UK researcher helps AI learn a bird’s individual song

Darin McNeil, Ph.D., an assistant professor of wildlife management in the University of Kentucky Martin-Gatton College of Agriculture, Food and Environment, is partnering with the University of Pittsburgh and others to help build an artificial intelligence tool that could give bird researchers a new way to follow one of Appalachia’s most vulnerable songbirds.

The research does not center on a specific bird species but on a specific bird. The project focuses on the cerulean warbler, a small blue bird whose numbers have fallen over time. While there are already apps available to locate a specific species, this project aims to enhance that capability.

“Scientists already have programs, and apps exist that can tell the phone user when they are hearing a cerulean warbler,” McNeil said. “We want to know which cerulean warbler is singing. We don’t just want to identify whether it is a cerulean warbler; we want to know which cerulean warbler it is.”

That distinction could open a new path for conservation research.

For years, scientists who wanted to keep track of individual birds had to catch them, band them with colored leg bands, then return again and again to see which birds came back. That work can be slow, difficult and intrusive, especially for a species like the cerulean warbler, which often stays high in the forest canopy.

McNeil said the new system could allow researchers to identify birds through the songs they already sing, without having to catch them at all.

The idea rests on something most people never notice. To the human ear, one cerulean warbler may sound much like another. A computer, though, can sift through tiny differences in pitch, timing and pattern.

McNeil said the process is not all that different from how people recognize familiar voices.

“This new method would allow us to tell one individual from another based on their song,” McNeil said. “You could identify these individual birds by their voices as well.”

That could give researchers a better sense of what is happening to the birds from year to year. If the same singers return to the same places, scientists can begin to measure survival and site fidelity without relying as heavily on traditional marking methods. They may be able to tell who stayed, who moved and who did not make it back. For a declining species, that kind of information matters, McNeil said. Furthermore, it can help researchers piece together why numbers are dropping and where protection efforts may count most.

The project, led by Ph.D. student Lauren Chronister at the University of Pittsburgh, is still in its early stages. Currently, the team is focused on showing that the concept works.  

McNeil said researchers are using parabolic microphones and other specialized equipment rather than ordinary smartphones. Although such technology could be incorporated into a phone app later, for now, the likely users are scientists in the field, not casual birders.   

“The cerulean warbler is where we’re starting, but it isn’t where this ends,” McNeil said. “If we can learn to recognize individual birds by voice, there’s no reason the same approach couldn’t work for other species that are just as hard to track.” 

The collaboration includes UK, the University of Pittsburgh, Indiana University of Pennsylvania, Massachusetts Department of Conservation and Recreation and Arkansas State University.