Research / engineering / creative practice

Sound is a hard problem. Good.

Music technology spans signal processing, hardware, machine learning and creative practice. Our members learn it by building, in small project teams with questions from members, companies, university chairs and artists.

01
Engineer
Classical DSP, real-time systems and hardware, from guitar pedals to chips.
02
Analyse
Transcription, source separation, tagging and search: understanding music that already exists.
03
Represent
How to describe a song for a computer: samples, notes, text or several views at once.
04
Create
Tools that help people make sound, while keeping rhythm, harmony and form coherent.
05
Question
Who controls a tool, what data shaped it, and whose creative intent stays visible.

Published

Papers, explained.

September 2026
A watermark inside the codec.
arXiv · cs.SD
An audio watermark written into a frozen neural codec’s latent space, so it survives the codec resynthesis that erases waveform watermarks.

The real scale

One song, millions of values.

A three-minute recording holds almost eight million samples, and its structure lives on several timescales at once: milliseconds shape timbre, seconds shape notes, minutes shape the song. That makes audio one of the most demanding signals to work with.

01 second: 44,100
03 minutes: ≈ 8M
Timescales: MS → MIN

From the bench