Abstract—The presence of obstruent consonants constitutes key landmark events with cues that indicate abrupt acoustic discontinuities in the speech signal. Such discontinuities allow further analysis and recognition to be performed in knowledge-based speech recognition systems. This paper describes an acoustical investigation on Thai obstruent consonant detection using average level crossing rate (ALCR) information. Simple and easy to compute, ALCR information alone was successfully used in an automatic speech segmentation system for English. Comparable and, in some cases, slightly better performance than the spectral-domain methods using the Mel frequency cepstrum coefficients (MFCC) was reported. However, ALCR has never been applied to Thai. As a result, the objective of the study is to apply ALCR information to ascertain its usefulness in detecting significant temporal changes involving obstruent consonants in Thai continuous speech. Preliminary results suggest that ALCR and RMS energy can be combined to detect the phonetic boundary between initial obstruent consonant and preceding/following vowel or final consonant of the preceding syllable. An experiment was conducted on a small speech corpus containing 21 sentences designed to highlight the occurrences of all 21 possible leading consonants in various syllable structures. The overall detection rate is 83.5% for data from four speakers. The proposed method also reduces the insertion error due to amplitude variations within a phonetic segment. 
 

Index Terms—average level crossing rate, automatic speech segmentation, Thai obstruent detection

Cite: Siripong Potisuk, "Obstruent Consonant Landmark Detection in Thai Continuous Speech," International Journal of Signal Processing Systems, Vol. 4, No. 3, pp. 214-219, June 2016. doi: 10.18178/ijsps.4.3.214-219