Personal authentication system utilizing voice distortion through guitar effectors

Abstract

In this study, we propose a personal authentication system using processed speech that can only be understood by those who know the original voice. The proposed system presents non-invertible processed speech signals, including distortion, equalization, and echo, to the voice representing the words registered by the legitimate user at the time of authentication, just like a signal processed through guitar effects pedals. The strength of the effect is set within the range where the legitimate user can barely be identified, and the combination of effect parameters can be changed each time it is presented, making it harder for everyone other than the legitimate user to break through. We confirmed the system's effectiveness by utilizing a dataset of Japanese words with defined word familiarity. Through an authentication experiment in which the system was automatically adjusted to an effect strength that was barely recognizable to legitimate users, it was found that legitimate users could identify 75.6% of the processed speech signals. In comparison, non-legitimate users could decode only 38.8% of the words. These results indicate the feasibility of a personal authentication system utilizing distorted speech signals.