This repository accompanies the paper Time-frequency localization of bird calls in dense soundscapes and contains two components:
birdbox/ |
Julia toolkit to train and run YOLO and RF-DETR bird call detectors on raw audio |
birdwatch/ |
Tool to annotate and analyze spectrograms in the browser |
BirdBox localizes bird vocalizations in time and frequency from raw audio using custom-trained YOLO11 and RF-DETR models. It includes detectors pre-trained on complex soundscapes from Singapore, along with tools to apply them to new recordings and train custom detectors.
See birdbox/README.md for setup and usage.
BirdWatch is a browser-based annotation and analysis tool for spectrograms that supports bounding-box labeling, playback of selected time–frequency regions, and evaluation of model predictions, with all processing performed locally in the browser.
See birdwatch/README.md for usage, or open the app directly: BirdWatch →
@misc{hexeberg2026timefrequencylocalizationbirdcalls,
title={Time-frequency localization of bird calls in dense soundscapes},
author={Simen Hexeberg and Fanghui Tong and Hari Vishnu and Mandar Chitre},
year={2026},
eprint={2606.10407},
archivePrefix={arXiv},
primaryClass={cs.SD},
url={https://arxiv.org/abs/2606.10407},
}MIT — see LICENSE.