This paper introduces an automated system to classify the severity of the cleft palate using machine learning, tailored to address the unique challenges of early detection in areas limited in resources. Using long-short-term memory (LSTM) networks, the system captures and analyzes temporal dependencies in speech data to classify severity into three levels. Mild, Moderate, and Severe. Unlike existing methods, our approach emphasizes accessibility and adaptability to underserved regions by focusing on speech-based diagnostics that do not rely on advanced medical imaging. The system achieved a training accuracy of 65.99% and a testing accuracy of 50.00%. Although these metrics indicate room for improvement, our work lays a foundation for integrating machine learning into the diagnosis of cleft palate, addressing gaps in resource accessibility and early intervention. Future enhancements will include transfer learning and hyperparameter optimization to refine accuracy and broaden applicability. This study aims to advance the field by providing an accessible and speech-driven diagnostic tool for underserved communities.