Audio Buffer London
Audio Buffer London is our monthly meet-up for audio developers and creative technologists—bridging the gap between academic theory and the gritty reality of releasing professional audio tools.
Led by Dr. Sinan Bokesoy, Audio Buffer is aimed at anyone developing software-hardware based music tools, be they students, composers, academics, and other creative technologists. In a world of online isolation, we provide a physical environment to network, and stress-test ideas.
Audio Buffer 06 will be kindly hosted by Goldsmiths, University of London, during their annual Manifold Art event and exhibition.
"Old, But Not Obsolete" Zero-Inference Generative Systems
Session #06 of Audio Buffer London shifts focus to the foundational algorithms of computational creativity.
In the current era of AI, the conversation is dominated by massive, pre-trained transformer models. Yet, for live generative music, these modern monoliths face a critical flaw: latency. Gradient descent over billions of parameters cannot happen in 5ms.
This brings us to tonight's thesis: Old, but not obsolete. The algorithms that truly drive live, interactive generative performance are rooted in the mid-to-late 20th century—Markov chains, Cellular Automata, Stochastic distributions, and Evolutionary computing. These architectures do not rely on a frozen corpus or massive server farms. They are live, rule-based, on-the-fly processes with near-zero inference time. Because learning and generation happen simultaneously, they are instruments you inhabit rather than oracles you query.
Session #06 will examine four distinct applications built on these classic concepts, proving that foundational models still hold the key to real-time generative sonic art.
Registered attendees will receive a free license of Cosmosf M31.
Presenters
Prof. Matthew Yee-King, "Clockwork" (Markov Chains) & "EvoSynth" (Genetic Algorithms)
Matthew presents two distinct systems spanning discrete logic and evolutionary computing. First, he introduces Clockwork, a custom Markov improviser that relies on explicit, real-time histograms built directly from the performer's input, implementing turn-taking as a physical quantity that charges and drains against a real clock.
He will also present EvoSynth, a system building upon concepts from his pioneering 2000 project AudioServe. EvoSynth uses interactive genetic algorithms to search the space of all possible configurations for FM/AM synthesizer circuits. By cross-breeding small local "islands" of genomes, the system allows users to seamlessly mutate and discover novel waveforms on the fly.
Dr. Sinan Bökesoy "Cosmosf M31" — Stochastic Distributions
Sinan presents his stochastic multi-scale MIDI event generator, Cosmosf M31. Rooted in the Xenakis lineage of aleatoric processes, M31 distributes discrete note events across varying time scales (macro, meso, micro) using a deep palette of stochastic functions. Instead of a pre-trained model guessing the next note, the performer continuously transforms a multi-dimensional parameter space in real time, navigating smoothly between rigidly ordered sequences and complex, disordered states.
Michael Wallis "Cela" — Cellular Automata & Expressive Interfaces
Joining us from QMUL, Michael presents Cela, a MIDI controller designed to bridge the gap between musical skill and instrument playing. At its core is an 8x8 grid of toggles driven by the principles of Conway's Game of Life to autonomously generate evolving melodies and chords. Unlike models relying on vast data histories, Cela's memory is simply the current grid state. Michael will demonstrate how these hyper-local interactions create complex global sequences, while giving the performer deep expressive control through tactile BPM and ADSR parameters, alongside a physical light sensor that dynamically modulates the generated sound qualities.