AI Based Multimodel Deepfake Detection System for Audio and News Text

This project presents an AI-Based Multimodal Deepfake Detection System designed to identify fake audio and misleading news text using advanced machine learning techniques. The system performs audio analysis using Mel Spectrogram processing and a custom-trained Convolutional Neural Network (CNN) model to extract important voice features and detect deepfake patterns. It supports common audio formats such as WAV and MP3 and generates a probability score to indicate whether the audio is real or fake, improving reliability and usability. For text analysis, the system uses BART-MNLI classification to evaluate the authenticity of news content through natural language inference. It also integrates NewsAPI to cross-check the given news with real-time trusted sources and applies linguistic pattern analysis to detect suspicious writing styles often found in fake news. Finally, a confidence scoring mechanism combines the outputs from both audio and text analysis to produce an overall prediction, making the system effective for applications such as media verification, cybersecurity,and fake news detection.