output, bypassing both AudioSource and mixer. Timeline integration adds channel-automation clips and eight score-generation clip modes, spanning hand-authored Score lines, an arpeggiator and chord generator, two step-sequencer variants, and generative Euclidean, stochastic, and swarm-based event generation.

Together with a UI widget system auto-generating a playable interface from a Csound document’s Cabbage GUI layout, cross-platform MIDI input, and a new vector-morphing layer built on the existing preset system, these additions move CsoundUnity toward an integrated, multi-platform environment for performing Csound instruments inside Unity – the same project is portable across desktop, mobile, and web. We close with directions for future work, including native audio output.

Idýl is a functional, reactive programming language designed to describe and program time. Inspired by Antescofo and by Faust’s principle that “everything is a stream”, it treats time as a first-class value and change as the object of computation, rather than as a side effect of a general-purpose runtime. Definitions are stateless by default; state lives explicitly inside temporal blocks driven by a high-accuracy, drift-free scheduler. The language is generalist by intent: composition, performance, live-coding, and the control of arbitrary devices. It reaches the outside world through modules, C++ extensions that expose external APIs to Idýl code. Its reference module targets Csound: the scheduler drives one or more independent Csound instances, sequencing i-events, writing control channels, and polling channels back into the reactive graph. We present the language, its temporal semantics, the Csound bridge, and the engine that executes it.

This paper reports on the reconstruction of Cordelia, a domain-specific language for performance (live coding), composition and notation based on Csound. Where the original implementation relied on hand-written, string-based mini-parsers to interpret its compositional syntax and translate into Csound langauge, the rewritten system is built on a formal Lark grammar processed by an Earley parser with a dynamic lexer and prioritised token classes, supporting LilyPond-style pitch notation alongside Cordelia’s own rhythmic and instrumental syntax. A seven-stage compiler pipeline replaces the previous single-pass conversion from text to Csound code, with a stage querying outward into a plugin-based corpus. The paper situates these changes within Cordelia’s founding principle of resource efficiency and argues that formalising the grammar has not constrained the language’s character but clarified it, making explicit a structure that was previously implicit.

This paper presents a programmable Csound sound engine for machine-listening-driven real-time audio processing in live electronics. The system integrates Max/MSP, FluCoMa, a multilayer perceptron (MLP) regression model, and Csound in a layered architecture for audio analysis, parameter mapping, and sound processing. Max/MSP handles audio input, control scheduling, and OSC/UDP communication, while FluCoMa extracts acoustic features. The MLP maps these features to two continuous performance-state coordinates, which a subsequent mapping layer converts into eight digital signal processing (DSP) control values. Training takes place before performance. Network parameters and feature-standardization parameters are saved in separate JSON files and loaded for inference during performance. Csound applies delay, spectral processing, nonlinear transformation, random modulation, and spatial diffusion to the input audio according to the received control values. The paper describes data transfer, parameter mapping, and smoothing across the layers and examines the processing workflow through model-driven offline replay of a complete recording. The resulting design provides a programmable connection between machine listening and real-time electronic sound processing.

One algorithmic composition may require a Csound score, engraved notation, real-time control, and a rehearsal click. Authored separately, their timelines drift. Temporal System is a Wolfram Language paclet that instead compiles one immutable store of typed entities on a rational beat timeline through backend-specific contracts. It emits Csound synthesis, beta MusicXML 4.0, OSC control, and click artifacts that remain synchronized because they share that store. Conversion to seconds, samples, or hertz occurs only at render time. Csound notes use stable named instruments in external .orc files; curves become k-rate signals declared against score p-fields. The click backend derives rehearsal audio from the same meter and tempo and reuses the Csound serializer. We describe the temporal, semantic, and rendering-contract layers, their practical trade-offs, and the limits of this proprietary authoring environment within an otherwise open-source ecosystem. The archived supplement exposes the reported outputs pending paclet release.

