Research On Ultrasonic Attack And Defense Of Voice Control System In Multiple Scenarios

In recent years,with the development of artificial intelligence and Internet of Things(Io T)technology,voice interaction has become one of the main ways of humanmachine interaction,and voice control systems(VCS)have been widely popularized in various electronic devices.Users can easily interact with devices and perform many operations through voice commands.However,the security and privacy issues related to VCS are becoming increasingly serious.Ultrasonic attacks are a form of attack that injects attack commands into VCS using carefully constructed ultrasonic signals that are inaudible to the human ear.Attackers modulate the voice commands into the ultrasonic frequency band and use the nonlinear characteristics of the microphone at the receiving end to achieve demodulation and recover the attack signal,thereby injecting malicious voice commands.This thesis explores the method of transmitting ultrasonic signals through solid media,extends the implementation scenarios of ultrasonic attacks,and proposes an active defense scheme for existing defense measures’ shortcomings.Existing ultrasonic attacks use air as the propagation medium.The propagation characteristics of ultrasonic waves in the air limit the attack distance,require high emission power,have strong directionality,and cannot exceed the line of sight.These limitations make it difficult for ultrasonic attacks to be applicable to complex scenarios in real life,such as scenarios with greater distances or scenarios with obstructed attacki ng devices and target devices.Therefore,the existing ultrasonic attacks have a strong threat but limited actual harm.This thesis implements three improvements based on existing ultrasonic attacks.First,a new ultrasonic attack mode is designed and implemented,which uses Lamb waves as the waveguide to propagate through a solid thin plate,and uses PZT piezoelectric material instead of high-frequency speaker as the transmitting device to expand the attack scenarios.Second,deception of speaker verification is achieved by using two speech synthesis methods,voice splicing and voice conversion,increasing the attack’s effectiveness.Third,attack characteristics are hidden by signal design to evade existing defense mechanisms,improving the attack’s concealment.This thesis explores the factors that affect the attack effects in three scenarios: ultrasonic propagation in the air,ultrasonic propagation through the unobstructed thin plate,and ultrasonic propagation through the obstructed thin plate.The results show that the ultrasonic attack propagated through the solid thin plate has a longer attack distance,is less affected by background noise,can achieve omnidirectional attack without being affected by obstructions on the plate,requires less power,and can achieve an attack distance of 900 cm with a power of only0.8 W,which has significant advantages compared with ultrasonic propagation in the air.Finally,due to the high cost or bypassing of existing defense solutions,this thesis proposes an active ultrasonic interference system based on the principle of sound wave interference,which cancels or weakens the malicious signals of ultrasonic attacks through real-time phase inversion and noise addition.Experiments have verified that the system can effectively protect VCS when normal speech signals are minimally affected.In addition,usage suggestions are proposed to improve the security of VCS by analyzing the limitations of ultrasonic attacks,helping legitimate users defend against ultrasonic attacks without external defense device.