AI Listens to Bird Song in Sierra Nevada Project

Researchers are using artificial intelligence to decode the natural landscapes of the Sierra Nevada in California, reported Scientific American. The AI uses machine learning technology to distinguish bird songs and assigns them to the right species.

In the summer, there are 2,000 recorders planted in the Sierra Nevada mountain range to help advocates and researchers understand how they can help conserve the ecosystem and help bird species thrive.

The project is expected to generate millions of hours of audio. Human ears alone can take a long time to sift through the hours of audio to identify bird calls and pinpoint them to the correct species. With the help of artificial intelligence, the time it takes to listen to the data can be cut significantly.

Sierra Nevada Project AI Listens to Bird Song

Machine learning expert Stefan Kahl at the Cornell Center for Conservation Bioacoustics and Chemnitz University of Technology in Germany used a neural network to build BirdNET, an avian sound recognition system.

Ecologist and Cornell University postdoctoral researcher Connor Wood said, “Audio data is a real treasure trove because it contains vast amounts of information,”

The Sierra Nevada project leader added, “We just need to think creatively about how to share and access [that information].” The AI system has the potential to identify species from their songs found in thousands of hours of audio within a day.

The Sierra Nevada team will use BirdNet to help analyze and document the recordings. However, this existing technology has limitations in that it is semi-automatic. It uses spectrograms for known calls to identify birds according to their song.

This is effective for some species, but may not offer clear answers to others, especially as birds were found to have “regional” dialects.

India-based ecologist V. V. Robin noted, “In other places, for rarer species or ones that don’t have well-classified data, [BirdNET] doesn’t work as well.”

To address these issues, researchers at Cornell Lab of Ornithology and New York University’s Music and Audio Research Laboratory collaborated to create BirdVox, a bird-call identification system. This new AI identifies bird calls from background noises and other calls.

BirdVox specializes in species found in the United States, while BirdNET contributed around 3,000 species found in Europe and North America. Currently, there are 4.2 million recordings of around 10,000 species available.

While AIs like BirdNET and BordVox have contributed a lot to the studies, they are yet to improve to become more efficient, effective, and specific.