RHAPSODY
收藏资源简介:
RHAPSODY是一个包含13,364个播客集的音频数据集,每个集都与YouTube的“最常重播”功能提取的段落级突出得分相匹配。该数据集旨在用于播客集的突出部分检测,将任务定义为段落级二分类任务。数据集涵盖了来自不同领域的播客,平均每集约30分钟长,包含约5,000个单词。数据集创建过程包括从YouTube上收集最受欢迎的播客创建者的数据,使用YouTube的“最常重播”功能来识别突出部分,并使用语音识别和文本摘要技术来生成数据。该数据集适用于播客内容理解的研究,旨在帮助听众快速定位播客集中最有兴趣的部分。
RHAPSODY is an audio dataset containing 13,364 podcast episodes, each paired with paragraph-level prominence scores extracted via YouTube's "Most Replayed" feature. This dataset is intended for highlight detection in podcast episodes, framing the task as a paragraph-level binary classification task. The dataset encompasses podcasts across diverse domains, with each episode averaging 30 minutes in duration and approximately 5,000 words. The dataset creation workflow includes collecting data from top podcast creators on YouTube, identifying highlight segments using YouTube's "Most Replayed" functionality, and generating the dataset through speech recognition and text summarization technologies. This dataset is applicable to research on podcast content understanding, aiming to help listeners quickly locate the most engaging segments within podcast episodes.

- 1Rhapsody: A Dataset for Highlight Detection in PodcastsYonsei University, The University of Texas at Austin, New York University · 2025年



