EcoEar: edge acoustic AI, proven in the wild
- Client:
- NSW Smart Sensing Network (project initiator)
- Collaborators:
- International Centre for Neuromorphic Systems, Western Sydney University (Dr Saeed Afshar); Biodiversity Monitoring Services; Nuvutech; Neurabuild
The problem
In 2022, wildfires tore through koala habitat across New South Wales. Ecologists needed to find and count survivors across vast, remote terrain. Their only tool was passive recorders that captured everything, leaving researchers to hand-review more than twelve hours of audio a day. Slow work, when conservation could not wait.
Our approach
Neurabuild joined the project with Dr Saeed Afshar and the International Centre for Neuromorphic Systems at Western Sydney University. We worked alongside the researchers to turn their neuromorphic acoustic algorithms into a rugged field device. Architecture led by Dawid Loubser, development led by Graham Withey, hardware with EcoEar engineer Richard Powrie. Three years, three hardware generations: 3D-printed first prototypes, then a redesigned PCB and power system, then a field-ready build. We found the hard problems in the field, including a memory-corruption fault under real conditions, and engineered them out.
Outcome
EcoEar is a sub-500g, IPX7-rated device that runs up to 21 days on four AA batteries and detects koala calls on-device, with no cloud connection. Forty units were deployed in the field in 2024. Instead of sifting twelve hours of audio a day, ecologists retrieve an SD card after a few weeks and receive tagged, GPS-located, timestamped calls. Validated in forests, sanctuaries, and bushland.