In September 2024 the U.S. Food and Drug Administration authorized the first over-the-counter hearing-aid medical device software, turning a pair of consumer earbuds into a regulated medical device and making an audio signal-processing algorithm into a treatment whose behavior must be verified against a known output, not simply tuned by ear. This paper’s focus is to build the artifact the paradigm asks for: a dependency-light hearing-aid prescription engine written entirely as a Csound orchestra. From one editable row of audiogram thresholds it computes, at initialization time, a per-band NAL-R/RP insertion-gain target and a wide-dynamic-range compression (WDRC) law and applies them in an eight-band compressor — the whole device readable in a single sitting and portable, through libcsound, from a desktop rig to on-ear silicon. We verify it at two levels. On the bench, offline Csound renders establish that the core is deterministic to the sample (a bit-exact golden file), that its initialization-time prescription reproduces the published NAL-R formula exactly, and that its input/output behavior follows the prescribed compression law up to a fixed maximum-output ceiling. In situ, we couple the same orchestra to a manikin through an ordinary pair of AirPods — an interchangeable stand-in for any suitable transducer a finished product would use — and measure the delivered sound with a VeriFit 2 real-ear system against NAL-NL2 targets for a standard N3 hearing loss [25]. At a 65 dB SPL conversational input the aided response matches target with a root-mean-square deviation of 4.2 dB, excellent through the core speech region; the same measurement shows a high-frequency over-shoot consistent with a limitation the core’s own documentation predicted in advance. We frame the companion-interface and distributed co-processing ideas as hypotheses rather than results, recast azimuth-based source separation as a candidate alternative to directional-microphone beamforming for speech-in-noise, and place the whole exercise honestly inside the regulatory apparatus of IEC 62304, ISO 14971, and ISO 13485, with the road to clinical validation named but not yet walked.

The workshop explores self-generative musical structures based on recursive instrument calls in Csound 7. Participants investigate different control strategies and non-hierarchical processes, combining coding with artistic decision-making.

Somatic Continuum is an interactive audiovisual installation in which a visitor’s face, hands, heartbeat, and presence become the instrument and art piece itself. It has no buttons and no controllers. A camera reads the face, a hand tracker reads the hands/arms, and a heart monitor reads the pulse. All of it plays a live Csound orchestra of more than twenty instruments, spread across many sonic worlds and held together in a shared key so that it can always sound harmonically connected. Nothing can sound wrong, but nothing happens without engagement: the more the visitor engages, the richer the soundscape becomes. Each person who interacts with the piece brings a new sonic world into being, and everything they see and hear unfolds in time with their own heartbeat. They can record their own voice and layer it in, and this interactive installation allows the audience to make their own proper art that could not exist without them. Somatic Continuum also allows users to take images, videos and audio recordings of themselves inside the piece so that users can take with them their own proper artwork they created. For artists and designers that want to custom make their own sounds and connect it to Somatic Continuum, this installation sends MIDI data so that whatever instrument or sound they want to use for their personal expression is possible, and that MIDI data allows more complex iterations of sonic exploration within Csound and with other audio programs. Nothing is recorded in advance, and nothing is fixed. In placing generative and embodied Csound synthesis at its center, the work engages the algorithmic and different sonic processes created by Gottfried Michael Koenig and James Tenney.

Workshop participants rehearse and perform a work for multiple mobile phones using a web app. The app is a csound web assembly version of a native app initially created for an ensemble of concert musicians to perform microtonal music. Players launch the app from a local server accessed via a Wi-Fi Router. Concert sound relies on venue acoustics rather than brute force amplification. Multiple hand-held mobile phones will form a malleable speaker array. Instruments are easy to play, quick to learn and tuned to a 25-note microtonal scale enabling musicians to explore uncharted harmonic spaces rarely visited using standard musical instruments.

Participants use Meta Quest 3 headsets to interact with sound objects in a shared virtual environment powered by Csound and CsoundUnity. Sounds are manipulated through gestures within a collaborative networked jam session.

The workshop will culminate in the world premiere of a networked solo whose shared virtual environment draws on Sicilian landscapes, architecture, and mythology through AI-generated visual worlds. As the session concludes, participants will be invited to join in and play along.

The workshop introduces the spectral music of James Tenney, featuring a performance of Saxony (1978) alongside a collective performance workshop of Harmonium #7.

Harmonium #7 is built on the harmonic series and sustained, justly-tuned tones. Participants are invited to design and bring their own Csound instrument(s) capable of sustaining tones with precise pitch control (derived from harmonic ratios), volume adjustment, and, where possible, expressive control over timbre. Two modes of participation are offered: a physical setup, where participants connect their laptops or audio interfaces to a shared mixer and PA system, and a network-based setup, where instruments are compiled and controlled remotely via UDP/OSC messages sent to a master computer. The latter also serves as a live demonstration of Csound 7’s new UDP server functionality for OSC-based instrument control.

The workshop introduces the spectral music of James Tenney, featuring a performance of Saxony (1978) alongside a collective performance workshop of Harmonium #7.

Saxony by James Tenney is a fine example of process-based electronic music, originally realized using two tape recorders in a feedback loop. This performance presents a new realization of Saxony created in Csound, utilizing an electronic wind controller rather than a saxophone. The Csound implementation centers on wave terrain synthesis as the primary method of sound generation. The feedback loop is recreated digitally, providing precise control over delay time, feedback, and other parameters. In addition, this realization ensures that the 32 required overtones of the low E♭ sound at their exact pitches. The piece is performed live by Tarmo Johannes using the Sylphyo, an electronic wind controller that provides expressive, breath-driven control over synthesis parameters.

