DeepGuard AI: A Multi-Modal Deepfake and Misinformation Detection Framework Using Hybrid Deep Learning and Audio-Level Feature Fusion

The rapid advancement of deep learning techniques has significantly enhanced the realism of synthetic media, commonly referred to as deepfakes. While these technologies offer benefits in controlled applications, they pose serious threats such as misinformation dissemination, identity fraud, and digital manipulation. This paper presents DeepGuard AI, a multi-modal deepfake detection framework that independently analyses image, audio, and video content, complemented by a knowledge-based misinformation verification module. The proposed system employs Convolutional Neural Networks (CNNs) for image-based analysis, a hybrid audio detection mechanism that integrates waveform and spectrogram representations, and a video detection model based on a convolutional backbone with temporal feature modelling. A key contribution of this work is the hybrid audio feature fusion approach, which enhances robustness by combining temporal and frequency-domain characteristics. Furthermore, the system incorporates a semantic verification component that analyses extracted textual content using a large language model API to validate contextual authenticity. Experimental results demonstrate that the proposed approach achieves improved performance across multiple modalities, particularly in audio detection, validating the effectiveness of hybrid feature learning. The system provides a scalable and practical solution for real-world deepfake detection and misinformation analysis.