The increasing sophistication of voice cloning technologies has led to serious challenges in voice-based authentication and forensic analysis, particularly for regional languages such as Marathi. This study presents a system for detecting cloned versus original speech using acoustic as well as spectral-temporal features. While features such as formant frequencies, pitch, jitter, and shimmer were also explored, these voice quality measures exhibited limited discriminative power when used independently, particularly against advanced synthetic voices. Therefore, the system emphasizes Mel-Frequency Cepstral Coefficients (MFCC) and its dynamic derivatives (delta and delta-delta), which provide a more robust representation of speech patterns. A Gaussian Mixture Model (GMM) classifier is used to distinguish between original and cloned Marathi voice samples based on MFCC features. The system demonstrates high accuracy and reliability, validating the effectiveness of spectral-based analysis combined with GMM modeling in detecting voice spoofing.