Defending Against Voice Spoofing: A Robust Software-Based Liveness Detection System

Abstract

The recent proliferation of smartphones has been the primary driving factor behind the booming of voice-based mobile applications. However, the human voice is often exposed to the public in many different scenarios, and an adversary can easily "steal" a person's voice and attack voice-based applications with the help of state-of-the-art voice synthesis/conversion softwares. In this paper, we propose a robust software-based voice liveness detection system for defending against voice spoofing attack. The proposed system is tailored for mobile platforms and can be easily integrated with existing mobile applications. We propose three approaches based on leveraging the vibration of human vocal cords, the motion of the human vocal system, and the functionality of vibration motor inside the smartphone. Experimental results show that our system can detect a live speaker with a mean accuracy of 94.38% and detect an attacker with a mean accuracy of 88.89% by combining three approaches we proposed.