BirdNET

Recognise birds by their song, from an ordinary phone.

AI inside the tool Biodiversity Education and knowledge

You record a song and the app suggests the species. Beyond curiosity, the recordings help researchers follow where birds live and how changes to their habitat affect them.

And where does AI come in?

A neural network analyses the spectrogram of the recording. Researchers use the same tool to sift through thousands of hours of automatic recordings that no team could listen to.

The AI label

Model or technique: Convolutional neural network on spectrograms (EfficientNetB0 for version 2.4, 6,522 species).
Model published: Yes
Where it runs: On the user's computer or phone; lighter versions run on small devices.
Data: Annotated recordings, notably from Xeno-canto. Models published under CC BY-NC-SA 4.0, no commercial use.
Role of people: Every detection is for the user to check, with tools built for that.
Frugality: A 50 MB model; BirdNET-Tiny targets microcontrollers.

“Not filled in”: MAUB could not check it with the project. Know more? Write to us.

Something wrong or out of date? Write to contact@subalpine.it.