Sixth International Conference on Spoken Language Processing
This paper presents Thai monophthongs recognition. The monophthongs were qualitatively recognized by the 3-state left-to-right continuous density hidden Markov model. The LPC cepstral coefficients were used as feature which represented specch signal. The temporal cepstral derivative was additionally utilized in order to compare efficiency of the additional feature with that of the single LPC cepstral coefficients. The number of coefficient orders was varied in order to determine an appropriate order. Thai single, double, and triple polysyllabic words were used in this research. The 18 monophthongs from the polysyllabic words were qualitatively recognized as 9 different vowels. The highest recognition rate of the single feature obtained from 18-order LPC cepstral coefficient is 86.983 percent, while the recognition rate of the 16-order LPC cepstral coefficient accompanied by temporal derivative is 94.580 percent. The misclassification is examined and concluded that this resulted from excessively overlapped distributions of vowels in low and in back vowel group respectively.
Bibliographic reference. Maneenoi, Ekkarit / Jitapunkul, Somchai / Ahkuputra, Visarut / Thathong, Umavasee / Thampanitchawong, Boonchai / Luksaneeyanawin, Sudaporn (2000): "Thai monophthong recognition using continuous density hidden Markov model and LPC cepstral coefficients", In ICSLP-2000, vol.4, 620-623.