Datasets:

lighthouse-emnlp2024
/

CASTELLA_CLAP_features

npz dict	__key__ stringlengths 13 18	__url__ stringclasses 2 values
{"features":[[0.32028162479400635,0.13711793720722198,0.2178938090801239,0.057061754167079926,-0.014(...TRUNCATED)	./zY3icUyMdh8	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.5468708276748657,0.7032186985015869,-0.05993366241455078,-0.018820997327566147,0.045(...TRUNCATED)	./-2va_YGrK_4	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.6369049549102783,0.536095917224884,0.21774634718894958,0.2592740058898926,0.24336147(...TRUNCATED)	./_GI7meqlYZk	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.4778326749801636,0.27810317277908325,0.20423077046871185,0.16321846842765808,0.33478(...TRUNCATED)	./KJhGuhNHToA	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.7288158535957336,0.195418119430542,0.36413347721099854,-0.082061268389225,0.19499891(...TRUNCATED)	./Ls1zyPjs3k8	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.4318028688430786,0.4769047498703003,0.00005314976442605257,0.4175373315811157,0.1965(...TRUNCATED)	./3Qzk1nQ3a7Q	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.32028162479400635,0.13711793720722198,0.2178938090801239,0.057061754167079926,-0.014(...TRUNCATED)	./KvrcRMfFzOE	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.10982418060302734,0.7713618278503418,-0.4591480493545532,0.2551789879798889,-0.10882(...TRUNCATED)	./pmso1Ye6isk	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.42791223526000977,0.5555799603462219,0.22977358102798462,0.13306060433387756,0.07321(...TRUNCATED)	./UCy1BEx8jBE	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)
{"features":[[0.4593592882156372,0.5794185400009155,0.08515777438879013,0.16417160630226135,0.191399(...TRUNCATED)	./OjECKYqsM8g	"/tmp/hf-datasets-cache/medium/datasets/28188483982468-config-parquet-and-info-lighthouse-emnlp2024-(...TRUNCATED)

End of preview. Expand in Data Studio

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},
}

Downloads last month: 11

Size of downloaded dataset files:

1.33 GB

Size of the auto-converted Parquet files:

1.34 GB

Number of rows:

5,742

Paper for lighthouse-emnlp2024/CASTELLA_CLAP_features

CASTELLA: Long Audio Dataset with Captions and Temporal Boundaries

Paper • 2511.15131 • Published Nov 19, 2025 • 1