Floating on Their C of Sound: Remembering Barry & John is a solo Csound/XR performance, six minutes long, written in memory of Barry Vercoe, who died on 15 June 2025 in Tauranga, New Zealand, and John ffitch, who died in June 2026 in Bath, England. Alone on stage in a Meta Quest 3 headset, Richard Boulanger floats through a single CsoundMeta world built from what the two men actually left behind: their own voices, drawn from archival interviews and talks, their own music, and images from their lives and their work, the MIT Experimental Music Studio, Csound’s source code, a Bath office, all surfacing and receding inside the world the performer alone inhabits. What the headset shows the performer is projected for the audience to see as they listen, so the audience watches the performer remember. Granulated fragments of Vercoe’s 1976 Synapse for Viola and Computer and of the networked duet Boulanger and ffitch wrote and performed together drift like driftwood on an ocean that is also the four decades of Csound the two of them, in their very different ways, kept afloat. The piece is the performer’s own way of floating with both men once more before passing them, and the gift they gave him and everyone who works in Csound, on to whoever performs or hears it next. There is no fixed script: which voice, which fragment of music, which image rises to the surface, and when, is left to change from one performance to the next.

Bella Tenor (2018) is a work for tenor saxophone and embedded Csound live-electronics that uses algorithmically organised verbal instructions to guide an improvisation. A Csound patch thereby plays the instructions for parameters like dynamics, duration, sound content, register, and tempo, to shape the performance without fixed notation. All electronics are built as a portable setup into the saxophone, including microphone, loudspeaker, and a Bela board. The sound is processed in real time using the COSMO library and played back through the loudspeaker in the saxophone’s bell, blending acoustic and electronic sound.

Convergent Series is a fixed media quadraphonic electroacoustic composition conceived as a tribute to Gottfried Michael Koenig and James Tenney. The work draws on two foundational currents in computer music: Koenig’s serial parameter organization and Tenney’s spectral and psychoacoustic explorations. The piece unfolds in four continuous sections—Emergence, Dialogue, Ascent, and Convergence—in which spectral drones, FM synthesis cells, Shepard-Risset tones, and rhythmic percussion layers coalesce from silence, engage in polyrhythmic counterpoint, and dissolve back into acoustic space. The title refers to the mathematical notion of a convergent series, in which successively smaller terms sum toward a finite whole—an analogue for two composers whose ideas, developed independently across decades, converge in the contemporary practice of algorithmic and spectral computer music. The harmonic language is derived from upper partials of the overtone series (partials 7, 9, 11, 13, and 15 of A0 = 27.5 Hz), yielding microtonal pitch material that exists outside equal temperament. FM synthesis employs irrational modulator-to-carrier ratios (φ, √2, √3, √5, π, e) to generate inharmonic, metallic timbres. Rhythmic organization in the central sections draws on additive metric patterns (2+3+2 / 3+2+3) that accelerate progressively across the duration, reaching a climactic density before dissolving. The quadraphonic space is treated as a compositional dimension, with independent azimuth rotation arcs for each layer and asymmetric front/rear reverb processing creating a sense of acoustic envelopment and spatial emergence.

With this work, I present a fantastical interpretation of a musical phenomenon that emerged in the United States at the dawn of the 1980s: the rise of house and techno music in the iconic cities of Chicago and Detroit.

In a spirit that is both celebratory and solemn, this live performance invites the audience to dance without ever needing to move.

Drawing from the cultural roots of house and techno, the project also places great emphasis on the concept of aesthetic boundary-pushing. Midnight at the club champions unclassifiable music, offering an open invitation to the fringes of alternative and unconventional electronic music: acousmatic, IDM, ambient and hyperpop.

Written in memory of the great musician and electronic pianist Eugenio Giordani, Simple_test is divided into “rooms” that host materials and gestures of various categories, which recur cyclically by typology, varying in content each time. The materials include brief quotes from at least six of Giordani’s electroacoustic works and other material, such as audio in C-Pong.csd format, musical performances, or human footsteps drawn from a hypothetical library, real and unpublished, created by the artist. The original work was conceived for 3D reproduction within a Sonosphere.

