A listening tool that uses on-device machine learning to map the sounds of a place over time.
- Client
- Studio initiative
- Year
- 2026
- Disciplines
- Artificial Intelligence, Software & Applications, Research & Innovation
- Industry
- Technology
- Deliverables
Listening, not recording
Audio is analysed in short windows and discarded instantly. What remains is a description — “blackbird, 82% sure, 06:12” — never the sound itself.
The most important number on screen is how unsure the model is.
Project notes
Overview
Murmur is a self-initiated studio project exploring what AI can do for attention rather than against it. Leave a phone on a windowsill; come back to a map of the birds, rain, traffic and voices it heard.
Challenge
Making machine listening understandable and trustworthy — with no audio ever leaving the device.
Approach
A small on-device classifier, a visual language built from sound waves, and an interface that always shows its uncertainty.
Process
Field recordings in twelve locations, model evaluation sessions with sound ecologists, and many paper prototypes of the “listening diary”.
Outcome
A working prototype app, an open dataset card and a short design paper on honest confidence displays.
- Research
- Hana Ito
- ML engineering
- Jonas Varga
- Design
- Capsul Lab
Next project →
Salt & Stone
