AI-based DeepFake Detection

Abstract

Deepfake technology raises questions regarding security and disinformation by making it possible to create incredibly lifelike fake photos and videos. In order to identify deepfakes involving particular public figures, this paper suggests a real-time AI-based system that makes use of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks. Following frame extraction and preprocessing (such as face detection and alignment), CNN-based spatial feature extraction and LSTM-based temporal analysis are carried out by our system. Included is an audio analysis module that looks at cloned voice patterns using CNN/RNN and spectrogram features. The entire pipeline is implemented as a web application, which uses adaptive preprocessing and data augmentation to support varying lighting conditions and video quality. We anticipate that the system will attain detection accuracy on par with the most advanced techniques found in the literature. The web interface lets you upload or stream stuff and get authenticity checks right away.