Research On Ultrasonic Attacks To Voice Assistant Intelligent Equipment

Nowadays,with the increasing popularity of voice control systems,the functions of the voice interface become more and more powerful.Voice control system is used in mobile payments,smart home control and even in automatic driving system.This thesis is to conduct research on intelligent phonetic system from the aspect of hardware for the security of the upper software coupling.The research direction of the thesis is to design the voice command that a human ear cannot hear,i.e.,Silent Ultrasound Attack.Such a sound is modulated into the ultrasonic range(e.g.,f > 20 KHZ).By using the nonlinear effect of microphone circuit,voice commands are modulated to high frequency signals that the human ear cannot hear to attack smart devices(such as mobile phone).At the same time,based on the principle of FFT,CVI system analysis function is used to analyze the high frequency signal to measure the signal-to-noise ratio(SNR),to indicate that the high frequency signal attack is real and effective.Since the speech recognition system such as MIUI and Google Now has become a popular method of human-computer interaction,and has tied various system into a voice controllable system(VCS).In the research work,this thesis will use voice synthesis technology and extract features from the recorded voice to prove that the modulated voice command can be successful demodulated and understood by the voice recognition system.Experiments are conducted to use the ultrasonic attacks that cannot be recognized by the popular speech recognition systems to attacks mobile phones,tablets,and other smart devices with microphone,including Xiao Mi,Huawei and other brands.By conducting some attacks for proof of concept,a series of voice command are used as the input,including to open Face time on iPhone’s Siri mode,and to attack the Xiao Mi mobile phone to make calls.Finally,the solutions to hardware and software defense are proposed,and it is suggested to redesign voice control system so as to identify the voice attacks that the human ear cannot hear.