Deepfake detection over different media types using deep learning algorithms
B. Arshath, M. Balaji, A. Bhavan Surya, L. Hariprasath · 2024
The goal of this project is to create a deepfake detection model that is adaptable to several media forms, such as photos, videos, and audio recordings, by utilizing state-of-the-art deep learning techniques. It tackles the problems of detecting deepfakes in deteriorated video material and subtle audio alterations by combining convolutional neural networks (CNNs) for image and video analysis with Mel-frequency cepstral coefficients (MFCCs) for audio. The algorithm successfully learns to discriminate real content from altered content with 91 percent accuracy after thorough training on a variety of datasets. It also expands its capabilities to text detection, using a CNN architecture with FastText word embeddings to distinguish between text created by humans and text generated by bots. It performs better than existing deep learning models. This all-encompassing strategy emphasizes how powerful deep learning methods and sophisticated algorithms are at stopping the spread of false information and deepfake technology.