Deepfake detection for preventing Audio and Video frauds using Advanced Deep Learning Techniques

Deepfakes, powered by generative adversarial networks (GANs), automate the creation of deceptive videos. To combat this menace, the study assembles an advanced model that fuses ResNeXt, Long Short-Term Memory (LSTM), and ResNet architectures, renowned for their efficacy in handling visual and temporal aspects, for the detection of deepfakes in audio and video content. Pre-processing, facilitated by a Multi-Task Cascaded-Convolutional Neural Network (MTCNN), accurately extracts facial regions. The model undergoes rigorous evaluation across three key datasets which are, FaceForensics++(FF-DF), Celeb-DF, and Facebook Deepfake Detection Challenge (DFDC), affirming its real-world readiness and exceptional accuracy. The combined models consistently deliver highly precise results, maintaining their robustness against evolving deepfake technologies.