PhD Thesis
Instant Harmonies — Enhancing Music with Web-Based Just Intonation Tuning Algorithms
Maynooth University · Supervised by Prof. Joseph Timoney · Expected December 2026
My doctoral research explores how machine learning can enable real-time adaptive tuning of musical audio — specifically, how to shift music from the standard 12-tone equal temperament (12-TET) system to Just Intonation, a tuning system based on naturally occurring harmonic ratios that many musicians consider more consonant and expressive.
The core challenge is speed: retuning must happen in real time, with latency low enough that musicians can play and hear corrected audio without perceptible delay. To achieve this, I developed ensemble ML prediction models combined with n-gram fingerprinting techniques that anticipate upcoming notes and pre-compute tuning adjustments before they are needed.
Low-Latency Algorithm Design
Designing algorithms that operate within strict real-time constraints for audio processing, where even milliseconds of delay are perceptible to users.
Ensemble ML Prediction
Combining multiple predictive models to achieve higher accuracy in musical note prediction, enabling pre-computation of tuning corrections.
N-gram Fingerprinting
Adapting natural language processing techniques to musical sequences, identifying patterns in note progressions to predict what comes next.
Production ML Deployment
Taking ML models from research notebooks to production systems — real-time inference, WebSocket communication, and browser-based audio processing.
A production-grade system for real-time music analysis and adaptive audio tuning. Instant Harmonies is the practical realisation of my PhD research — a fully deployed, publicly accessible web application that demonstrates ensemble ML prediction combined with ultra-low-latency audio processing.
The system listens to MIDI input in real time, predicts the harmonic context using ensemble ML models, and applies Just Intonation tuning corrections with imperceptible latency. It was built end-to-end by a single engineer, from the ML pipeline to the browser-based frontend.
An AI-powered music coaching application that provides personalised, intelligent feedback to everyday musicians. Unlike existing apps that only detect pitch and timing accuracy, HarmoniAI combines real-time audio analysis with LLM-powered reasoning to deliver qualitative coaching — explaining why something sounds a certain way and how to improve.
Being developed under Intonation Labs Pte. Ltd. Launching as a web application first, targeting the Southeast Asian market.
MIDI Adaptive Tuning Strategies by Means of AI-Based Struck-String Interaction in Ubimus
Su, R., Timoney, J., & Keller, D. · Ubimus Journal · August 2025
Just Intonation and Inharmonicity in Struck-String Interaction
Su, R., Timoney, J., & Keller, D. · Proceedings of the Ubiquitous Music Symposium 2024 (UbiMus 2024) · October 2024