HMM-GMM classification for articulation disorder correction among Algerian children

Abstract EN

In this paper, we propose an automatic classification for Arabic phonemic substitution using an Hidden Markov Model / Gaussian Mixture Model (HMM / GMM) systems.

The main objective is to help Algerian children in the correction of articulation problems.

Five cases are analyzed in the experiments, 20 Arabic words are recorded by a 20 Algerian children, with age range between 4 and 6 years old.

Signals are recorded and stored as wave format with 16kHz as sampling rate, 12 Mel Frequency Cepstral Coefficients (MFCC), with their first and second derivates, respectively / and // are extracted from each signal and used to the training and recognition phases.

The proposed system achieved its best accuracy recognition 85.73 %, with 5-stats HMM when the output function is modelled by a GMM with 8 gaussian components.