Comparison of Syllable and Phoneme Modelling of Agglutinative Tamil Isolated Words in Speech Recognition
Ibralebbe Mohamed Kalith *
Department of Mathematical Science, Faculty of Applied Sciences, South Eastern University of Sri Lanka, Sri Lanka.
David Asirvatham
School of Computing and IT, Taylor's University, Subang Jaya, Selangor, Malaysia.
Ali Khatibi
Graduate School of Management, Management and Science University, Malaysia.
Samantha Thelijjagoda
Department of Information Systems Engineering, Sri Lanka Institute of Information Technology, Sri Lanka.
*Author to whom correspondence should be addressed.
Abstract
Aim: In this paper, the emphasis was on improving the automatic speech recognition of Tamil speech by applying syllable and phoneme as a sub-word unit. Agglutinative complex words in Tamil are described by showing their element in the building of the sub-word such as the syllable and the phonemes. This present study used the Hidden Markov Model (HMM) based speech recognition system that was created using CMU Sphinx speech recognition toolkit. An effective consonant-vowel six-segment (CVS-6) algorithm was designed to syllabification of the Tamil isolated words and experimentally investigated its speech recognition accuracy. The database used in this study was designed using a maximum of 160 isolated words, representing 430 syllables and 216 unique syllables.
Results: Through the experiment, the syllable-based model achieved a mean recognition rate of 93.84 (standard deviation, 5.02) compared to 91.37 (standard deviation, 6.26) achieved by a phoneme-based model.
Conclusion: It was concluded from this research that the syllable-based model using the CVS-6 algorithm is a good choice and can be used in the development of sub-word modelling of isolated words, which is an effective sub-word modelling of medium and large vocabulary ASR Tamil language.
Keywords: Hidden Markov model, phoneme-based, syllable-based, ASR, Tamil language, Consonants Vowel Segmentation-6 Algorithm