VoxCeleb

VoxCeleb是由牛津大学视觉几何组创建的大规模文本无关说话人识别数据集，包含超过100,000条来自1,251位名人的真实世界语音数据。数据集通过自动化流程从YouTube视频中提取，涉及视频下载、面部跟踪、主动说话人验证和面部验证等步骤。该数据集旨在为说话人识别技术提供大规模、多样化的训练和测试资源，适用于高安全性系统认证、法医测试等多个领域。VoxCeleb is a large-scale text-independent speaker recognition dataset created by the Visual Geometry Group at the University of Oxford. It contains over 100,000 real-world speech utterances from 1,251 celebrities. The dataset is extracted from YouTube videos via an automated workflow that includes steps such as video downloading, face tracking, active speaker verification and face verification. It aims to provide large-scale and diverse training and testing resources for speaker recognition technologies, and is applicable to multiple fields such as high-security system authentication and forensic testing.