Performance evaluation of phonotactic and contextual onset-rhyme models for speech recognition of Thai language

This paper proposed two acoustic modelings of the onsetrhyme for speech recognition. The two models are Phonotactic Onset-Rhyme Model (PORM) and Contextual Onset-Rhyme Model (CORM). The models comprise a pair of onset and rhyme units, which makes up a syllable. An onset comprises an initial consonant and its transition towards the following vowel. Together with the onset, the rhyme consists of a steady vowel portion and a final consonant. The experiments have been carried out to find the proper acoustic model, which can accurately model Thai sound and gives higher accuracy. Experimental results show that the onset-rhyme model excels the efficiency of the triphone for both PORM and CORM. The PORM achieves higher syllable accuracy than the CORM 2.74 %. Moreover the onset-rhyme models also give a more efficiency in term of system complexity compared to the triphone models.