Detecting Fake Audio of Arabic Speakers Using Self-Supervised Deep Learning

One of the most significant discussions in forensics is Audio Deepfake, where AI-generated tools are used to clone audio content people’s voices. Although it was intended improve lives, attackers utilized maliciously, compromising public’s safety. Thus, Machine Learning (ML) and Deep (DL) methods have been developed detect imitated or synthetically faked However, suffered from massive training data excessive pre-processing. To author’s best knowledge, Arabic speech has not yet explored with synthetic fake audio, very limited challenged fakeness, which imitation. This paper proposed a new Deepfake detection method called Arabic-AD based on self-supervised learning techniques both Additionally, contributed literature by creating first dataset single speaker who perfectly speaks Modern Standard (MSA). Besides, accent also considered collecting recordings non-Arabic speakers evaluate robustness Arabic-AD. Three extensive experiments were conducted measure compare well-known benchmarks literature. As result, outperformed other state-of-the-arts lowest EER rate (0.027%), high accuracy (97%) while avoiding need for training.