Abstract

Voice communication continues to be a primary channel for both personal and institutional interactions, including banking, healthcare, and enterprise operations. However, the trust inherently placed in voice calls has led to a significant rise in voice-channel exploits such as vishing, caller-ID spoofing, robocalling, and more recently, AI-driven voice cloning attacks. These threats increasingly bypass traditional spam filters and blacklist-based defenses due to the use of VoIP infrastructure, dynamic number spoofing, and realistic synthetic speech. This paper proposes the SecureCall Intelligence Engine (SCIE), an integrated and adaptive defense framework designed to detect and mitigate malicious, fraudulent, and AI-generated voice calls in real time. SCIE operates across three complementary layers: (i) telephony metadata analysis to detect anomalies in call origin, behavior, and frequency; (ii) audio spoof and deepfake speech detection using spectro-acoustic signal analysis; and (iii) semantic and contextual analysis of transcribed speech to identify social-engineering attempts. A decision-fusion model aggregates risk scores from all layers to determine appropriate actions—allow, warn, or block—while minimizing disruption to legitimate communication. Experimental evaluation on a diverse dataset of legitimate, spam, replayed, and AI-generated calls demonstrates that SCIE achieves a detection rate of approximately 96% while maintaining a false-positive rate below 3%. These results highlight the potential of the proposed system to significantly reduce financial loss and identity compromise associated with voice-channel fraud. SCIE is designed for scalable deployment across PSTN and VoIP networks, supporting continuous learning to adapt against evolving threats.