BeVo: Behavior-Vocalization Dataset

BeVo is a behavior-conditioned animal vocalization dataset for generating animal sounds from natural-language descriptions of behaviors, social interactions, and contexts.

BeVo contains 31,191 vocalization events from 6 species, organized into 42 vocalization anchors, with approximately 524.3 minutes of audio.

The behavioral conditions are associated with recordings through shared vocalization anchors and should be interpreted as anchor-level behavioral conditions, rather than event-level ground-truth descriptions of individual recordings.

Project Page: https://persimmontian.github.io/BCAVG

Split	Goat	Hyena	Marmoset	Meerkat	Zebra	Zebra Finch	Total
Train	2,826	10,216	4,800	4,475	353	2,298	24,968
Validation	360	1,226	600	564	62	311	3,123
Test	400	800	600	600	200	500	3,100
Total	3,586	12,242	6,000	5,639	615	3,109	31,191

Dataset Structure

BeVo/
├── README.md
├── LICENSE
├── dataset.svg
│
├── data/
│   ├── source_datasets.csv
│   ├── train/
│   ├── validation/
│   └── test/
│
├── conditions/
│   ├── train_validation/
│   ├── test_assignments.json
│   └── generation_prompt.md
│
└── behavior_cards/
    └── behavior_cards.xlsx

• data/ contains the audio events, organized by split, species, and vocalization anchor.
• conditions/train_validation/ contains the behavioral-condition pools used for training and validation.
• conditions/test_assignments.json contains the fixed audio-condition assignments used for the test set.
• conditions/generation_prompt.md contains the prompt used to generate behavioral conditions.
• behavior_cards/behavior_cards.xlsx contains the literature-grounded behavior cards used as the behavioral knowledge source for condition generation.
• data/source_datasets.csv records the original bioacoustic data sources, publications, persistent identifiers, and source-specific licensing information.

Audio files are organized as:

data/<split>/<species>/<anchor>/*.wav

Source Data and Licensing

BeVo integrates recordings from six public bioacoustic datasets covering spotted hyena, meerkat, common marmoset, goat, plains zebra, and zebra finch.

Detailed information about the original datasets, publications, persistent identifiers, and licensing terms is provided in:

data/source_datasets.csv

BeVo is a mixed-license dataset. Third-party audio retains the license or terms of use of its original source dataset. BeVo-created materials are released under the terms described in the repository-level LICENSE, to the extent that the BeVo authors hold the relevant rights.

Please consult both LICENSE and data/source_datasets.csv before redistributing or reusing the audio.

Citation

If you use BeVo in your research, please cite the dataset:

@misc{deng2026bevo,
  title     = {BeVo: Behavior-Vocalization Dataset},
  author    = {Deng, Xinlong and Tian, Yong and Guan, Jian and Kong, Qiuqiang and Jiang, Jie and Cao, Yin},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Xinghour233/BeVo}
}