Optimizing Voiceprint Modelling for Biometric Authentica-tion and Security: Applications in Public Safety and Surveillance

Document Type : Original Article

Authors

National Higher Polytechnic School of Douala, University of Douala, Douala, Cameroon

10.21608/ijt.2025.392606.1115

Abstract

A pioneering approach for biometric authentication leveraging voiceprint modelling is metic-ulously engineered here for public security and various surveillance contexts. A hybrid archi-tecture underpins the system melding deep neural networks and hidden Markov models thereby facilitating robust extraction of acoustic features from Mel frequency cepstral coeffi-cients. Developing a voice authentication model capable of operating effectively in real time amidst high background noise and considerable speaker variability is undertaken. Methodology adopted here weds supervised learning with hierarchical Markov models alongside deep neural networks' remarkable generalisation capabilities pretty effectively. Experimental results show remarkably high performance in degraded conditions with accuracy rate hovering just above 95% fairly consistently. Model's originality stems from capacity to thwart voice cloning and deepfake attacks while generally adhering quite rigidly to stringent privacy standards. An ex-ploratory analysis of linguistic biases was undertaken ensuring algorithmic fairness quite vig-orously in a deeply multilingual societal context. Future developments being pondered system integration with multimodal biometrics and deployment on cloud infrastructures happens slowly for enhancing scalability purposes basically. A substantial leap forward occurs here in domain of secure voice authentication with reliability somehow bolstered significantly.

Keywords

Files

Share

How to cite

Statistics