Jazz

Computational Jazz Research

Despite being one of the most influential musical traditions of the last century, jazz has received surprisingly little scientific study — largely because improvisation makes it difficult to analyse and because most jazz performances aren’t written down.

This research project tackled these challenges by building a large collection of annotated jazz recordings and using computational methods to understand what makes jazz distinctive. We created a database of 45 hours of jazz performances and used it to:

• Identify musical style: Develop systems that can recognise individual performers based on their musical choices, revealing what actually defines a musician’s distinctive sound
• Decode rhythm and timing: Explore the hidden patterns behind jazz concepts like “swing” and “feel,” connecting scientific analysis to what jazz musicians and listeners have always known intuitively
• Study coordination: Observe how musicians actually listen and respond to each other in real time, even when that coordination is deliberately disrupted
• Generate new music: Create AI models capable of producing jazz in the styles of particular performers or subgenres

This project was completed at the CMS as part of Huw Cheston’s PhD research (2021–2025), funded by a Vice-Chancellor’s Award from the Cambridge Trust. His thesis is available to read online.