Sniffing Computer-usage Behaviors Through Acoustic Side Channel

Nowadays,with the rapid development of computer technology,commercial computers have been widely used in government agencies,military systems,companies and schools.With the application of commercial computers in the social production activities,people's working efficiency has been greatly improved.However,there are security and privacy issues also attracted more and more attention.As a part of the security problem,the side-channel attack has always been a research area that draw so much attention.Normally,In a side-channel attack scenario,the attacker uses the special device to collect the magnetic field change,vibration,temperature,sound and so on that emitted by the target.Then the attacker eavesdrops valuable information by analyzing the data.And with the development of the acoustic sensing technology,there have been some researchers use the acoustic emanation and relevant technology to implement side-channel attack.They eavesdrop on the entered content by combining the sound collected by the microphone with other emanations.Consequently,there exist security problems that lead to keystroke content leakage when using the keyboard and mouse.Nonetheless,it's not all of the problem.Computer has been widely and frequently used in our modern daily life for business or entertainment.People can enjoy a variety of services such as text entry,programming,shopping,gaming,and so on with a computer kit.Although computers have facilitated our lives to a great extent,we uncover an ignored fact that behaviors of using a computer can be eavesdropped by an attacker through sniffing emitted acoustic signals from keyboard and mouse.Compared with other side-channel attacks demonstrated in previous works,behavior eavesdropping via acoustic side channel has lower requirement in terms of hardware device and is easier to implement in real world.In this paper,we design and implement a system,namely Behavicker,to validate the feasibility of this kind of attacks.Different from conventional activity recognition,Behavicker infers high-level computer-usage behaviors based on the recognition of basic keyboard and mouse events with a semantics preserving multi-scale learning scheme.Real-world experiments show that Behavicker can recognize six Meta events with an accuracy of 88.3% in an office environment,and can infer computer-usage behaviors with an accuracy of 82.7%.