Year: 2024
Type: article
Abstract: This research aims to enhance the detection of audio deepfakes by developing a real-time, highly accurate methodology that addresses existing technological and ethical gaps in the field. Employing advanced algorithms for feature extraction, the study innovatively utilizes a multifaceted approach by integrating an MFCC-based SVM classifier, which achieved a remarkable 97.28% accuracy, and a Neural Network with attention mechanisms, with a 91.04% accuracy rate. A novel aspect of our methodology is the use of multiple models in tandem to verify the authenticity of input audio, significantly boosting the reliability of detection. Leveraging the 'For-Original' dataset for exhaustive training and validation, our methods have shown exceptional effectiveness in distinguishing genuine audio from synthetic counterparts. These findings not only demonstrate significant improvements in existing deepfake detection techniques but also introduce a novel approach to comparative model analysis. This contribution is pivotal in advancing the field of digital media integrity, offering new avenues for ensuring the authenticity of audio content in the era of sophisticated digital forgeries.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217448687
