README.md

metadata

license: cc-by-4.0
language:
  - en
tags:
  - multimodal
  - audio-retrieval
  - moment-retrieval
size_categories:
  - 1K<n<10K
task_categories:
  - audio-text-to-text

CASTELLA CLAP features

This repository contains audio and text features of CASTELLA dataset extracted by CLAP.

• Using these features, we can reproduce the audio moments retrieval using CASTELLA, which is used in lighthouse.
• Please also check demo page.

How to Download?

Run the following script:

from huggingface_hub import snapshot_download

repo_id = "lighthouse-emnlp2024/CASTELLA_CLAP_features"
local_dir = "./"

downloaded_path = snapshot_download(
    repo_id=repo_id,
    repo_type="dataset",
    local_dir=local_dir,
    allow_patterns="*.tar.gz",
)

How to Use on Lighthouse

The .tar.gz files should be decompressed by following shell commands:

mkdir -p {LIGHTHOUSE_PATH}/features/castella/clap
mkdir -p {LIGHTHOUSE_PATH}/features/castella/clap_text
tar -zxvf clap.tar.gz -C {LIGHTHOUSE_PATH}/features/castella/clap
tar -zxvf clap_text.tar.gz -C {LIGHTHOUSE_PATH}/features/castella/clap_text

Citation

@article{munakata2025castella,
  title={CASTELLA: Long Audio Dataset with Captions and Temporal Boundaries},
  author={Munakata, Hokuto and Takehiro, Imamura and Nishimura, Taichi and Komatsu, Tatsuya},
  journal={arXiv preprint arXiv:2511.15131},
  year={2025},
}