Abstract:
Nowaday, many researches of Thai speech recognition have been conducted. Most ofthem are emphasize on tone and vowel recognition. And most of them are word-basedsystems which have a limit number of words or syllables that they can recognize. Thisproject proposes a phoneme-based recognition system for Thai speech recognitionsystem. By the assumption that the information of tone and vowel have an effect to theleading consonant recognition, the proposed system has designed for recognizing Thaileading consonant sound from the phonemes, tone information, and vowel informationof the input syllable. The proposed system can recognize all possible Thai leadingconsonant phoneme by using hidden markov model (HMM) and neural network (NN)in the recognition process. Three experiments have been conducted on more than 12000Thai syllable sounds. In the first experiment, we use only leading consonant signal asthe input to the recognition system. In the second experiment, we use leading consonantsignal together with vowel information as the input to the recognition system. And thethird experiment use leading consonant signal together with vowel information and toneinformation as the input to the recognition system. The error rate of first experiment is35.14 percent. The error rate is reduce to 25.64 percent after add the vowel informationto the input of the system. And 18.96 percent error rate is obtained by adding the toneinformation to the input of the system. We conclude that when the information ofleading consonant signal, vowel signal, and tone signal was used to create the model,we will get the better model performance.