Facestar

Facestar数据集旨在在大规模和高质量的环境中进行视听建模研究。核心数据集功能: 10小时的高质量视听语音数据音频记录在安静的环境中，分辨率为1300x1600的视频，每秒60帧，一个女性和一个男性说话者自然语音: 所有数据都是视频会议设置中的会话语音全脸可见性: 扬声器在说话时对着镜头

The Facestar Dataset is intended for audiovisual modeling research in large-scale, high-quality contexts. The core functionalities of the dataset are listed below: 1. 10 hours of high-quality audiovisual speech data: audio recorded in quiet environments, video with a resolution of 1300×1600 at 60 frames per second, featuring natural speech from one female and one male speaker; 2. All data consists of conversational speech captured in video conference settings; 3. Full facial visibility: Speakers face the camera directly while speaking.

搜集汇总

背景与挑战

背景概述

Facestar是一个用于大规模视听建模研究的高质量数据集，包含10小时由一男一女说话者在安静环境中录制的自然会话语音数据，视频分辨率为1300x1600、帧率为60fps，且说话者面部完全可见。该数据集由Meta Reality Labs于2022年发布，适用于视听语音增强等研究领域。