Detection of Deepfake Audio Using Deep Learning

作者：A. Lakshmi, V. Sindhuja, B. Meghana, Gautam Gupta · 年份：2024 · DOI：10.1109/icces63552.2024.10859752 · 被引用次数：3 · 研究领域：Digital Media Forensic Detection、Music and Audio Processing、Generative Adversarial Networks and Image Synthesis

Deepfake technology employs AI algorithms to create misleading digital information, such as photos, movies, and audio recordings. Deepfakes can now create realistic-looking content, making detection more challenging. Most studies on detecting audio deepfakes use various machine learning and deep learning algorithms with the ASVSpoof dataset, despite recent advancements in video deepfake detection. MFCCs are used to extract useful acoustic information. The dataset was constructed using a text-to-speech model and then partitioned into a few datasets. Datasets are divided into subsets according to bit rate and audio length. The experimental results demonstrate that support vector machines (SVMs) outperformed other machine learning (ML) models in terms of accuracy. Our objective is to enhance the model's efficiency and detection accuracy through the use of deep learning techniques with CNN and LSTM neural networks. Using Generative Adversarial Networks (GAN) as a classifier, that can determine whether audio is real or fraudulent. And a Graphical User Interface (GUI) was further implemented for a better visual representation of the output, that is obtained