A COMPREHENSIVE SYSTEM FOR DETECTING DEEPFAKE VIDEOS AND AI-GENERATED TEXT

Abstract

Artificial Intelligence (AI) has changed from simple rule-based systems to complex machine learning models. However, this progress has also made things complicated by introducing the problem of fake content that has spread all over digital platforms. Fake contents in text and video formats pose serious threats like dissemination of misinformation to manipulating public opinion. To combat this challenge, a comprehensive detection system is implemented that uses three different AI models which are Dessa for deepfake detection, LLMDet for finding the source of text and DeepSpeech for converting speech into-text. Each model serves a vital function; whereas Dessa relies on sophisticated algorithms to differentiate between real and manipulated clips; LLMDet identifies computer-generated texts; while DeepSpeech precisely converts spoken words into written texts. This system also consists of a reporter module that generates JSON-formatted structured reports encompassing findings and interpretations from each detection task. The system was tested using both synthetic and natural data which include videos created by synthesia.io tool as well as those downloaded from online sources. Through rigorous testing, it was demonstrated that the multi-model AI is highly effective with solid performance results. This work can help researchers who study fake videos and AI-generated texts. It brings together different AI models to solve difficult problems, which could be useful in spotting and dealing with fake content.