Abstract
The rapid advancement of AI-based voice synthesis and cloning technologies has introduced significant security challenges in modern digital communication systems. Voice-based cyberattacks such as vishing (voice phishing) and impersonation scams have increased by over 70%, resulting in severe financial losses and emotional distress. As AI-generated synthetic voices become increasingly realistic and indistinguishable from human speech, traditional voice authentication and verification methods are no longer sufficient to ensure security. This paper proposes an intelligent deepfake voice detection system leveraging audio signal processing and machine learning techniques. The system extracts discriminative acoustic features such as Mel-Frequency Cepstral Coefficients (MFCC), pitch variations, and spectrogram-based patterns from voice samples. These features are then used to train a classification model capable of distinguishing between genuine human voices and AI-generated deepfake audio. The proposed approach enhances the accuracy and robustness of cybercrime detection by effectively identifying manipulated voice samples. Experimental evaluation indicates that the system achieves up to 97.5% detection accuracy, significantly outperforming traditional methods, thereby strengthening secure authentication mechanisms in digital communication environments. The study contributes to the development of reliable AI-driven cybersecurity frameworks for mitigating voice-based cyber threats.