Cancelable Voiceprint Template Based on Chaff-Points-Mixture Method
Wenhua Xu, Minying Cheng · 2008
Template security is crucial since biometrics don't change, and privacy concerns cause users to be fearful of intrusions by government and other organizations. This paper proposes an approach to address these two concerns. "Chaff Points'' are added to the MFCC feature matrix such that it is difficult for an adversary to separate the genuine points from the mixture. A prime accumulator is designed to help the server to obtain the raw feature, and a non-linear transformation of the MFCC is captured to prevent a dishonest server to abuse the raw feature. Unlike previous works that analyzed fingerprints limited to one dimension, this paper focuses on the voiceprint that use two dimensional feature. In addition, the server not only can authenticate the entity by template match, but by polynomial reconstruction. Experimental results demonstrate the validity of this approach, and attack probability analysis models a secure system.