Recognition test on highly newly robust Malay corpus based on statistical analysis for Malay articulation disorder

Abstract

In designing the Malay language database for articulation disorder, the priority is more on Malay alveolar target words where the important set of words had been used for therapy training exercise especially for the patient at Sekolah Kebangsaan Pendidikan Khas (SKPK), Johor Bahru [9]. The use of manual or traditional technique by speech-language pathologist (SLP) at SKPK is not efficient anymore because it can lead to time consuming and require a lot of involvement of SLP for each therapy session for the ratio of 2:1000 of SLP to patient. Therefore this paper describe the computerized technique that been use in speech recognition where few experiment had been conducted in the process of building the Computer-based Malay Language Articulation Diagnostic System that can be use specifically for speech articulation disorder. The technique use for statistical and processing the word behind this system is Hidden Markov Model (HMM). From the total 108 target words that been collected, few words been selected to run the experiment by using voice sample of real patient The experiment results shows the accuracy of the recognition rate has achieved about 97% from the overall sample and few words can be claimed as 'major spoken' mistake that always happen in speech articulation disorder case. The experiment regarding to voice sample evaluation had also been done to find the total accuracy rate for Malay alveolar consonants.