Unmasking the Truth: A Deep Learning Approach to Detecting Deepfake Audio Through MFCC Features
Islam Altalahin, Shadi AlZu’bi, Assal A. M. Alqudah, Ala Mughaid
Abstract
Deepfake content is artificially created or altered using artificial intelligence (AI) methods to appear real. Synthesis can include audio, video, images, and text. Deepfakes may now produce content that looks normal, making it more difficult to identify. Significant progress has been made in identifying video deep fakes in recent years; However, most of the investigations into voice deep fake detection have used the ASVSpoof-2019 dataset and several machine learning and deep learning algorithms. This research uses machine-based and deep-learning approaches to identify fake audio. Melted frequency cepstral coefficients (MFCCs) are used to extract the most useful information from the sound. We choose the 2019 ASVSpoof dataset, which is the latest reference dataset. Experimental results show that Convolutional Neural Networks (CNN): (CNN-LSTM) outperformed other machine learning (ML) models in terms of accuracy, achieving an accuracy of up to 88%.