Luce Matino’s sole sources are video and audio materials that were captured during a seven-minute walk through the northern Italian city of Cuneo on a July morning in 2025. Initially I didn’t respond enthusiastically to the materials; perhaps I was anticipating more overtly touristic results. But when I returned to them many months later, I began to be drawn to details in the light and shadow cast through the Venetian arcades, the sense of purpose in people beginning their day and the sense of location imbued by street architecture and graphics. The washy soundscape captured was similarly initially unyielding, but as I listened closer, I found nuggets of interest in the subtly shifting acoustics as I moved in and out of the arcades as well as lurking features such as the clatter of coffee cups, energetic chatter, a passing sports car, a musical jingle played through a street PA system accompanied by a petrol generator and the screeching of the wheels on a delivery trolley. The subtle details in the video I was noticing could be brought to the fore by time-stretching so that the eye can explore the scene at leisure. Careful colour grading brought out the light and shadow. Audio events are not synced with their sources and in fact I wanted to reflect the disordered, overlapping and looping manner in which memory functions. A key technique applied to the audio was to reverberate it with itself and often using an approach in which the window of the reverberant impulse is moving along with the source to be reverberated. This approach was also manipulated so that some of the sound features described above – bells, screeches, bangs, speech fragments and melody – could function as musical motifs as they periodically recur, albeit in slightly transformed versions, each time.

Dedicated to Saint Cecilia, this piece draws inspiration from Raphael’s The Ecstasy of Saint Cecilia, contrasting the destruction of worldly instruments with the divine vision of martyrdom. The work reworks the Gregorian antiphon Cantantibus organis across ito four continuous sections, utilizing positional rhythmic notation with a complex, Lachenmann-inspired graphic score. Central to the composition is the use of Csound, which serves as a vital engine for intricate electronic processing. By enabling bespoke spectral delays and specialized reverberation effects, Csound facilitates a deep integration between real-time synthesis and fixed media. This digital architecture complements the guitar’s extensive timbral palette, encompassing harmonics, bowing, percussive strikes, and innovative bottleneck techniques.

Paranoise is a performance for analog synthesizer and Csound live processing, built around an accumulative feedback system that continuously transforms itself. Noise is not added material here but the system’s constitutive condition: what functions as signal at one point in the loop returns as interference at another, with the performer acting as the main interference source. Serge-type function generators feed three nested feedback loops that amplify their own irregularities. A parallel accumulation process gathers small grains of sound by amplitude and “noisiness”, reordering them in a buffer. Once the buffer fills it may at any moment begin to self-destruct — searching for “pure” sounds that, by construction, are not there. Paranoise ends in this failure: the impossibility of separating noise from its other.

fleeting experience (2025) – electroacoustic – fixed media – ambisonic.

Respiri, silenzio: soglie proposes a model of interplay that departs from conventional gesture–response paradigms to explore the parallel coexistence of two heterogeneous respiratory regimes: the human breath (a low-register wind instrument) and a digitally generated respiratory process based on real-time physical modelling. This digital breath is conceived neither as an effect nor as an extension of instrumental gesture, but as an autonomous respiratory process endowed with cyclic dynamics, internal state, temporality, and inertia. From this forced cohabitation emerge actions that do not produce immediate perceptible responses, but operate through accumulation and threshold-crossing, feeding the slow state variables of the system. As accumulation increases, the digital breath crosses continuous thresholds that trigger irreversible transformations—metallicity, spectral densification, and rarefaction—manifesting as a spiral trajectory that progressively makes coexistence with the performer’s gesture impossible. Interplay unfolds as a process of asymmetric co-evolution operating across different temporal scales. The performer acts within conditions progressively redefined by the evolving state of the system, while the digital respiratory flow continues to transform through temporal persistence and spatial expansion. Rather than unfolding through a predetermined succession of events, musical form emerges from the continuous evolution of the system’s state through accumulation and irreversible transformation. In this respect, the work extends a process-oriented lineage of composition — associated with Gottfried Michael Koenig and James Tenney — into the domain of physiological rather than sonic material.

“Xoaniña” is the galician word for ladybug, and this piece of music portrays a part of the cycle of life of this beautiful and intriguing insect. The sketches of the nodes of the matrices of my studies on scanned synthesis reminded me of the dark spots on their shells. The piece was written entirely in Csound, using the CsoundQt IDE, and uses no external samples. It uses exclusively scanned synthesis instruments which are then recorded on the fly onto f-tables to be stretched, looped and granulated. It has been written for a quadraphonic system (4.0).

Interconnected devices open immediate access, where everything is within reach, visible, audible, perfectly calibrated to us. Targeted content, direct sounds, lights that carve out a tailor-made space. A precise point, designed to be perfect. So perfect as to exclude us from the rest. What should unite instead contracts into fragments, individual experiences, enclosed within the boundary of a device. Perhaps there is something that no one can perceive alone.

This work presents a real-time Csound instrument controlled through a machine-learned timbre space. A 57-parameter synthesis engine is sampled, analyzed with spectral descriptors, compressed into a three-dimensional latent representation, and mapped back to Csound parameters for performance. The piece explores continuous transformations between noise, inharmonic resonance, saturation, and granular texture, treating timbre as both musical material and control structure.

