Chinese Talking-Face Dataset
收藏资源简介:
该数据集由京东和香港大学的研究团队构建,旨在促进中文环境下的说话人脸生成研究。数据集包含约1100个高质量视频,总时长达130小时,视频来源于Bilibili和抖音平台。数据经过严格筛选,确保每个视频仅包含一个可见人脸,且音频与说话者身份一致。数据集的应用领域主要集中在说话人脸视频生成,特别是唇音同步和视觉质量的提升。通过该数据集,研究者可以训练模型以实现更精确的唇音同步和高质量的视频生成。
This dataset was constructed by a research team from JD.com and The University of Hong Kong, aiming to facilitate research on talking face generation in Chinese-language contexts. It contains approximately 1,100 high-quality videos with a total duration of 130 hours, sourced from Bilibili and Douyin platforms. The dataset underwent strict screening to ensure that each video contains only one visible human face, and the audio matches the speaker's identity accurately. The primary application domains of this dataset center on talking face video generation, particularly lip synchronization and visual quality enhancement. With this dataset, researchers can train models to achieve more precise lip synchronization and high-quality video generation.

- 1JoyGen: Audio-Driven 3D Depth-Aware Talking-Face Video Editing京东(JD.Com, Inc.)和香港大学 · 2025年



