All ETDs from UAB

Advisory Committee Chair

Nitesh Saxena

Advisory Committee Members

Jeremy Blackburn

Karthikeyan Lingasubramanian

Jonathan Voris

Yuliang Zheng

Document Type

Dissertation

Date of Award

2018

Degree Name by School

Doctor of Philosophy (PhD) College of Arts and Sciences

Abstract

The ubiquitous nature of consumer devices like smartphones, laptops, smartwatches and other smart devices in the current era has led to increased privacy concerns from side-channel attacks that tend to exploit the seemingly benign components of these devices to eavesdrop on confidential data. In particular, many of these devices host increasingly complex hardware sensors for delivering a rich and personalized user experience by consuming a significant amount of personalized information but overlook the possible privacy leakage introduced due to lack of secure implementation of these sensors. The side-channel attacks targeting such privacy leakage are low-cost in nature, using off-the-shelf equipment and do not require significant computing power for recovering secure information from the eavesdropped data. In this work, we dissect the vibration side channel that allows an attacker to compromise the user's privacy by leaking secure information, specifically their speech characteristics. Since audio is a form of vibration, falling within the audible frequency range, we consider the audio channel as a subset of the vibration channel. We establish a relationship between the audio and vibration communication channels and show how it facilitates an attacker to eavesdrop on vibration effects generated by the audio communication channel and vice-versa. We study these attacks in the context of smart mobile devices such as smartphones that offer a low cost and convenient way to eavesdrop on both audio and vibration side channels. On the other end of the security spectrum, we leverage the vibration side channel to be used to counteract against automated voice impersonation attacks that aim to imitate a user's voice features in the audio domain. As per our analysis, the relationship between audio and vibration side channels allows us to utilize the vibration side channel for this purpose. In addition, we address the problem of designing a practical and low-cost defense mechanism that can be effectively applied to most of the privacy leakage scenarios involving audio side-channel eavesdropping. Our dissertation offers the following contributions: First, we perform a security analysis of the vibration side-channel in the context of speech privacy by analyzing the threat of external vibration-based eavesdropping on human and machine speech signals using embedded motion sensors on mobile devices. Second, utilizing this threat analysis, we construct an eavesdropping attack using the vibration side-channel that exploits the motion sensors of the smartphone to extract speech information from its inbuilt loudspeakers. Third, building upon the relationship between vibration and audio domain, we leverage the defensive use of vibration side-channel by constructing a practical and efficient voice authentication model that utilizes the speech vibrations to uniquely perform speaker classification and is resilient to voice impersonation attacks (in contrast to existing state of the art voice authentication mechanisms). Fourth, we use the audio side-channel attack to eavesdrop on rhythmic password-based authentication systems and offer sound masking as a countermeasure to our proposed attack. Fifth, we study acoustic eavesdropping attack on keyboards (in the context of random passwords and PINs) in both local and remote eavesdropping scenarios (such as over a VoIP call). We test the effectiveness of sound masking as a potential defensive mechanism against the previously studied local and remote acoustic eavesdropping attacks on keystrokes. Sixth, we investigate an acoustic eavesdropping attack on vibration-based pairing mechanisms in local remote eavesdropping settings and offer noisy vibration pairing as a low cost and effective countermeasure against the eavesdropping attack. Overall, our work reveals a significant correlation between audio and vibration channels that has major implications on the security of a system. In the context of speech privacy, the vibration side-channel can act either as a threat to speech privacy or as a voice authentication enhancement measure. In the context of secure operations such as pairing and password entry, acoustic side-channel attacks are effective in both local and remote eavesdropping settings. However, they can be effectively resisted by a relevant sound masking approach that makes it hard for an eavesdropping attack to extract confidential information from the eavesdropped signal thereby preserving the privacy of the communication channel.

Share

COinS