Who's Your Birdy
Identify birds by their song, entirely in your browser.
- Year
- 2026
- Type
- AI
- Platform
- Web
- Status
- Live
- Built with
- TypeScript, BirdNET, ONNX Runtime, WebAssembly, Web Worker
Runs the BirdNET v2.4 model locally in WebAssembly. No backend, no upload: the audio file never leaves your machine. Non-commercial, as the model's licence requires.
Built on BirdNET v2.4 from the Cornell Lab of Ornithology and TU Chemnitz. The model, about 52 MB, downloads once and is cached. Inference runs in WebAssembly inside a Web Worker, and the app then works offline.
The interface is built around the spectrogram. It includes BirdNET's geographic and seasonal filter, and live listening from the microphone.
Scores are checked against the official implementation. One known gap: on the zero-padded last window of a file whose length is not a multiple of 3 seconds, a score can differ by up to 0.017, against a 0.001 target. No detection changed across the 35 padded windows tested.
The model is licensed CC BY-NC-SA 4.0, so the project is non-commercial.
