In the remote reaches of Canada, thousands of recording devices are quietly gathering the sounds of nature—millions of hours of audio documenting birds and amphibians. The Alberta Biodiversity Monitoring Institute alone deploys hundreds of acoustic recording units, or “ARUs”, across Alberta each summer. These recordings they collect are a treasure trove of biodiversity data, but a major hurdle remains: turning sound into usable information at scale.

That’s where HawkEars comes in. Developed by computing scientist Jan Huus with support from the ABMI and Biodiversity Pathways, HawkEars is a deep learning model designed specifically to recognize the calls of 328 Canadian bird species and 13 amphibians. Unlike large global models like BirdNET, which aim to identify thousands of species across diverse regions, HawkEars takes a regional approach. By narrowing its focus and training on strongly labeled, high-quality data from Canadian species, the model achieves much higher accuracy and recall—essential for ecological monitoring where missed detections can mean missed conservation opportunities.

The team recently published a paper introducing HawkEars, both as a classifier for modeling species and as a framework for training new classifiers for targeted objectives. A custom AI-enabled search tool was designed by Jan to sift through recordings and select the best training samples for each species. HawkEars also includes a custom submodel for the low-frequency Ruffed Grouse drumming that otherwise goes unclassified.  When tested against expert-labeled datasets, HawkEars detected an average of two more species per minute than global classifiers, BirdNET and Perch.

The implications for conservation are significant. More accurate species detection feeds directly into better occupancy and density models, enabling researchers to estimate populations and track changes over time. It also enhances biodiversity monitoring, improves the detection of rare and cryptic species, and speeds up ecological assessments. For scientists and land managers, HawkEars offers an open-source, customizable tool that improves efficiency without sacrificing accuracy. And while HawkEars was designed with Canada in mind, the framework is adaptable. Researchers can apply the same methods to build high-performance classifiers for other regions or taxa.

If you’re interested, HawkEars is available through GitHub and the ABMI will be integrating it into the WildTrax platform in the near future to make it even easier for users to apply HawkEars to their own acoustic data!