Persona Torrent is a performative piece for three speaking performers and three-channel fixed-media electronics.. The performers speak tutorial-like texts that promise fast transformation: learn it fast, fix it fast, become it fast. Their speech is controlled by notification sounds from the electronics, which trigger them to speak, stop, freeze, or repeat fragments of text. What begins as friendly and familiar online language gradually becomes fragmented, compulsive, and overloaded. The piece ends in a collapse of speech, notifications, and pink noise.

This composition presents a creative exploration of human materiality and resilience, inspired by Yuval Harari’s The Ultimate Experience and Joseph Heller’s Catch-22. By leveraging DSP via Csound and SuperCollider, the work integrates electronic and acoustic soundscapes to mimic the flexible, cartilaginous nature of the human body. The piece investigates the extreme objectification of the human form within the machinery of war, utilizing fundamental musical parameters—rhythm, noise, and sound masses—to reflect profound contingency and unconscious orchestration. Spatialization techniques, specifically asymmetrical panning, are employed to aurally depict the physical dichotomy of war’s toll, translating visual fragmentation into an immersive acoustic experience. Structurally atomized sound particles intentionally disrupt the listener, creating a psychological friction that is subsequently resolved through unexpected silence and pure sonic matter. Ultimately, the composition transcends the notion that “Man was matter”, asserting the dignity, fragility, and enduring elasticity of the human spirit amidst conflict.

“Don’t Say Cats” is a live performance for resonating box and live electronics, built on the Spatial Agents instrument, in which the timbre of a performed gesture governs its position in space. Five piezoelectric contact microphones mounted on a hand-built resonating box feed a single Csound engine hosted inside Max through the csound6~ object. Each input is reduced in real time to five audio descriptors (short-term energy, transient density, spectral centroid, spectral spread and spectral flux) that drive five autonomous “agents” through a second-order Ambisonic field, decoded to an eight-loudspeaker horizontal ring or to a file-free binaural render. Spectral stillness and purity pull a source toward the centre, while spectral change and noisiness push it to the periphery, so that a sustained sine tone is heard as a still central point and an evolving noisy texture as a fast peripheral orbit. The decoder is written as user-defined opcodes in the ACN/SN3D convention up to third order and can be rebuilt at run time from a JSON description of the loudspeaker layout. The system is functional and validated.

This piece explores the morphological transformation of the human voice into instrumental timbres, centering on the concept of sonic metamorphosis as a narrative adventure. All sounds are synthesized and simulated using Csound, integrated with recorded acoustic materials. Frequency and harmonic analyses inform the synthesis process, enabling a seamless continuum between vocal and instrumental identities. The work investigates perceptual thresholds of timbral recognition, inviting listeners on an immersive journey through gradually shifting sonic landscapes.

Postcard from Sertões is an experimental performance project that establishes a dialogue between Almeida Prado’s solo viola work Sertões (1978) and the compositional approach of James Tenney’s Postal Pieces (1971). The project explores cultural hybridity by fusing the technical-aesthetic references of the Brazilian rabeca with contemporary electroacoustic practices. Methodologically, the performance distills Prado’s “transtonal system” [1] into four essential sound blocks, reinterpreted as clangs, primary musical units of unitariness and segregation [2]. These blocks are manipulated in real-time using Csound, where four custom-designed looper tracks process the instrumental input to evoke memories from the symbolic territory of the Brazilian sertão. The result is a “hybrid listening zone” that investigates the phenomenal nature of sound through timbral experimentation and real-time algorithmic interaction.

This composition is born from a thought on the essence of the movement and tries to explore the perceptive possibilities of different types of dynamism. Sound is naturally perceived as a dynamic material and the space in which it is diffused can be inclined to favor the tendency to a perceptive characteristic rather than another. Through the management of this inclination, it is possible not only to organize the direction of sound, but also to modify its timbre characteristics in relation to time.

Lumina is an autonomous five-minute stereo fixed-media composition developed from materials created for IHM, a larger work for four cellos and live electronics based on three historical operatic representations of Dido by Henry Purcell, Domenico Sarro, and Niccolò Piccinni. The composition investigates the persistence of vocal identity after fragmentation. A corpus of recorded vocal material is segmented into syllabic units through a Praat-assisted procedure. Each fragment is analysed using MFCC data and eight spectral descriptors: centroid, crest, flatness, kurtosis, roll-off, skewness, slope, and spread. The resulting feature vectors define a multidimensional timbral space in which constrained stochastic paths generate sequences according to spectral proximity rather than textual syntax or melodic continuity. These sequences are articulated through a hierarchical temporal structure derived from interacting periodic cycles and rendered in Csound as dense streams of overlapping, windowed sampling events. The resulting form moves continuously between partially recognisable vocal traces and more fused, quasi-instrumental textures. The stereo field contributes to this process by differentiating simultaneous layers, shaping transitions, and controlling the perceived convergence or separation of the materials.

