An annotated set of audio recordings of birds in Western Canada

These datasets were used to evaluate the real-world performance of our classifier, HawkEars, as described in the associated manuscript. We processed the same datasets with BirdNET and Perch to compare the performance of HawkEars relative to other available community classifiers. Two datasets are provided, one to evaluate community composition and one to assess vocal activity. All recordings were collected with Song Meters (Wildlife Acoustics SM2, SM2+, SM3, SM4 units) between the years 2013 and 2019. The recordings were collected both during the day, primarily at dawn, and at night to ensure representation by both diurnal and nocturnal bird species. Similarly, most recordings were collected during the breeding songbird season in Canada from mid-May to mid-July with the remaining recordings collected from mid-February to mid-May to capture early breeding resident species like owls.

Passive acoustic monitoring is rapidly emerging as a dominant approach for studying acoustic wildlife, with neural networks used as an increasingly common and promising approach for extracting detections of particular species from acoustic recordings. Existing options for avian classifiers include small custom models for focal species or large models that attempt to classify the entire global avian community, which suggests a possible tradeoff between classifier performance and species coverage. We argue that building domain-specific classifiers for particular geographic regions provides improved performance in exchange for reduced species coverage. Our manuscript presents HawkEars, a regional avian classifier for Canada that includes 314 bird and 13 amphibian species.