Whether in the city park, on a walk in the woods or in the garden at home - when spending time outdoors, we are often accompanied by a familiar sound: Birdsong. But who exactly is chirping in the trees above us? BirdNET has the answer.
The application is a joint project of the Chair of Media Informatics at Chemnitz University and the Cornell Lab of OrnithologyScientists from Mittweida University of Applied Sciences were also involved.
AI recognises birdsong
BirdNET is a research platform that helps recognize birds based on their calls. All you need to do is record the birdsong. The application generates a spectrogram from the recording, which is used to analyse and identify the bird's voice.
At heart of the analysis software lies an AI-supported algorithm connected to a server at Chemnitz University of Technology. The artificial intelligence was previously trained with around 50,000 recordings of bird calls - over 350 hours of test material in total.
The application can currently recognize around 3,000 of the best-known species and offers innovative tools for conservationists, biologists and birdwatchers.
The software can be used with Arduino microcontrollers, the Raspberry Pi, smartphones, web browsers, stand PCs and cloud services.
BirdNET-App
An Android and iOS app is available for smartphones.
Users can simply use the microphone on their smartphone to record bird calls and have BirdNET identify them. The AI also takes the location and time of the recording into account when analyzing the data.
All observations can be saved in the app and shared with friends.
Birdwatchers can thus easily and effectively recognize who is singing above their heads.
The app has over one million downloads and users far beyond national borders.
Citizen-Science Platform
The software also benefits from the involvement of the general public because new training data is made available by the users of the app themselves. Whenever a user records and identifies a bird call with the help of the application, they generate a new (anonymised) data set that can be used to train the AI further. The users themselves therefore actively contribute to the development process of BirdNET .
Further Development
The experience gained from working on BirdNET provides valuable findings for our project. Using these understandings, we want to improve the BirdNET model in regard to bird species threatened by wind power. BirdNET hence provides the technical and algorithmic basis for the project DeepBirdDetect.
More about BirdNET
BirdNET-Website
TU Chemnitz
Cornell Lab of Ornithology
BirdNET on Twitter
BirdNET on Youtube
BirdNET on mdr Wissen
EN 
DE