Skip to content

Releases: birdnet-team/BirdBox

BirdBox v1.0.0 - First Stable Release

25 Jun 19:38
@El1as42 El1as42
v1.0.0
ab07301

Choose a tag to compare

View all tags
BirdBox v1.0.0 - First Stable Release Latest
Latest

BirdBox is a deep learning system for detecting and evaluating bird calls in audio recordings. It applies YOLO object detection on PCEN spectrograms to localize vocalizations in time and frequency.

Highlights

  • 282 species supported across pre-trained regional and multi-region models
  • YOLO-based detection on PCEN spectrograms with sliding-window processing for arbitrary-length audio
  • Multiple model formats: .pt, .onnx, and .engine
  • Batch and single-file inference via CLI and Python API
  • Song reconstruction to merge temporally adjacent detections into continuous vocalizations
  • Built-in evaluation: F-beta score analysis, confusion matrices, and optimal threshold search
  • Flexible outputs: JSON (with algorithm metadata), CSV, Xeno-Canto Annota-JSON, and Raven Selection Table
  • Interactive demo for quick testing without a local setup

Getting started

Documentation: https://birdnet-team.github.io/BirdBox/

Requirements: Python 3.12 (see Installation)

License

MIT, see LICENSE

Assets 2
Loading