Audio Detection
Updated acoustic detection with top-5 species ranking.
Upload a bird audio clip to preview the latest LightBirdNet workflow, including 5-second mono analysis, log-Mel feature extraction, and ranked confidence output.
Upload Input
Upload a bird audio clip
Upload a field recording and BirdLens will process a 5-second mono window using the updated LightBirdNet acoustic pipeline for top-species ranking.
Drag and drop your audio clip
Or choose a file manually for a premium preview of the BirdLens detection flow.
Supported formats: WAV, MP3, OGG, FLAC, and M4A. Audio is resampled to 22.05 kHz, converted to mono, and normalized into a 128x128 log-Mel window.
Recording UI Placeholder
Live recording can slot in here next.
The interface reserves space for browser capture so an end-to-end record-and-detect flow can be added after the production inference endpoint is connected.
Ready when your input is.
Upload a bird audio clip to preview the premium result state, including confidence, alternatives, and concise scientific framing.
Detection Notes
Input
Single uploaded recording (5s analysis window)
Model
LightBirdNet (log-Mel spectrogram classifier)
Output
Top-5 species probabilities
Preprocessing behavior
Clips are converted to mono at 22.05 kHz, then represented as 128-bin log-Mel spectrograms to keep inference consistent across uploads.
When confidence improves
Predictions improve when one bird is dominant in the foreground and environmental noise does not mask repeated phrase structure.
How to read the result
Read the top match as the leading hypothesis, then compare the remaining ranked candidates against location, season, and behavior notes.