MAV-Celeb

该数据集是一个包含英语、印地语和乌尔都语等多语种说话视频的视听资料集，这些视频是从网络媒体平台上收集而来的。为了分析语言效应，数据集被分为两个版本（V1和V2），其中包含了真实世界场景的视频，且视频质量受到无约束的噪声退化影响。该数据集涵盖了154个不同的身份，以及多种真实世界的场景。其研究任务是在多语言环境中进行人脸与声音的关联分析。

This dataset is an audiovisual corpus consisting of multilingual speech videos in languages including English, Hindi, and Urdu, which was collected from online media platforms. To investigate language effects, the dataset is divided into two versions, V1 and V2. Both versions contain videos captured from real-world scenarios, with their quality degraded by unconstrained noise. This dataset includes 154 unique identities and spans a diverse set of real-world scenarios. The core research task associated with this dataset is face-voice association analysis within a multilingual context.