An Architecture for Voice-Based Authentication and Authorization with Deepfake Detection

Abstract

Voice biometrics offer a convenient and secure authentication method, but the rise of sophisticated deepfake technology presents a significant challenge. This work presents an architecture for voice-based authentication and authorization that integrates deepfake detection to mitigate this risk. This paper explores the design of this cloud-native architecture, leveraging Amazon Web Services (AWS) services for orchestration and scalability. The system combines cutting-edge Al models for robust voice-printing and real-time deepfake analysis. We discuss multi-factor authentication (MFA) strategies that provide layered defense against unauthorized access. Two specific use cases are explored: identity verification and secure approval of banking transactions. This paper addresses key considerations for real-world deployment, including system resiliency, cost-effectiveness, and the efficiency of the Al models under varying conditions. We evaluate the architecture's suitability as a two-factor authentication (2FA) solution, focusing on the accuracy of deepfake detection and the rates of false negatives and false positives.