Deepfake audio is growing so convincing that it threatens our ability to trust phone calls, voice notes, and news broadcasts. To counter this, we combine state-of-the-art features, such as Mel Frequency Cepstral Coefficients (MFCCs) with lightweight signal descriptors–spectral centroid, bandwidth, roll-off, and zero-crossing rate–and three Fractal Dimension measures: Hurst R/S exponent, Katz fractal dimension (FD), and Higuchi FD. These fractal features capture subtle self-similar patterns in genuine speech that synthetic voices still struggle to replicate.