Voice-recognition technology is getting better and better, but if you’re using it for security applications such as spoken passwords, it is not infallible. A team of researchers at Florida State University in Tallahassee may have a solution for spoofing attacks on voice biometrics, however — and it involves sonar.
Called VoiceGesture, the system developed by the team reappropriates your smartphone as a Doppler radar, transmitting a high-frequency sound from the device’s speaker and then listening to the reflections on the microphone when a person says their passcode. Compared to some of the weaknesses involved in regular voice biometric systems, such as the risk of someone impersonating your voice or using a recorded sample, it’s far more effective.
“Our system evaluation involves 21 users with thousands of passphrases and three types of smartphones: Samsung Note 5, Note 3, and Galaxy S5,” Jie Yang, an assistant professor in the Department of Computer Science, told Digital Trends. “Experimental results show that VoiceGesture achieves over 99 percent spoofing attack detection accuracy at around one percent equal error rate (EER).”
Florida State isn’t the only research institute currently investigating ways to make voice biometrics more secure. Recently, we covered a research project from the University of Michigan, which utilizes an accelerometer-based wearable accessory — currently a necklace, earbuds, or glasses attachment — to measure the unique skin vibrations in a person’s face, throat, or chest when they talk.
But what makes VoiceGesture stand out is the fact that it requires no additional hardware in order to work. Yang says that it can be integrated with existing smartphone operating systems and mobile apps for more secure device login, without having to change the physical devices themselves. He told us that the team has already approached Google, who are reviewing the proposed technique. “We also plan to reach out to other smartphone manufacturers like Samsung and Huawei,” Yang said. Should it prove as secure as the team claims, let us hope this turns up on a next-generation smartphone in the near future.
A paper describing the work, titled “Hearing Your Voice is Not Enough: An Articulatory Gesture Based Liveness Detection for Voice Authentication,” is available to read here.
