HawkEars: Jan Huus’s Lecture on his AI birding tool

Left to right, Marian McLennan, Jan Huus, and Hilde Huus

On November 3, members were treated to an excellent talk on AI and birding. Mr. Huus has an extensive background in computer science and as an avid birder.

With three other scientists, Jan has developed and published a scientific paper, through the University of Alberta, entitled HawkEars: - a regional, high-performance acoustic classifier. To the layman’s ear, this sounds very complicated, which it is, but Jan explained the concept and application in a way that we could all understand.

Essentially, the team accessed 425,000 training samples from various sources such as the MacAulay Library at the Cornell Lab of Ornithology and Xeno-canto, and others. They also used Google Audioset to gain samples of non-avian sounds, including mammals (domestic dogs, coyotes, wolves, and squirrels).

From this enormous sample size, they selected 50 species of birds and randomly selected about 200 recordings per species. These recordings were used to tune an initial AI model, which was later extended to 334 bird species and 13 amphibian species. It was interesting to learn that not all birds vocalize. For example turkey vultures do not, and some species, such as mourning doves, can be identified by the sound from their wing strokes.

For active birders, the HawkEars technology can be applied on a small scale by employing listening devices that can be attached to trees, for example. These machines can record and process the spectrograms of numerous types of birds which leads to accurate and rapid recognition of species present in a particular geographic area in a way that traditional birders could only imagine! Many of us who are unfamiliar with, and possibly uncomfortable with the development of Artificial Intelligence, may find it reassuring to learn about a positive and beneficial application of this advanced technology.