Csound EEG Duet #2 is a live electroacoustic work for two performers, each wearing a Muse Athena EEG headband and streaming Delta, Theta, Alpha, Beta, and Gamma band data via OSC into Csound from four sensors (TP9, AF7, AF8, TP10). The four streams per band most often generate four independent voices or four-part harmonies; in solo passages, they instead shape the detuning, mix, and filtering of a single four-oscillator instrument. MIDI keyboards and faders let each performer trigger, gate, and transpose the sound, so the EEG data becomes audible only when a key is pressed, at a rate the player sets. Across eight differently mapped Csound orchestras, the piece moves between unquantized harmonies drawn directly from the fluctuating brainwave data and more focused, quantized, reharmonizing textures. Each performer runs The Csound Brain Lab, which hosts the EEG connection and projects EEG-responsive, audio-reactive visuals synchronized to the selected orchestra. Over 8 minutes, the performance moves from raw, exotic harmonic material toward focused chordal writing and a brief melodic solo, then dissolves back into the unpredictability of the original data streams.

All the sounds that make up this composition come from the processing of a short vocal sample “Sta” (0.371 seconds). My goal is to generate from this small sample both sounds with an unmistakable connection to the source and sounds that differ from its origin (bells, percussion sounds, abstract sounds, etc.). I used all these generated sounds to create a true musical discourse, supported by the following sounds processing techniques:

• Morphing;
• Time and spectral stretching;
• Formant synthesis;
• Harmonics used and processed through additive synthesis and frequency modulation;
• Distortions;
• Transpotition and superimpositions;

These sounds are generated using Csound software on a Linux platform.

What does it mean to live inside an equation that has no solution? Symbol X: Six Unbalanced Equations for Solo Violin and Live Electronics is a work for live violin and live-processed electronics in six scenes, each built around a mathematical symbol that portrays the problems and dilemmas women face in different social contexts. The electronics are driven entirely by a wrist-worn IMU controller that the composer built and wears during performance, routing gesture data into Csound in real time; there is no fixed medium and every sound is generated live. Beginning with the undefined equation of 1/0 and ending in the self-evident simplicity of 1+1=2, the piece traces a path from the unease of being watched toward a temporary space of equality.

Written on the occasion of the centenary of the death of Giacomo Puccini, Butterfly Effect is a composition that revolves around quotations of themes drawn from Puccini’s opera, reworked, assembled, and shaped by a simple question: what would have happened if Madama Butterfly had not taken her own life? This hypothesis leads us to reflect on the potential ramifications and the different directions that the story of the young Japanese geisha might have taken if its main events had been altered. By imagining an alternative ending in which Butterfly chooses to face reality and pursue a new path in life, the work confronts us with questions such as: how would her relationship with Pinkerton have developed? How would it have influenced her relationship with her son? And what impact would it have had on her experience of migration and adaptation to a foreign culture? Without the tragic choice of suicide, the plot of Madama Butterfly could have taken an unexpected turn, opening up new scenarios and conflicts. These hypotheses—explored through the theory of the so-called butterfly effect—invite us to consider the infinite possibilities of an alternative narrative, while at the same time preserving the dramatic power and deep reflection offered by Madame Butterfly’s tragic decision.

Echoes of the Unseens is an electroacoustic co-composition realized through a hybrid duo performance system that combines live coding with analog electronics.

The piece gives rise to sonic entities through real-time gesture, translating an industrial nightscape—a threshold between earth and concrete—into a shifting sound-image. Its dramaturgy draws on Pierre Henry’s Entités, featuring two protagonists engaged not in strict counterpoint but in a dialogue closer to sonic gesture. The urban environment and its nocturnal mystery are informed by Charles Ives’s Central Park in the Dark, while the visual imagination of Edward Hopper shapes the treatment of solitude and stillness within a vast, largely static soundscape.

Positioned between solitary performance and the evocation of a plastic tableau, the piece stands as a hymn to minority, to the unseen, to those kept hidden.

Transmutational Interplay for Piano and CsoundLive is a solo piano piece reshaped live by a second performer processing the piano signal in real time on CsoundLive, a browser-based Csound 7 performance environment built around 22 time-based and 12 spectral effects, Muse2 EEG and MIDI-mapped controls, and an internal live-input, sampler, and player mixing architecture. No sampled or pre-recorded material is used at any point; every sound in performance is the piano, reshaped as it is played. The accompanist draws on seven of CsoundLive’s time-based effects: a flanger, a rhythmic gate, a resonator, a pitch harmonizer, a shimmer reverb, a real-time granular processor, and a slow spectral “accordionizer” that stretches and compresses the piano’s harmonic content the way a bellows moves air, bringing each in and out across a single emotional arc: an unsettled, searching opening; a harder passage of real hesitation; the gradual arrival of warmth and reassurance; and a more open, hopeful close, voiced as though the performer’s present self were speaking back to an earlier, less certain one. The two performers work like a jazz duo: the pianist states the piece’s motives, and the accompanist answers, stretches, and leans into them live, so the balance of the seven effects shifts from one performance to the next even as the piano’s material stays fixed.

