Exploiting Sound Masking for Audio Privacy in Smartphones
Yu-Chih Tung, Kang Geun Shin · 2019
Privacy leakage via malware's unauthorized audio recording has become an emerging threat to mobile users. To counter this threat, we propose SafeChat which prevents unauthorized recording without requiring new audio privacy settings in operating systems. SafeChat utilizes sound masking to differentiate audio information between authorized and unauthorized recording apps. Specifically, SafeChat enables authorized recording apps to recover more private/secret information than the unauthorized apps even though both of them record identical audio signals from the same microphone. We have implemented SafeChat as an Android chat app. Our experiments on several commodity phones have shown that SafeChat can make an up-to-26dB difference in signal strength between authorized and unauthorized recording apps. This difference reduces the accuracy of state-of-the-art speech recognition engines, like Google Speech API, to less than 0.1% in understanding the unauthorized recording while comprehending the authorized recording with high accuracy. Moreover, none of the 317 testing participants we recruited online could comprehend the masked speech. Our usability study shows that only 35% of the participants were aware of this threat of privacy leakage and 60% of them wanted to use SafeChat to protect their private/secret information from unauthorized recording.