Final Consonant Segmentation for Thai syllable by Using Vowel Characteristics and Wavelet Packet Transform

Abstract

Since Thai final consonant is unique comparing with other languages and plays key role in recognizing the Thai syllables, segmentation of the final consonant phoneme from the vowel is needed and capable of decreasing the amount of recognition patterns and also improving the recognition accuracy. This paper presents a technique to separate the final consonant phoneme from Thai syllable by exploiting the vowel characteristics and Wavelet packet transform. In this method, ending of the vowel phoneme (starting of the final consonant) is considered by vowel characteristic, which has the highest energy in the syllable. The frequency range having this qualification is selected as vowels. It is then employed to determine the filter for vowel signal. The Wavelet packet transform that is appropriate for discriminating vowel (high frequency and long period) from final consonant phoneme (low frequency and short period) is used as the filter. And the ending of vowels frequency signal component is considered to be the segmentation point of the final consonant. The experiments have been performed by 4,350 samples of syllable recorded from 15 males and 15 females. The experimental results gained the 92.89 % accuracy.