DeepShield: AI-based Real-Time Deepfake Scam Prevention using Multi-Modal Voice and Video Fraud Detection

Abstract

Deepfakes generated using advanced AI techniques, including Generative Adversarial Networks (GANs) and diffusion models, have become increasingly realistic, posing serious risks to digital trust, financial security, and public safety. Existing detection methods often focus on a single modality or require significant computational resources, limiting their practical deployment. This paper presents DeepShield, a real-time multi-modal deepfake detection platform that integrates video, audio, and URL-based scam analysis within a unified framework. The system employs EfficientNet-B0 for video analysis, spectral feature-based audio detection, and URL inspection using domain reputation, SSL validation, redirects, and HTML content analysis. Built with a Flask REST API and React frontend secured by JWT authentication, DeepShield achieved over 87% detection confidence while maintaining real-time performance across diverse input sources.