Investigation on Secure Speaker Verification Based on Vocal Liveness Detection

Period: April 2014 – March 2017
Grant type: Japan Society for the Promotion of Science, Grants-in-Aid for Scientific Research (B)
Issue number: 26280066
URL: https://kaken.nii.ac.jp/ja/grant/KAKENHI-PROJECT-26280066/

Among biometric authentication, voice-based biometric authentication (speaker verification), which can be easily used with a microphone, has attracted attention and has been introduced into society. However, it has been reported that speaker verification can be spoofed by exploiting the dramatic improvements in voice synthesis and voice conversion technology in recent years. Therefore, in this research, we pioneered the introduction of liveness detection technology to speaker verification, which determines whether the subject of authentication is truly a living human. We constructed the world's first large-scale spoofed voice corpus and also held an event to compete for the liveness detection accuracy of speaker verification systems. Furthermore, we demonstrated that it is possible to distinguish whether a voice is spoken by a human or played on a speaker by using pop noise, a phenomenon in which the voice waveform is distorted by exhalation.