About BirdNET

Featherframe doesn't identify birds. BirdNET does. Featherframe shows what it hears.

What it is

BirdNET is an AI that identifies birds by their sound. You give it audio, and it tells you which species are singing, with a confidence score for each.

It's made by the K. Lisa Yang Center for Conservation Bioacoustics at the Cornell Lab of Ornithology and Chemnitz University of Technology in Germany, led by Stefan Kahl. It's free, and its code is open source.

How it works

1. Listen. A microphone records all the time.
2. Slice. The audio is cut into 3-second pieces.
3. Picture it. Each piece becomes a spectrogram: an image of the sound, with pitch going up and time going across.
4. Recognize it. A neural network, trained on a large library of labeled recordings, looks at the image and scores every species it knows.
5. Filter. Species that don't live near you at this time of year are set aside, using eBird's records for your location and week.
6. Report. Anything above your confidence threshold is a detection: species, time, and confidence.

The current model, BirdNET V2.4, knows about 6,500 species worldwide. It runs comfortably on a Raspberry Pi.

It isn't perfect. It mistakes some calls, and it hears insects, frogs, and other noises. A higher confidence threshold means fewer mistakes and fewer detections. For a species it has never heard before, Featherframe waits for a confident detection, or a second one at least 10 minutes after the first, so a one-off mistake isn't shown.

How to get it

	What it is	Works with Featherframe
BirdNET-Pi	A 24/7 listening station on a Raspberry Pi, with a web dashboard. Use the maintained Nachtzuster fork.	Yes
BirdNET-Go	A newer take on BirdNET-Pi, written in Go. Runs on a Pi, a Linux box, or Docker.	Yes
BirdWeather	A worldwide map of BirdNET stations. Stations can use BirdNET-Pi, BirdNET-Go, or BirdWeather's own PUC.	Yes
BirdNET app	Record a bird with your phone to identify it.	No
BirdNET-Analyzer	For researchers: analyze folders of recordings.	No

If you're new to BirdNET, start with BirdNET-Go on a Raspberry Pi 5 with a USB microphone. See BirdNET hardware.

Licenses

BirdNET's code is MIT-licensed. Its trained models are CC BY-NC-SA 4.0: free for personal and research use, not for commercial use.

Read more

• The BirdNET paper: Kahl et al., Ecological Informatics, 2021
• BirdNET-Analyzer documentation
• Detection sources: connect your station to Featherframe