George Howard Redpath

FPGA Modular Synthesizer Platform · SystemVerilog

An FPGA-based Eurorack modular synthesizer platform with a novel approach to total recall — addressing one of the longest-standing limitations of modular synthesis. Custom hardware and FPGA architecture designed from the ground up, combining analog signal path design with digital intelligence.

Full 3U 104HP Eurorack case designed from scratch. Case and custom knobs 3D modelled in OpenSCAD and 3D printed. Custom front panels with silkscreen artwork, Thonkiconn jacks, and panel-mounted LEDs for visual feedback. The case is a purpose-built enclosure with a custom-designed PSU (bus board, ribbon cables, regulated ±12V and +5V rails) designed to house the complete system as a self-contained instrument. Every aspect of the physical design — panel layout, knob spacing, ergonomics — is considered alongside the electronics.

Multi-layer PCB design with mixed-signal layout: analog signal paths kept separate from digital FPGA logic, dedicated ground planes, and careful power regulation. The analog front end uses precision op-amps and DC-coupled I/O for full Eurorack voltage range (±12V rail, ±5V signal). Custom codec/DAC stage for high-fidelity audio conversion between the analog domain and FPGA fabric.

GPU-Accelerated 3D Magnetic Tape Simulation · C++/Metal

Real-time magnetic tape saturation emulator using GPU-accelerated 3D simulation. Models the nonlinear hysteresis of analog tape across a 65,536-point 3D grid (64×256×4 — representing 6mm tape width, 5cm buffer, and 10μm oxide depth) with spatial diffusion using a 7-point stencil. Jiles-Atherton hysteresis model runs per-grid-point with Langevin function anhysteretic magnetisation. Includes tape cutting, splicing, looping, variable speed playback, and degradation effects.

Waveguide Uilleann Pipes Synthesizer · C++/Metal

Physics-based virtual instrument modelling Irish uilleann pipes using digital waveguide synthesis with Metal GPU acceleration. Complete paraphonic pipe set: chanter (9-hole melody), 3 drones (bass/baritone/tenor), and 3 regulators (keyed chords). Nonlinear reed excitation via Bernoulli equation, Kelly-Lochbaum scattering junctions at tone holes, five bore profiles (cylindrical, conical, hyperbolic, parabolic, stepped), material damping, and bell radiation modelling. Metal compute shaders process 7 waveguide pairs + 9 tone holes + 13 regulator keys at 96kHz.

Virtual Analog + Neural Hybrid Architecture · C++/Python/ONNX

JUCE audio plugin emulating the EMS VCS3/Synthi diode ladder filter using a hybrid architecture: physics-based virtual analog modelling with neural network enhancement for authentic nonlinear characteristics. The VA model implements a 4-pole ladder topology with bilinear transform and tanh soft saturation. A WaveNet model trained on 750 impulse responses captured from hardware learns only the residual difference between the VA model and real hardware, running via ONNX Runtime. Three filter modes (LP/HP/BP), three slope options (18/24/48 dB/oct). Includes separate Python research environment for DSP prototyping and ML training pipeline.

Lexicon-Inspired Dual-Tank FDN Reverb · C++

Professional-quality algorithmic reverb inspired by the Lexicon 224/480L. Dual 8-delay feedback delay networks with Hadamard orthogonal matrix mixing, RT60-based decay using modified Sabine equation, 4× modulated allpass prediffusion, and 7-tap early reflections. Includes time FX section (delay, shimmer with pitch-shifting, chorus), HRTF crossfading, anti-blowup loop energy monitoring, and safety limiter. Celtic-inspired UI with golden ratio spiral and triskele elements. Evolved from a 24-hour WebAudio/WebMIDI hackathon prototype into a full JUCE plugin.

Extended Karplus-Strong Guitar Synthesizer · Faust

Physics-based plucked string synthesizer implementing Julius O. Smith's Extended Karplus-Strong algorithm. Bidirectional delay line with linear-phase FIR3 damping, pick-position comb filtering, dynamic-level lowpass compensation, and T60 decay modelling. 32-step sequencer with E minor pentatonic arpeggio patterns, stereo LFO panning, and Zita reverb integration. Compiles to Web, JACK, VST, Daisy, and Bela targets.

Physical Model of Granular Collisions · MATLAB

Physical model of grains of sand colliding in space. Simulates the acoustic behaviour of granular particle interactions, synthesising the resulting sounds from first-principles physics rather than sample playback.

