This study presents a non-invasive, and low-cost method to automatically detect Williams Syndrome. It addresses the diagnostic delay arising from phenotypic similarities with common neurodevelopmental disorders, such as Down Syndrome. A total of 72 acoustic features were extracted, encompassing glottal source biomechanics, tremor indicators, and voice quality parameters. Unlike conventional voice analysis methods which typically rely on features such as F0, Jitter, Shimmer, Mel Frequency Cepstral Coefficients (MFCCs), and Cepstral Peak Prominence, our approach utilizes a broader set. Using a Random Forest classifier, we achieved a detection accuracy of 93.98%, revealing that Physiological Tremor and Neurological Tremor are the most effective features for distinguishing normotypic from non-normotypic individuals. In conclusion, the proposed method effectively differentiates Williams Syndrome from a normotypic group using a minimal set of interpretable features. Future work could expand this research to include neurodevelopmental and genetic conditions, such as Down syndrome and Smith–Magenis syndrome. By incorporating a broader range of neurological disorders, the model could aid in developing more comprehensive and robust differential diagnostic tools.