Deepfake audio is an emerging concern because of the huge implication for security, privacy, and trust. In this paper, we present an efficient framework for deepfake audio detection utilizing the OpenSMILE toolkit for feature extraction. Precisely, we used key speech characteristics, including prosodic and spectral features for training the classifier models. We employed several supervised machine learning algorithms, including Random Forest, Decision Tree, Support Vector Machine, and Long Short-Term Memory (LSTM), to differentiate between real and synthetic audio. The proposed approach has been tested on the Fake-or-Real dataset with different combinations of extracted relevent features. Experimental results prove that the best performance is obtained by the model Random Forest, giving an accuracy of 95.12% and 0.9739 AUC on the Norm dataset. Furthermore, the study reveals that using the top 30 features extracted from OpenSMILE can effectively detect deepfake audio with minimum computational overhead.