Cancelable speaker verification system based on binary Gaussian mixtures

Biometrie systems suffer from non-revocabilty. In this paper, we propose a cancelable speaker verification system based on classical Gaussian Mixture Models (GMM) methodology enriched with the desired characteristics of revocability and privacy. The GMM model is transformed into a binary vector that is used by a shuffling scheme to generate a cancelable template and to guarantee the cancelabilty of the overall system. Leveraging the shuffling scheme, the speaker model can be revoked and another model can be reissued. Our proposed method enables the generation of multiple cancelable speaker templates from the same biometric modality that cannot be linked to the same user. The proposed system is evaluated on the RSR2015 databases. It outperforms the basic GMM system and experimentations show significant improvement in the speaker verification performance that achieves an Equal Error Rate (ERR) of 0.01%.

I. Introduction

Biometric applications are gaining more and more popularity. Regarding the progress of biometric technologies, voice biometrics [1] has the advantage that it can be applied in diverse situations that cover mobile commerce and transactions due to the ubiquity of the microphone. However, there are some issues related to biometric systems [2]. In available biometric systems, the generated biometric templates are similar in different applications because of the employment of the same biometric feature and the same computational procedure. In speaker recognition systems, the speaker model formed from the same voice for different applications is similar. If the speaker model of an application is stolen, it can be exploited to access other applications. This is presenting a threat to privacy. In addition, since biometric characteristics are eternally associated with the user, it is impossible to replace the compromised template with a new one which means non-revocability [3]. Therefore, protection of biometric data is considered as an important requirement to avert privacy and security threats [4]. Currently, biometric templates are protected by storing them in an encrypted place. However, during the biometric comparison, they need to be decrypted inducing a weakness.