The Electropoem series is a collection of multichannel compositions that explore the connections between sound, space, and poetic expression. Each piece focuses on the spatial placement and movement of sound to create an engaging listening experience where the balance between sound and silence encourages attentive listening.

P l e n u m is a series of pieces devoted to the exploration of site- and instrument-specific constellations through electronics realized exclusively in Csound. The series investigates the phantasm of the organ’s full presence—a sonic effect traditionally achieved, in its ideal form, through the plenum: the organ registration in which several, potentially all, aliquot stops are added to the principal.

For P l e n u m (Groot Auditorium Leiden) I recorded several sound combinations of the 1998 Flentrop organ at Groot Auditorium, an instrument built after 17th-century Dutch models. The compositional process involves creating a plurality of microtonally shifted textures in which the recorded material is transposed by gradually increasing factors and enriched through multiple artificial plenum gradations. The aliquot strata are algorithmically generated to replicate the mixture structure of the Flentrop organ. At the formal level, echo sections play a key role, as they do in several Fantasias by Jan Pieterszoon Sweelinck.

In the co-presence of multiple organs within the same space, the question of the polyphony of space becomes immediately striking. Yet even within the spatiality of a single organ, the multiplicity of sound sources remains a crucial issue. Polyphony is a fundamental category of organ music; organ placement is an extremely complex subject, and the question of sound projection in the electronic realm is itself immense. A kind of mise en abyme therefore emerges in the relationship between organs, electronics, and spaces. I conceive my P l e n u m series as a contribution to this exploration.

The piece In memoria di Giuseppe Ganduscio is a tribute to a relevant but little-known Sicilian ethnomusicologist, poet, and political figure. Ganduscio (1925–1963) is known to me through a recording of an interview that Luciano Berio made in 1962, together with Cathy Berberian, Roberto Leydi, and others. On that occasion, Ganduscio sang and explained his personal criteria for preserving and transmitting Sicilian oral-tradition music. There are cases of literal preservation and transmission, but there is also, for example, a blessing song that he transformed into a political challenge, with the words “amo a canciari ’sta sorti mischina” (“we have to change this miserable fate”).

Here I sample his late song and this spoken text—his tenor voice alone being already evocative—and I transform the text myself, separating the text from its melodic contour and also applying the spoken text to the melodic contour of the song. Through LPC (Linear Predictive Coding), I use time-stretching, freezing, and the extraction of only the noisy or vocal component, as well as formant shifting. This can change, in a striking way, the perceived age of the virtual singer, from a child to an aged and decrepit voice. After a monophonic freeze on the upbeat of the melody, the spectral components are scattered through space in a powerful dynamic and spatial explosion.

The piece is dedicated to Luigi Pestalozza, a great figure in connecting the manifold social levels of music and music research.

As the title might suggest, this electroacoustic composition’s sources (street sounds, museum sounds, interior sounds, etc.) were captured while the composer was in Alghero, Sardinia, while enduring a fever. The composition’s collaging and mixing was completed entirely within Csound. The composer’s Csound instruments and opcodes automate and randomize how the field recordings are sampled and transformed.

This piece was composed imagining the neural path of an event. The information is initially perceived by the receptor organs and elaborated in a first neural stage. Subsequently the event is compared with the memory, in order to obtain an elaboration of the information coherent with the personal life experience. The brain data becomes more and more abstract and create the meaning of the information, shaping our thoughts, our impressions. In this way we create ourselves through an abstract genesis. The sound events in the composition are always related to the previous sounds and have been developed following an evolutionary process. The spatialization has been treated as a projection of the internal sound space, paying particular attention to the distribution of the spectrum, which aims at a sound perception in an increasingly intimate environment.

This proposal presents “Flourishing Blossoms Like a Dream”, a fixed-media audiovisual electroacoustic composition scored for 4-channel quadraphonic audio and synchronized digital video. The work explores the aesthetic dichotomy between ephemeral worldly prosperity and the enduring tranquility of memory through a tightly coupled hierarchical mapping between sound and image. Technically, the electronic audio layer blends pointillistic digital synthesis with granularly processed acoustic wind chime samples. Spatialization is driven entirely by a custom-developed suite of eight algorithmic quadraphonic panning plugins implemented natively within Csound. These algorithms—ranging from orbital trajectories to flocking systems—dynamically distribute the sonic textures across a four-channel speaker array to create a deeply immersive acoustic landscape that cross-modulates with the unfolding visual narratives.

