Recording the subtle syllables, notes and motifs that distinguish one bird species from another requires some pretty sophisticated gear. But for OSU researchers, collecting audio data in an old-growth forest last summer was a walk in the park compared with analyzing it. “It’s a lot of data,” reports Jed Irvine, a faculty research assistant in the OSU Bioacoustics and Machine Learning group.
Confronted with a terabyte of digital sound from the H.J. Andrews Experimental Forest, Irvine and a team of students in the College of Engineering are building a website that will let them borrow the ears of experienced birders to identify avian singers. These IDs will then be used to “teach” computers how to distinguish a robin from a Swainson’s thrush or a tree swallow for a study being led by forest ecologist Matt Betts.
The terabyte gets its name from the Greek word tera, meaning “monster.” The etymology is apt. Trying to grasp the size of a terabyte – a trillion bytes of computer data – is like trying to wrap your mind around the number of water drops in Crater Lake or sand grains on Cannon Beach. Besides their monstrous size, these audio files may contain all sorts of extra sounds, from streams to airplanes to distant highway traffic. Also complicating the task of automatically recognizing bird sounds is the fact that birds can sing in regional “dialects,” and some even mimic other species.
Where the Birds Are
“The bioacoustic team is developing software that will automatically identify bird species – perhaps even individual birds – so that we can assess population distribution on an ongoing basis,” Irvine explains. Then, without a hint of irony, he adds, “It’s a lofty idea.”
An amateur birder himself, Irvine came up with the idea of creating a website that would allow birders to upload bird images and audio for annotation and discussion. The bioacoustics and machine learning problem being worked on by professors Xiaoli Fern and Raviv Raich was a perfect task for the Web, Irvine says. Once the website was ready for the first round of user testing, he asked local birders to test-drive the site by listening to sound clips online and then posting species IDs for each clip. By the end of June, he had gotten 85 identifications from about a dozen volunteers.
In the next phase of website development, he hopes to make the experience of using the site as close to the experience of “birding by ear” as possible. Each online session will be designed as a “birding trip” into the forest, where volunteer birders can employ their knowledge of birdsong to further the goals of science.
“The next iteration will be more interesting,” says Irvine, who started birding with his dad as a kid in New Jersey. “We want to make the site addictively fun for birders, so that we can get as many bird sound snippets identified as possible.”