Context-independent acoustic models for Thai speech recognition

The main purpose of this paper is to investigate and construct the suitable acoustic models for Thai speech recognition. Many methods are presented in this paper including single phone modeling, base phone modeling, and tone modeling. The context-independent models and gender-independent models are often used in the acoustic modeling. Many modeling techniques are given from the linguistic-phonetic knowledge of Thai language. Experiments to evaluate the recognizer performance in word recognition are performed. The word utterances are used as training and testing data in our experiments. The experiments measure the word accuracy of each modeling methods when a number of mixtures are varied. The best accuracy is 87.03% from the single phone models without tone modeling (31 models) and 82.03% from the base phone models with tone modeling on vowels (171 models).