FreedomIntelligence/TalkVid

TalkVid是一个大规模、多样化的开源数据集，用于音频驱动的说话人头合成。该数据集包含7729位不同说话者的超过1244小时的HD/4K视频。它覆盖了15种语言，年龄范围广泛（0-60+岁），并具有高质量的视频（1080p和2160p分辨率）和全面的质量过滤。数据集提供了丰富的上下文，包括全身的上半身画面，以及高质量的字幕和全面的元数据。

TalkVid is a large-scale, diversified open-source dataset for audio-driven talking head synthesis, featuring over 1,244 hours of HD/4K footage from 7,729 unique speakers. It covers 15 languages and a wide age range (0–60+ years) with high-quality videos (1080p & 2160p) and comprehensive quality filtering. The dataset provides rich context with full upper-body presence and includes high-quality captions and comprehensive metadata.