Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

 

A STUDY OF THAI SYLLABLES TO THE EFFECTIVENESS IN MACHINE VERIFICATION OF SPEAKERS

 

TITLE A STUDY OF THAI SYLLABLES TO THE EFFECTIVENESS IN MACHINE VERIFICATION OF SPEAKERS
AUTHOR APICHAT TRONGSUKSON
DEGREE MASTER OF SCIENCE PROGRAMME IN COMPUTER SCIENCE
FACULTY FACULTY OF SCIENCE
ADVISOR SUPACHAI TANGWONGSAN
CO-ADVISOR CHOMTIP PORNPANOMCHAI
SUKANYA PHONGSUPHAP
 
ABSTRACT
This research studies the relative effectiveness of Thai syllables in machine verification of speakers to find the most effective Thai words that have high distinguishing power for use in the system decision phase. The study was conducted by a series of experiments used to examine the unique properties of a speaker’s vocal tract and their spoken word effect on the speaker verification system. The speaker verification system used in this research is a text-dependent type utilizing the global speaker normalized log likelihood score, using the Mel-frequency cepstrum coefficient to create representatives of speech or feature extraction, and using the Gaussian Mixture Model to generate a speaker model and a global speaker model of each word. Finally, a threshold defined for a comparison is used to decide whether to accept or reject the speech. Based on the experimental results, it was found that sound of Thai words (Thai speech) does impact the efficiency of the speaker verification system performance. In particular Thai speech that is believed to be a factor for speaker verification is that of nasal words which are pronounced through one’s nose, like “น”, “ม”, and “ง”. In addition, there are also words with long-sound or mixed-sound vowel, such as “นอง”, “เหน่ง”, “เน้น”, etc. Thus, Thai nasal words are the most suited words to the verification task. This conclusion is based on the testing data of one hundred speakers and eleven nasal words which had the lowest error rate of False Rejection (FR) of 0.34% and 0.34% of False Acceptance (FA). Further study might better affirm the experimental consequences of this research for higher reliability, perhaps through a mathematical model to prove that boarder frequency words are the most promising for the speaker verification system.
KEYWORD SPEAKER VERIFICATION / GLOBAL SPEAKER MODEL / THAI NASAL WORD/ NORMALIZED LOG LIKELIHOOD SCORE

 

Go to Top

 

ICT Building, Mahidol University, 999 Phuttamonthon 4 Road, Salaya, Nakhonpathom 73170 Tel. +66 02 441-0909 Fax. +66 02 849-6099
Mahidol University Computing Center, The Faculty of ICT, Mahidol University , Rama 6 Road, Rajathevi, Bangkok 10400 Tel. +66 02 354-4333 Fax. +66 02 354-7333