Near Real-Time Deepfake Audio Detection based on Speaker Diarization Techniques

The spread of the deepfake phenomenon, and particularly of the audio deepfake, imposes the design of reliable detection techniques. Continuous efforts are being made to design reliable methods and systems to stop the spread of deep-fake audio and guarantee their detection. In this paper, we propose a real-time processing pipeline for audio deepfake detection. The novelty of our approach is that our system fully analyzes audio signals, going beyond feature extraction and analyzing the time and frequency components of the signals. This detailed examination is essential for identifying authentic or fraudulent audio signals. To improve signal clarity and increase the accuracy of audio deepfake detection, we also focus on data augmentation. The results obtained show that the proposed pipeline provides a reliable way to distinguish between real (authentic) and fake (fraudulent) audio content.