🦆 Listening to the Sky: How AI Helps Study Birds
Around 4 billion birds migrate throughout North America every year. At night, they communicate with their flocks via feeble signals that last barely 50 milliseconds.
Researchers have attempted to trace migrations using these calls for years, but the human ear struggles to distinguish such subtle sounds.
BirdVoxDetect, an AI model developed over 8 years by a team of scientists from the U.S. and France, solved the problem.
It was trained to separate bird calls from background noise by analyzing thousands of hours of manually annotated audio recordings. Scientists compare it to teaching a smart speaker to respond to a specific keyword, but it's far more difficult with birds because each species has its own set of calls.
The neural network is as accurate as traditional radars at tracking migration volumes in real-world conditions. Furthermore, BirdVoxDetect outperforms GPS tags and visual estimates in identifying specific species within huge moving flocks.
Researchers now aim to train the AI to determine a bird's spatial position. In the future, this could become the foundation for a universal model for studying any animal based on its vocalizations.
More on the topic:
🐥 Merlin Bird ID—Shazam in the world of birds
