Recognition of Arabic speech sound error in children

The accurate and automatic recognition of speech sound errors in children is crucial to facilitate the early detection and correction of any faulty phonological process in their early life. This paper addresses the problem of speech sound error classification in native Arabic children when they wrongly pronounce Arabic words containing the letter r (pronounced as /ra/). We identify whether the speech sound error occurs when the letter appears at the beginning, middle, or end of the words. To classify the spoken words, we represent the speech signal with Mel Frequency Cepstral Coefficients (MFCC) features and then train a probabilistic classifier. We evaluate the performance of our proposed approach using a real-world database consisting of speech recordings from native Arabic speaking children. The proposed method achieves a classification accuracy of 71.75%, 77.20%, and 74.06% on average, for speech sound error with Arabic words containing the letter r at the beginning, middle, and end of the words, respectively. These results are superior to those obtained with Hidden Markov Model: another state-of-the-art method on the same dataset.