The audio in WoW-Bench comes from the Watkins Marine Mammal Sound Database and is shared with permission from the New Bedford Whaling Museum. Each request is reviewed manually, and the information you submit is shared with the Museum.

WoW-Bench

Questions and audio for "WoW-Bench: Evaluating Fine-Grained Acoustic Perception in Audio-Language Models via Marine Mammal Vocalizations" (Findings of ACL 2026).

[Paper] [Code] [Project Page]

metadata.jsonl: one line per question (1,777 in total) with question_id, file_name, question, choice and answer.
<benchmark>/<task>/<question_id>.wav: the audio for each question. file_name in metadata.jsonl is the path of the audio relative to the dataset root.

Download

hf auth login
hf download jaeyeonkim99/WoW-Bench --repo-type dataset --local-dir wow_bench

If you log in with a fine-grained token, enable read access to public gated repositories. See the code repository for the evaluation script.

Terms of Use

The data may be used for research and education only. Commercial use requires separate permission from the New Bedford Whaling Museum.
Do not redistribute the audio files. Anyone who needs them should request access on this page.
Credit the audio as "Watkins Marine Mammal Sound Database / New Bedford Whaling Museum", with a link to www.whalingmuseum.org for web use.

Acknowledgments

The audio recordings and associated metadata used in this work were sourced from the Watkins Marine Mammal Sound Database, Woods Hole Oceanographic Institution, and the New Bedford Whaling Museum. We gratefully acknowledge the New Bedford Whaling Museum for granting permission to use the database for research purposes.

Citation

@inproceedings{kimwow,
  title={WoW-Bench: Evaluating Fine-Grained Acoustic Perception in Audio-Language Models via Marine Mammal Vocalizations},
  author={Kim, Jaeyeon and Yun, Heeseung and Woo, Tony and Yang, Chao-Han Huck and Kim, Gunhee},
  booktitle={Findings of ACL},
  year={2026}
}

Downloads last month: 47

Total file size:

1.96 GB

Paper for jaeyeonkim99/WoW-Bench

WoW-Bench: Evaluating Fine-Grained Acoustic Perception in Audio-Language Models via Marine Mammal Vocalizations

Paper • 2508.20976 • Published Aug 28, 2025

You need to agree to share your Hugging Face information to access this dataset