PAMS: Improving Privacy in Audio-Based Mobile Systems
PAMS: Improving Privacy in Audio-Based Mobile Systems
复制标题
DOI：
10.1145/3417313.3429383
复制
发表时间：
2020-11
期刊：
影响因子：
--
通讯作者：
S. Xia;Xiaofan Jiang
中科院分区：
文献类型：
--
作者：
S. Xia;Xiaofan Jiang
智能手机和移动应用程序已成为我们日常生活中不可或缺的一部分。这反映在移动设备、应用程序和每年产生的收入的增长上。然而，这种增长也伴随着人们对用户隐私的日益关注，近年来发生了许多与智能手机和移动应用程序相关的隐私和数据泄露事件。在这项工作中，我们专注于改善基于音频的移动系统的隐私性。这些应用程序通常会监听环境中的所有声音，并可能记录应用程序可能不需要的隐私敏感信号，例如语音。我们推出 PAMS，一个用于移动应用程序的软件开发包。 PAMS 集成了一种称为概率模板匹配的新颖声源过滤算法，可生成一组隐私增强过滤器，使用学习到的这些声音的统计“模板”来消除无关的声音。我们通过将 PAMS 集成到睡眠监测系统中来证明 PAMS 的有效性，目的是消除系统正在监测的呼吸、打鼾和其他睡眠声音中的无关语音。通过将我们的 PAMS 增强型睡眠监测系统与现有的移动系统进行比较，我们发现 PAMS 可以将语音清晰度降低高达 74.3%，同时在检测睡眠声音方面保持相似的性能。
Smartphones and mobile applications have become an integral part of our daily lives. This is reflected by the increase in mobile devices, applications, and revenue generated each year. However, this growth is being met with an increasing concern for user privacy, and there have been many incidents of privacy and data breaches related to smartphones and mobile applications in recent years. In this work, we focus on improving privacy for audio-based mobile systems. These applications will generally listen to all sounds in the environment and may record privacy-sensitive signals, such as speech, that may not be needed for the application. We present PAMS, a software development package for mobile applications. PAMS integrates a novel sound source filtering algorithm called Probabilistic Template Matching to generate a set of privacy-enhancing filters that remove extraneous sounds using learned statistical "templates" of these sounds. We demonstrate the effectiveness of PAMS by integrating it into a sleep monitoring system, with the intent to remove extraneous speech from breathing, snoring, and other sleep sounds that the system is monitoring. By comparing our PAMS enhanced sleep monitoring system with existing mobile systems, we show that PAMS can reduce speech intelligibility by up to 74.3% while maintaining similar performance in detecting sleeping sounds.