DeepGuard: A Real-Time Multi-Modal Deepfake Detection and Classification System

The current state of Generative AI presents a critical point because users can easily merge fabricated media with authentic content to create deepfakes. The tools generate creative content but they create digital skepticism while spreading false information which damages forensic evidence. The research presents DeepGuard as a real-time multimodal system which identifies deepfakes through video, audio and image analysis. The system uses Convolutional Neural Networks (CNNs) to extract facial information from snapshots and Mel-spectrogrambased CNNs to detect speech patterns through temporal and spectral analysis. The selfsupervised Wav2Vec2-style module enhances the model's ability to detect audio manipulation. The system uses individual detectors to produce scores which merge into a confidence value that enhances both reliability and accuracy of classification results. The evaluation results show DeepGuard successfully identifies authentic content from manipulated material while maintaining its performance when compression artifacts and background noise affect the data. The system operates in real-time through its modular design which allows for simple expansion and maintains flexibility against evolving deepfake creation techniques. The research team plans to study how to merge different detection methods with automatic forgery type identification to enhance security applications.