Audio Deepfake Detection by using Machine and Deep Learning
H. Hakan Kılınç, Faruk Kaledibi
Abstract
Fake voices are one of the hottest topics in cyber security, forensics, and social media. There are a variety of usage scenarios, from speech disorders to fake news to telephone and financial fraud. With products using artificial intelligence technology such as Google Audio LM, it is possible to produce realistic, well-structured, and consistent sound sequences. These products, although synthetic, can accurately replicate intonation, accents, and other unique features by mimicking the human voice. A solution using machine and deep learning methods to recognize fake voices is proposed. In the feature extraction stage, Mel-frequency cepstral coefficients (MFCCs) are used. Then these features are classified using machine and deep learning-based models. According to the results obtained, the sound is judged to be real or fake.