Optimized Feature Selection and Classification of Arabic Speech Disorders: A Case Study of Letter /r/ Difficulties

Abstract The reliable and automatic recognition of speech sound problems is critical for assisting in the early detection and treatment of defective phonological processes in children. This study addresses the issue of speech sound error classification in Arabic children when they mispronounce Arabic words, including the letter r (pronounced /ra/). A determination was made regarding whether a speech sound problem existed when the letter appeared at the beginning, middle, or end of words. The speech signal was characterized by different classifier models using the number of times and frequency features to aid in the automatic diagnosis of speech disorders in children. Utilizing a real-world library of voice recordings, the highest accuracy of 92.4% was achieved using a bagged trees classifier with a combination of effective frequency features under the holdout method.