Hybrid approach DTW/HMMC for the recognition of the isolated Arabic words

Hidden Markov models (HMMs) are stochastic models capable of statistical learning and classification. In this paper a new approach based on Gaussian mixture hidden Markov models (GHMM) is introduced, evaluated and compared with traditional approach for isolated word recognition system. A new parameter is introduced in the HMM as defined as the prototype word represented the class word, is calculated in the training phase by the iterative algorithm based the dynamic warping time technique (DTW). The full and diagonal covariance matrix is used for word recognition. The experiments concluded the potentiality from the hybrid models, compared to the traditional models.