American biologists have made the work of ornithologists and bird enthusiasts easier. A team from West Virginia University, in collaboration with the U.S. Fish and Wildlife Service, has trained a neural network to recognize individual wild birds by their songs with up to 90% accuracy.
Traditional audio monitoring had many limitations — for example, it was difficult to determine whether one bird or several were singing in a forest. For a long time, scientists had to capture birds, band them, or attach radio transmitters to track their survival and migration.
The neural network was trained on recordings of the blue-winged warbler — a rare migratory species whose population has been declining each year. The system converts sound into spectrograms and analyzes details that are inaudible to the human ear, such as pause lengths, frequency variations, and unique syllable combinations in trills.
Tests have shown that the model can correctly identify bird species even when the recording quality is low. In the future, such technologies will allow ecologists to monitor rare birds around the clock and more accurately estimate their populations.