Attack on practical speaker verification system using universal adversarial perturbations
This work exposes a critical security flaw in speaker verification systems used in authentication scenarios, enabling targeted attacks that bypass replay detection, which is incremental but practically impactful for system security.
The authors tackled the vulnerability of practical speaker verification systems by crafting universal adversarial perturbations that, when played separately while an adversary speaks, cause the system to misidentify the adversary as a target speaker with a 100% success rate in physical experiments, while only increasing the word error rate for speech recognition by 3.55%.
In authentication scenarios, applications of practical speaker verification systems usually require a person to read a dynamic authentication text. Previous studies played an audio adversarial example as a digital signal to perform physical attacks, which would be easily rejected by audio replay detection modules. This work shows that by playing our crafted adversarial perturbation as a separate source when the adversary is speaking, the practical speaker verification system will misjudge the adversary as a target speaker. A two-step algorithm is proposed to optimize the universal adversarial perturbation to be text-independent and has little effect on the authentication text recognition. We also estimated room impulse response (RIR) in the algorithm which allowed the perturbation to be effective after being played over the air. In the physical experiment, we achieved targeted attacks with success rate of 100%, while the word error rate (WER) on speech recognition was only increased by 3.55%. And recorded audios could pass replay detection for the live person speaking.