Audio Hotspot Attack: An Attack on Voice Assistance Systems Using Directional Sound Beams and its Feasibility

Abstract

We propose a novel attack, called an "Audio Hotspot Attack," which performs an inaudible malicious voice command attack, by targeting voice assistance systems, e.g., smart speakers or in-car navigation systems. The key idea of the approach is to leverage directional sound beams generated from parametric loudspeakers, which emit amplitude-modulated ultrasounds that will be self-demodulated in the air. Our work goes beyond the previous studies of inaudible voice command attack in the following three aspects: (1) the attack can succeed on a long distance (3.5 meters in a small room, and 12 meters in a long hallway), (2) it can control the spot of the audible area by using two directional sound beams, which consist of a carrier wave and a sideband wave, and (3) the proposed attack leverages a physical phenomenon i.e.,non-linearity in the air, to attack voice assistance systems. To evaluate the feasibility of the attack, we performed extensive in-lab experiments and a user study involving 20 participants. The results demonstrated that the attack was feasible in a real-world setting. We discussed the extent of the threat, as well as the possible countermeasures against the attack.

Development of an Acoustic Spot Data Communication System Using Multiple Parametric Array Loudspeakers

“…There are some researches considering to emit audio to a specific area using multiple loudspeakers [3,4]. For example, Iijima a Correspondence to: Yoshiaki Taniguchi.…”

“…For example, Iijima a Correspondence to: Yoshiaki Taniguchi. E-mail: y-tanigu@info.kindai.ac.jp Faculty of Science and Engineering, Cyber Informatics Research Institute, Kindai University, 3-4-1 Kowakae Higashiosaka 577-8502, Japan et al [4] considered to use two parametric array loudspeakers to send voice commands to a specific area where a smart speaker is located. In their method, audio wave are divided into carrier wave and sideband wave, and separately transmitted.…”

When Speakers Are All Ears: Characterizing Misactivations of IoT Smart Speakers

“…A number of researchers have carried out activation attacks on smart speakers, including injecting inaudible and invisible commands into voice assistants using lasers [30], directional amplitude-modulated ultrasound beams [31], or hidden voice commands unintelligible to human listeners [32]. In this paper we specifically focus on activations which are unintentional, or based on misclassifications under normal operational circumstances.…”

Using Inaudible Audio and Voice Assistants to Transmit Sensitive Data over Telephony

“…In another technique that allows a command to be inaudible to humans but recognized by VAs, researchers have used ultrasonic sound carriers to hide the commands from being heard by human [3], [9], [16], [18], [32], [37], [38], [40], [41]. Zhang et al [40] modulated voice commands on ultrasonic carriers (i.e., frequency ≥ 20 kHz) to achieve inaudibility.…”