LOL: localization-free online keystroke tracking using acoustic signals

Contents typed via keyboards prove to be vulnerable to attacks based on acoustic emanations analysis. However, previous works achieve the attacks under controlled environment, e.g., neglecting the noises or requiring the keyboard to be located in fixed locations. In this study, we present a localization-free online keystroke tracking system (LOL), which enables people to use prior knowledge obtained from the keyboard in one location to recognize real-time keystrokes of the same type of keyboard in any other places, despite various background noises. Combined with support vector machine, we design an detection model to separate keystroke signals from noises. By analyzing the properties of acoustics transmission, we propose an angle-based sampling method with a single microphone to decrease the dependence on certain locations, and it also increases the diversity of signals in the meantime. Our real-world experiments demonstrate a 99.47% keystroke detection rate, a 97.27% recognition accuracy under ideal condition, and an 84.55% content recovery accuracy despite changing locations of the keyboard. Most commercial off-the-shelf sound recording devices, e.g., smartphones, can be used in our system to record acoustic emanations from keystrokes. LOL could attract more community to study security of keyboard devices and promote users to enhance privacy protection awareness.