TalkingMachines Video Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://aaxwaz.github.io/TalkingMachines/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了150万个高质量、以人为中心的高清视频片段，这些片段经过了场景检测、美学评分、动作评分以及文本内容的筛选。视频片段主要聚焦于单一说话者，且每段至少持续4秒钟，支持多种不同的宽高比。该数据集的规模达到了150万个片段，旨在支持音频驱动的视频生成和唇形同步任务。
This dataset contains 1.5 million high-quality human-centric high-definition video clips. These clips have undergone screening via scene detection, aesthetic scoring, action scoring and text content filtering. Primarily focusing on a single speaker, each clip has a duration of at least 4 seconds and features a variety of aspect ratios. Comprising 1.5 million clips in total, this dataset is designed to support audio-driven video generation and lip-sync tasks.
