MAVS数据集

MAVS数据集是由挪威科技大学和印度理工学院合作开发的多语言视听智能手机数据集，旨在支持开发和评估先进的单模态或视听说话人识别系统。该数据集包含103名参与者（70名男性和33名女性），在三个不同会话中录制，考虑了不同的现实世界场景，并使用了五种不同的智能手机设备。数据集中的每个参与者用三种语言（英语、印地语和孟加拉语）说出六句话，每种语言三句。此外，数据集还包括两种类型的演示攻击，即物理访问攻击和逻辑访问攻击，用于评估生物识别系统的鲁棒性。MAVS数据集的应用领域包括但不限于生物识别验证系统的开发、多模态生物识别技术的研究以及演示攻击检测方法的评估，旨在解决生物识别系统在智能手机环境中的通用性和鲁棒性问题。

The MAVS dataset is a multilingual audio-visual smartphone dataset developed collaboratively by the Norwegian University of Science and Technology and the Indian Institute of Technology, aiming to support the development and evaluation of advanced unimodal or audio-visual speaker recognition systems. It comprises 103 participants (70 males and 33 females), with data collected across three distinct sessions covering various real-world scenarios, and five different smartphone devices were utilized for recording. Each participant in the dataset utters six sentences in three languages: English, Hindi, and Bengali, with three sentences per language. Additionally, the dataset includes two types of presentation attacks, namely physical access attacks and logical access attacks, for evaluating the robustness of biometric systems. Application scenarios of the MAVS dataset include, but are not limited to, the development of biometric verification systems, research on multimodal biometric technologies, and evaluation of presentation attack detection methods, with the goal of addressing the generality and robustness issues of biometric systems in smartphone environments.