| license: cc-by-4.0 | |
| > [!WARNING] | |
| > The authors request this repository remains private while the data and paper details are finalised | |
| ## WABAD: A World Annotated Bird Acoustic Dataset for Passive Acoustic Monitoring | |
| Original dataset available and full authors list: https://zenodo.org/records/15629388 | |
| Dataset corresponding author: Cristian Pérez Granados (cristian.perez@ctfc.cat) | |
| Additional dataset curation: Ben McEwen (b.j.mcewen@tilburguniversity.edu), Corentin Bernard (corentin.bernard@lis-lab.fr) | |
| ## Dataset Curation | |
| The original dataset (available on Zenodo) is composed of mostly one-minute audio files (.wav) with corresponding Audacity and Raven Pro annotations corresponding to each individual audio file. | |
| This curated version of the dataset has been adjusted for ease of use: | |
| - Segmented audio into 3-second files (.wav) | |
| - Multi-label annotations combined with site metadata in parquet files | |
| - Train and test splits (80/20 %) predefined by respective .parquet files | |
| ### Segmentation Method | |
| Raven Pro annotations onset/offset times were converted to a species list (column: species) and one-hot encoding (column: labels) for 3-second segment for which a minimum vocalisation length of 0.5-second is present within the segment. 3-second segments are non-overlapping. | |