Leveraging GenAI for Biometric Voice Print Authentication

This paper presents the development of a secure voice authentication system that delivers an inclusive solution for all users, including those with disabilities. Leveraging a Text-Dependent Active Verification process, the system combines a spoken passphrase with voice biometric coefficients and audio vector embeddings for reliable user verification. A vector database is used to efficiently store data and perform similarity retrieval. Initially, the system achieves a 71% spoof detection accuracy, ensuring that only genuine samples proceed to the embedding stage, where it attains a 55.21% accuracy in vector embedding and similarity retrieval. Furthermore, this approach paves the way for user-specific voice-controlled environments. Overall, this study underscores the transformative potential of voice biometric authentication by merging cutting-edge signal processing with sophisticated machine learning techniques, setting a benchmark for future research in balancing robust security measures, user convenience, and ethical inclusivity.