Serge Resonant Equaliser for Bela · C++

Implementation of the Serge Resonant Equaliser for the Bela embedded audio platform. The Serge ResEQ is a classic analog module design — a bank of resonant bandpass filters that can be driven into self-oscillation, functioning as both an equaliser and a sound source.

LM13700 OTA Analog Circuit Design

Analog circuit recreation of the OTO Biscuit synthesizer/effects processor using LM13700 operational transconductance amplifiers. Detailed analysis and recreation of the original circuit topology with focus on reproducing the nonlinear characteristics of the OTA-based signal path.

Binaural & Stereo Audio Processing · Harman Corporation

Research and development of binaural and stereo audio processing in collaboration with Harman Corporation. Work on spatial audio rendering techniques for consumer and professional audio applications.

EMS Stockholm + NTNU/AAU Research

Comprehensive body of work in binaural spatial audio. Head-trackable HRTF rendering with real-time adaptation to listener movement (EMS Stockholm). Woodworth spherical head ITD model, Brown-Duda ILD filters, fractional delay interpolation, and HRIR synthesis. MediaPipe-based head tracking with IRCAM-style room acoustics. Multiple implementations spanning research prototypes through to production-grade real-time systems with VST-grade parameter smoothing.

EMS Stockholm — Guest Composer Residency

Neural network models using WaveNet and CNN architectures to emulate analog synthesizer filters. Deep learning captures the nonlinear behaviour of analog circuits, producing convincing digital recreations of analog filter characteristics. Research conducted during a guest composer residency at Elektronmusikstudion (EMS) Stockholm.

Real-Time Emotion Detection → Modular Synthesis · C++/Python

Real-time emotion detection driving modular synthesizer control voltage. JUCE plugin version uses OpenCV face detection, 68-point landmark tracking, and Mini-Xception CNN via ONNX Runtime to output 7 CV channels, each modulated by a physics model matched to the emotion: Lorenz attractor (angry), Duffing oscillator (disgust), mass-spring-damper (fear), modal resonator (happy), breath model (sad), struck membrane (surprise), damped harmonic (neutral). Companion Python version uses MediaPipe FaceMesh with OSC output to Max/MSP. Optimised for Expert Sleepers ES-9/ES-8 DC-coupled outputs to Eurorack.

Adaptive VR/Mobile Sensory Platform · Unity/Godot/C++

An adaptive platform building empathy between neurodivergent and neurotypical individuals through shared, personalised sensory experiences. Phenotype-based audio adaptation using Dunn's Sensory Processing Framework. Integrates multiple DSP research modules: bioacoustic synthesis (cricket stridulation, avian syrinx), physical guitar, teorainn reverb, Physical Tape saturation, RTNeural inference, and binaural spatialisation. Generative music system using Markov melody generation and Eno-style coprime loop techniques. VR version (Unity/Meta Quest 3) and mobile version (Godot 4 with C++ GDExtension audio engine).

Polychromatic Microtonal Isomorphic Controller · Python

Microtonal isomorphic controller for the ERAE II hexagonal touch surface. Supports arbitrary N-TET tuning systems (12, 19, 31, 53, 106-TET), multiple isomorphic layouts (Wicki-Hayden, Tonal Plexus), and colour schemes (hue wheel, polychromatic harmonic series, chromaplane). MPE MIDI and OSC output. Direct hardware integration via ERAE II SysEx protocol for real-time LED feedback on the touch surface.

Proteus-Style Exploration Game · Godot 4.5

3D exploration game with generative bioacoustic music. Procedurally generated environments where the soundtrack emerges from simulated ecosystems — the music is the world.

Live Audio Visualisation Dashboard · Python

Real-time audio visualisation dashboard for the terminal featuring VU meters, oscilloscope display, and system metrics. Live audio analysis rendered directly in the terminal.

OTO Machines · Plinky · C++/JUCE

Suite of JUCE-based audio plugins for managing presets and interfacing with hardware synthesizers via MIDI SysEx. OTOInterface for OTO Machines devices (Biscuit, BIM, BAM, Boum) with Pro upgrade path, PlinkyInterface for the Plinky synthesizer, PlinkyBridge for anti-aliasing Plinky USB audio, and Plinky-VST for direct DAW integration.

