13th National Seabirds Meeting: DeepGreen & Biphonia Unveil Frugal AI for Shearwater Recognition 
 

DeepGreen and Biphonia are introducing a lightweight AI demonstrator for real-time Shearwater recognition at the 13th National Seabirds Meeting in Bonifacio, Corsica. The solution is built with Eclipse Aidge and is specifically optimized to run on the AudioMoth microcontroller, requiring only about 10kB of RAM—making it ideal for field deployment in severely resource-constrained settings.
Biphonia’s expertise in terrestrial bioacoustics led to practical biodiversity monitoring tools that bridge fundamental research and ecological challenges. This collaboration blends acoustic know-how with frugal technology, delivering a robust, efficient classifier for accurate species identification locally, without reliance on remote servers or the cloud.

An excerpt of the demonstration video presented on this occasion, produced by Marielle Malfante and Louis Lerbourg from CEA-List, offers a glimpse into the system’s deployment and performance under real-world, constrained conditions. Their work highlights the practical feasibility of running AI directly on low-resource hardware in the field.

Special thanks go to Paul Peyret and Juliette Linossier for their contribution to the paper submitted to AAAI  “Smart Passive Acoustic Monitoring: Embedding a Classifier on AudioMoth Microcontroller.” This project demonstrates that powerful conservation technology can also be lightweight, scalable, and energy-efficient.