Extrapolation de la campagne is a research and electroacoustic music project in which the concepts of space and nature lie at the heart of the musical thinking.

In this project, musical space is treated as a fundamental parameter, alongside pitch, timbre, intensity, and duration. From conception to realization, Extrapolation de la campagne was initially produced using a three-dimensional sound spatialization system based on ambisonic technology (3rd order HOA) and a network of twenty-four loudspeakers. To ensure the work’s longevity and wider artistic dissemination, a stereophonic version was subsequently produced.

From a poetic perspective, the project is motivated by an imaginative, artistic and musical exploration of acoustic spaces in natural environments. Indeed, one of the main aesthetic challenges of this new project is to imagine, from the recording of natural soundscapes, new musical spaces that seem close to nature, but that sound synthesis metamorphoses into another field of unheard perceptions. In this sense, Extrapolation de la campagne can ultimately be considered a hybrid and surrealist composition in which electronic sounds musically blend with field recordings made in rural area.

Ton-Leben can be translated to English as either Tone-Life or Tone-Lives. It works with a reduced constellation of (almost) one pitch only and aims to discover and walk through different stages, gestures, energies, as part of this tone’s experience(s) of life.

Hardly anywhere are the processes of formation and dissolution of a landscape as clearly visible as in Iceland. Heat and matter there shape the landscape, just as frost, water, and wind decompose it and transform it into something new. The laws of gravity and inertia of mass, in turn, have an ordering effect on what has previously been dismantled. Disintegration and the breaking up of rigid structures are not something negative here, but rather a completely natural part of a permanent, everyday process of transformation. The laws of formation of the landscape are reflected in “sonic landscapes” that – created purely from electronic sounds – develop around the listener. The electronic sounds, spatialized using the Ambisonic method, evoke landscapes of the soul that have their physical, real counterparts high in the north.

This fixed-media work is based on a compositional framework that foregrounds the interaction between sound materials and musical gestures. Materials are conceived as sound entities carrying latent structural relationships, which may remain compositional implicit while becoming perceptually manifest through their audible effects. Gestures function as temporal processes that articulate these relationships and shape the listener’s perception. The work integrates concrete and synthetic, tonic and non-pitched sound materials, organized through strategies of contrast and balance. Gestural processes are formalized as parametric sequences generated via a computer algebra system, with sound synthesis realized in Csound. Although produced using digital technologies, the compositional approach aligns with the tape studio tradition, emphasizing detailed post-processing in both the time and frequency domains as an integral part of the compositional process. “Undercurrents” was composed and premiered in 2026 at ICMC, Hamburg.

This composition investigates the notion of technological and computational obsolescence as a lens through which to reflect on human obsolescence, particularly in relation to mechanical, artistic, and intellectual labor. Within a social and economic climate increasingly oriented toward efficiency and automation, human beings are placed in constant competition with artificial systems designed to be faster, tireless, and free from the limitations that characterize human labor. The piece employs the figure of an android reaching the end of its operational life cycle as a central metaphor for this condition. Its final moments of activity are evoked through states resembling confusion, anger, resentment, despair, and acceptance, not as genuine emotions, which a machine is incapable of experiencing, but as the outcome of internal calculations and self-assessments of its own operational status. This process paradoxically humanizes the android, placing it on the same existential plane as the human beings it was designed to replace, and turning it into a metaphor for the condition of contemporary individuals increasingly rendered superfluous within industrial and productive contexts. Compositionally, the work is built primarily through granular synthesis, complemented to a lesser extent by additive and subtractive techniques. Concrete industrial and metallic sound sources are granulated and processed into disjointed, rigid rhythmic structures that evoke a mechanical system in decline, while colder, more abstract synthetic textures suggest a form of artificial consciousness. The piece unfolds as a continuous friction between mechanical rigidity and timbral instability, tracing the progressive deterioration of the android’s physical and cognitive functions.

This fixed-media electroacoustic work is rendered in Csound using an orchestra that generates pipe-organ audio with Modartt’s Organteq plugin and processes it with the Mverb2020 reverberation plugin. Organteq is a physical model of a pipe organ and does not use samples. The entire score is generated and rendered by a single Python script. Bars of music from the musical dice game traditionally attributed to Mozart are played in sequence, but each bar is repeated a randomly chosen number of times, as in Terry Riley’s In C. Every few bars, the music progresses to a randomly selected chord. Periodically, the current chord is used as a pivot for modulation to a new diatonic scale, and the generated notes are quantized to that scale. The Organteq registration is also changed periodically.