Voice based biometric authentication using collapsing classes discriminative space transform

A novel method of voice based biometric authentication system is proposed in this paper. The proposed method utilizes the distance between i-vectors to minimize the intra-class variation and maximizing the inter-class variation. This problem is formulated as a convex optimization problem, such that the i-vectors of the same speaker collapsed to a single point. On the other hand, i-vectors of the impostors are pushed away by a large margin. Additionally, the distance matrix is itself learned using the data (i-vectors of the speakers). The transformation matrix obtained herein linearly transforms the input i-vectors into a new discriminative space. The i-vectors in the new transformed space give improved speaker authentication performance. Speaker authentication experiments are conducted on NIST 2004 and YOHO speaker verification databases. Experimental results obtained indicate a reasonable improvement in the performance, when compared to state of the art voice based biometric authentication systems.