Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

THAI SPEECH RECOGNITION USING SYNTACTIC AND SEMANTIC ANALYSIS

 

TITLE THAI SPEECH RECOGNITION USING SYNTACTIC AND SEMANTIC ANALYSIS
AUTHOR RONG PHOOPHUANGPAIROJ
DEGREE DOCTOR OF PHILOSOPHY PROGRAMME IN COMPUTER SCIENCE(INTERNATIONAL PROGRAMME)
FACULTY FACULTY OF INFORMATION AND COMMUNICATION TECHNOLOGY
ADVISOR SUKANYA PHONGSUPHAP
CO-ADVISOR SUPACHAI TANGWONGSAN
CHONTIP PORNPANOMCHAI
 

ABSTRACT

Thai speech recognition has been actively studied for more than a decade. Many methods used in speech recognition of foreign languages, such as English and Chinese, are applied to develop Thai speech recognition systems. Although there are several existing methods used to recognize a Thai speech utterance, they cannot achieve high multiple-keyword recognition results and the results are not accurate enough to understand the utterance. Furthermore, only recognizing speech may not be enough for speech applications that require semantic information. Hence, a method using syntactic and semantic analysis through grammatical phrase-based patterns, and utterance concept summarization is proposed to improve speech recognition and understanding performance.The proposed method is comprised of two main parts: speech recognition and concept summarization. In the first part, to reduce acoustic variation in Thai speech recognition, gender identification from the Thai speech signal is proposed to allow a speech recognizer to select acoustic models that fit a speaker. The speech recognizer uses the acoustic models represented by Continuous Density Hidden Markov Models (CDHMMs) and predefined grammatical phrase-based patterns to convert speech signals to a word sequence consisting of a key-phrase. In the second part, the concept expressed by the spoken utterance is summarized using Thai syntactic and semantic analysis to obtain a key-phrase and an utterance topic. In the grammatical part, the key-phrase can be extracted by using the grammatical phrase-based patterns while in the ungrammatical part, the utterance topic can be extracted by using semantic analysis through machine-learning topic identification. According to the experimental results, gender identification rate of around 98.00% can be achieved using a Thai spoken syllable and a neural network. The proposed method can achieve a higher accuracy of 98.03% in recognizing key-phrases, compared with the use of N-gram language models, which can obtain an accuracy of 92.35%. Topic identification rates of 97.88% are attained using support vector machines (SVM). An utterance concept summarization rate of 95.91% can be achieved using Thai syntactic and semantic analysis. The experimental results indicate that the proposed method is an efficient method for Thai speech recognition and understanding.


KEYWORD THAI SPEECH RECOGNITION / THAI SYNTACTIC AND SEMANTIC ANALYSIS / GENDER IDENTIFICATION / SPOKEN LANGUAGE UNDERSTANDING / CONCEPT SUMMARIZATION

 

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