zachz/AV-SpeakerBench-PC

AV-SpeakerBench-PC是一个来自AV-SpeakerBench论文(Nguyen et al., 2025)的视频对分类数据集。每条数据包含两个视频片段(video1, video2)和一个二进制标签，标签表示这两个视频片段是否来自同一源视频(同一说话者)或不同源视频(不同说话者)。数据集经过去重处理，包含2048个平衡的视频对，这些视频对来自YouTube源视频ID。数据集格式包含视频和整型标签，适用于视频分类任务，语言为英语。

AV-SpeakerBench-PC is a video pair classification dataset from the AV-SpeakerBench paper (Nguyen et al., 2025). Each row contains two video clips (video1, video2) and a binary label indicating whether they come from the same source video (same speaker context) or different source videos (different speakers). The dataset is deduplicated and contains 2048 balanced pairs generated from YouTube source video IDs. The data format includes videos and integer labels, suitable for video classification tasks, with English as the language.