Optimized Voice Spoofing Detection Using One Class Learning to Combat Identity Theft and Fraud

Abstract

Voice recognition is essential to secure authentication systems, with increasing digitization. But the rise of advanced spoofing attacks emphasizes significant flaws in modern technologies, which frequently fall short in the midst of emerging threats. By employing a One-Class Learning technique and focusing only on genuine voice samples to detect deviations indicative of spoofing, this work addresses these vulnerabilities. To increase the system's adaptability, the methodology combines autoencoders for pattern recognition with data augmentation techniques like pitch, speed fluctuations. High detection accuracy can be observed in the results, regardless of various difficult circumstances. The architecture provides a scalable solution by enabling deployment in resource-constrained devices