EarthSpeciesProject/animalspeak-pseudovox

AnimalSpeak Pseudovox

This dataset contains 346,907 short, silence-trimmed animal vocalization clips from the train-unseen split. WAV audio is included in data-20k/train-*.tar.

Quick start

from datasets import load_dataset

metadata = load_dataset("EarthSpeciesProject/animalspeak-pseudovox", split="train")

Download and extract the TAR shards from data-20k/. Load a row's WAV from audio/<audio_name>.

Metadata

Column	Meaning
id	Clip ID: pseudovox_ followed by 32 hexadecimal characters.
audio_name	WAV filename inside a TAR shard's audio/ directory.
duration_s	Clip duration in seconds.
species_scientific_gbif	Source species label.
custom_classifier_species, custom_classifier_max_p	Predicted species and classifier confidence.
birdnet_confidence	BirdNET confidence.
source_url	Original recording page, where available.

IDs identify rows; filenames locate audio. ROOTS references these exact IDs in source_audio_ids.

License

Source licenses and attribution requirements apply to the audio. See the repository's license metadata and original recording sources.

Project and attribution

This dataset accompanies BEANS-Next and ROOTS: Broadening Audio-Language Capabilities for Bioacoustics.

@misc{plachouras2026beansnextroots,
  title={{BEANS-Next and ROOTS}: Broadening Audio-Language Capabilities for Bioacoustics},
  author={Christos Plachouras and David Robinson and Marius Miron and Gagan Narula and Paul Laisné and Anthony L. T. Fine and Benno Weck and Ellen Gilsenan-McMahon and Diane Kim and Laura Hay Mack and Maddie Cusimano and Sara Keen and Lukas Rauch and Benjamin Hoffman and Emmanuel Chemla and Emmanouil Benetos and Johan Pauwels and Milad Alizadeh and Matthieu Geist and Olivier Pietquin},
  year={2026},
  eprint={2610.10663},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2610.10663}
}