ML Training Data Pipeline · C++/JUCE

JUCE audio plugin for capturing impulse responses from analog filters for machine learning training. Automated sweep generation, capture, and cataloguing pipeline that produced the 750-file training dataset used for the MLEMS Synthi Filter neural model.

Thread-Safe Remote Script · Python

Thread-safe Python Remote Script for Ableton Live exposing 220+ LiveAPI tools for external automation and control. Enables programmatic access to Live's internal API for custom control surfaces, automation scripts, and integration workflows.

Sequentix Cirklon Instrument Definitions

Instrument definition files (.cki) for the Sequentix Cirklon hardware sequencer. Comprehensive MIDI CC mappings for hardware synthesizers including Plinky (synth + sampler modes) enabling full parameter control from the Cirklon.

VCO Tuning Utility · C++ · macOS Fork

Maintained fork of the VCOTuner utility updated to work on modern macOS. Precision tuning tool for voltage-controlled oscillators in analog synthesizers.

Commercial Distribution System · Python

Licensing and activation infrastructure for commercial audio plugin distribution. Handles license generation, validation, and management across the plugin suite.

USB Audio Class 2.0 on STM32L476 · C

USB Audio Class 2.0 implementation for STM32L476 using TinyUSB. 48kHz/16-bit stereo streaming in adaptive mode. Ring buffer in dedicated SRAM2 bank, dual PLL clock architecture (PLLSAI1 for ADC, PLLSAI2 for audio I/O), DMA integration for codec/SAI.

Theory to Real-Time Implementation · Python + C++

End-to-end physical modelling of guitar acoustics: from theoretical analysis and parameter extraction in Python to real-time JUCE C++ plugin implementation. Covers string vibration modelling, body resonance, and coupling between physical subsystems.

Gaulossen Nature Reserve · BirdNET · Praven Pro

Field-based bioacoustic monitoring and analysis. Gaulossen Nature Reserve acoustic monitoring study. BirdNET batch analysis pipeline with YIN F0 estimation and LPC formant analysis. Praven Pro — a professional bioacoustics workflow combining Python automation with Raven Pro for BirdNET batch processing with F0/formant analysis.

Signal Processing · Simulation · Psychoacoustics

Research and coursework spanning acoustic signal processing, room acoustics, and digital simulation methods. Includes 2D Transmission Line Matrix pipe simulation, 1D digital waveguide synthesis of musical instruments, music box DSP analysis (multi-method F0 consensus with SHS, YIN, cepstrum), audio denoising, MFCC extraction, harmonic analysis, and psychoacoustic modelling. Interactive study notebooks with plug-in calculator sections and visual concept maps.

Music Information Retrieval Tutorials · Python

Music Information Retrieval tutorials with librosa. Google Colab notebooks covering beat tracking, harmonic-percussive source separation, MFCC, chromagram analysis, and time-frequency visualisation for audio signal processing education.

Professional Video Denoising · Python / Google Colab

Professional video denoising tool with an interactive Google Colab interface. Designed for film and media applications where clean footage is critical.

AI-Powered Photo Organisation · JavaScript

Intelligent photo organisation tool that sorts photos by date, categorises content using AI, and manages projects. Available as both desktop app and web app.

Academic Publishing Pipeline · Python

Tool for converting Jupyter notebooks to academic LaTeX papers with Overleaf and GitHub integration. Streamlines the path from research notebook to publication-ready manuscript.

Arqiva

Professional RF engineering work covering broadcast transmission infrastructure. RF system design, signal processing, transmission planning, and field engineering. This foundational experience in signal processing and systems engineering directly informs current DSP and FPGA audio work.

Elektronmusikstudion (EMS) Stockholm · LaTeX

Proposal for artist residency at EMS Stockholm using machine learning for the historic preservation of rare analog synthesizers — specifically the EMS VCS3, Synthi A, Synthi AKS, and Synthi 100. Only ~900 VCS3 units were ever manufactured; these instruments are rare, fragile, and geographically limited. Traditional software emulations fall short.

The proposed solution is a Neural Virtual Analog Hybrid Filter — a parallel architecture combining traditional VA DSP with a lightweight neural network (~50KB) trained on real analog hardware input/output pairs, running sample-by-sample in real time. A bespoke capture system enables one neural model per instrument component per day. The existing Fan synthesizer was built from just 3 days of previous access to EMS Stockholm.

Paper submission in preparation for the International Conference on New Interfaces for Musical Expression (NIME) 2026.