Building Automatic Speech Recognition Systems for Moroccan Dialect: A Phoneme-Based Approach

Building efficient acoustic models for dialects is a major challenge in Automatic Speech Recognition (ASR) systems. In this paper, we investigate the Moroccan Fessi dialect speech recognition system based on phoneme modeling. We employed a combined approach, including the Hidden Markov Model (HMM) and the Gaussian Mixture Model (GMM). Also, the ASR dialect specificity was analysed, including phonemes nature and phonetic inventory. Our results show the best performance was found by using 3 HMM and 4 GMM configurations, achieving an accuracy of 97.33%. Additionally, we observed that the digits containing voiced pharyngeal phonemes, particularly the phoneme /ʕ/, achieved the highest recognition rate, while words containing the phoneme /s/ exhibited multiple substitutions.