From chirps to melodies, the variety of birdsong resembles a grand musical composition.
However, ovenbirds — stout, terrestrial warblers — possess a very limited selection of songs.
The male ovenbird has only one song: a quick tune often referred to as teacher-teacher-teacher.
To the average listener, all ovenbirds appear similar, yet each bird presents its own unique rendition.
Currently, researchers are leveraging AI technology to monitor the wellbeing of ovenbird populations based solely on their signature songs.
Research published this month in the scientific journal Nature Methods utilized AI to recognize individual ovenbirds from audio recordings obtained in Alberta and Pennsylvania.
A singular focus
The ovenbird’s “particular” song list made it an excellent candidate for study, according to lead researcher Sam Lapp, a postdoctoral researcher at the University of Colorado.
“They are quite consistent in performing one specific song,” Lapp stated in an interview with CBC News.
“Many birds can develop a vocal signature during their lives, often including various songs in what we call a bird’s repertoire.
“However, the ovenbird has the advantageous trait of maintaining one very stable song over time, and it only sings that one version.”
Ovenbirds perform their main song in the spring to establish their territory and attract a partner, and this song remains a constant throughout their lives.
Under Lapp’s guidance during his tenure with the Department of Biological Sciences at the University of Pittsburgh, the research team analyzed recordings from nearly 130 locations in Pennsylvania collected over four years.
They trained an AI model to identify features that made each song unique and group similar songs together, assuming each bird would return to the same area of forest each spring.
The team reviewed 9,072 hours of audio to monitor 405 individual birds. They subsequently tested the model against a different set of recordings gathered from the mixed wood and boreal forests of Alberta during the spring breeding season of 2016.
Lapp noted that the model could associate songs with individual birds with an accuracy rate of up to 96 percent.
This method is significantly less labor-intensive compared to leg banding, where biologists capture birds, carefully attach bands around their legs, and then attempt to recapture them the following spring.
The idea for this research originated from collaboration with biologists in Alberta who were trying to locate birds across extensive areas using a grid of microphones.
Astonishing discovery
According to Lapp, human ears cannot differentiate individual ovenbirds in real-time because their vocal subtleties occur too rapidly.
However, when viewed on spectrographs — graphical representations of sound — the unique song of each bird becomes clear, Lapp explained.
“I began to wonder, if my eyes can perceive these patterns, can we train a machine to recognize these patterns and automate the process of individual recognition?”
“Even for those deeply interested in birds and species identification, it’s quite surprising and thrilling to realize that this level of information exists, which we have been overlooking … yet is present if one listens closely enough.”
Erin Bayne, a biology professor at the University of Alberta and co-author of the study, has been attentive to ovenbirds for many years and understands the challenges of annual population surveys.
During his PhD, Bayne devoted tens of thousands of hours to tracking banded birds, enabling him to distinguish individuals based on their unique vocal differences.
“What that bird is saying over there is slightly different from what that other bird is saying,” he remarked. “I could identify who was who.
“And they are remarkably courteous when they sing. It’s quite amusing, really. One bird will sing, and then the other waits until it finishes, only to immediately sing afterward.”
Birds communicate amongst themselves
Bayne indicated that the study illustrates that avian communication is much more intricate than most people realize, and that conversations are occurring even if we do not grasp the language.
Just as humans exhibit distinct speech characteristics, researchers are discovering that certain animals also develop their own unique vocalizations, according to Bayne.
“It’s not surprising that individual birds or animals possess detectable differences. The more we investigate, the more apparent it will be that this is the case.”
Though the model has its limitations, the researchers aspire for it to offer a less intrusive means of assessing the health of songbirds simply through listening.
The AI system is unable to accurately track the varied calls of female ovenbirds or the basic sounds produced by juveniles.
Additionally, the current model would not be suitable for monitoring species like magpies and mockingbirds, which frequently imitate new sounds or sing a wide array of different songs.
It would be most effective for birds that sing a single, consistent song throughout their lives — which encompasses about 30 percent of songbird species — as well as for certain species of bats and owls.
The researchers have made the program code available on open-source platforms to assist other biologists in tracking the survival of various songbird populations or even different species altogether.
Richard Hedley, a statistical ecologist at the Alberta Biodiversity Monitoring Institute who was not part of the research, stated that it holds tremendous potential.
Although AI acoustic monitoring has been employed for years to track birds at the species level, demonstrating its ability to differentiate between individual birds is a remarkable achievement, he noted.
There exists a vast array of acoustic recordings that could be utilized in innovative ways, he added.
“Catching a single bird can take many hours. Thus, the notion that we can achieve this by simply placing a microphone in a bird’s territory is highly promising.”
