Multimodal Biometric Authentication with Voice: Comparative Evaluation of 15 Systems by Merging Scores

This work extends a multimodal biometric authentication system based on score fusion through the incorporation of voice as a fourth modality, together with face, fingerprint and iris. Using the chimeric user technique to integrate heterogeneous databases, 15 different configurations (4 unimodal, 6 bimodal, 4 trimodal and 1 quadrimodal) are evaluated on 100 virtual identities. The results show that the inclusion of the voice raises the accuracy of trimodal combinations to up to 96.67%, reduces the FAR to 0.00% in one configuration, and achieves AUC-ROC of up to 0.9933. The correlation analysis indicates that voice presents high complementarity in relation to physiological modalities, contributing to more robust and balanced systems between security and usability.