# WABAD: World Annotated Bird Acoustic Dataset (3-second segments)

## Facts

- Provider: Cristian Perez-Granados, Esther Sebastian-Gonzalez et al.; curated by Ben McEwen, Corentin Bernard
- License: CC-BY-4.0 (https://zenodo.org/records/17293588)
- Size: 222 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/DBD-research-group/WABAD
- Access: public, readable without an account
- DOI: https://doi.org/10.5281/zenodo.17293588
- Category: Environmental Science
- Source: https://zenodo.org/records/17293588
- Page: https://scigantic.com/archive/050f1f4b-03b5-55cf-ad49-05c1c42bfa88
- Keywords: Environmental Science, bioacoustics, Parquet, Hugging Face

## About

WABAD, the World Annotated Bird Acoustic Dataset for passive acoustic monitoring, is a globally representative set of soundscape recordings annotated to species level by local experts. The original Zenodo release includes 5,047 minutes of audio with the start and end time and upper and lower frequency of each identified bird vocalisation, covering 91,931 vocalisations from 1,192 bird species recorded at 72 sites worldwide, with Audacity and Raven Pro annotations for each mostly one-minute wav file. It supports training and evaluating bird sound detection and species identification models (BirdNET-style classifiers, bioacoustic foundation models) across regions. This Hugging Face version, curated by Ben McEwen and Corentin Bernard, segments the audio into non-overlapping 3-second wav clips, converts Raven annotations into per-segment multi-label species lists (a species counts when at least 0.5 s of its vocalisation falls in the segment) and predefines 80/20 train and test splits. It holds 72 site folders (for example ARD, BAM, BERB), each with an audio.zip and train and test metadata Parquet files, plus a species_list.csv and a loading script (measured 2026-11-14; the Hub reports about 28 GB). The original dataset is published on Zenodo under CC BY 4.0; cite the WABAD data paper and DOI.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("DBD-research-group/WABAD")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `ARD/ARD_metadata_test.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/050f1f4b-03b5-55cf-ad49-05c1c42bfa88
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/050f1f4b-03b5-55cf-ad49-05c1c42bfa88
- Over MCP: `get_schema_card` with id `050f1f4b-03b5-55cf-ad49-05c1c42bfa88` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Environmental Science datasets](https://scigantic.com/topics/environmental-science)
- [Bioacoustics datasets](https://scigantic.com/topics/bioacoustics)
