A stochastic knowledge-based Thai text-to-speech system

We describe the development of our Thai text-to-speech (TTS) system. Thai TTS system transforms Thai texts to the sequence of appropriate sounds for Thai speech. Thai complexity requires approximately four hundred rules including main and specific rules to derive most pronunciations for our rule-based approach grounded on Thai syllable structure analysis. An exception dictionary that covers anomalous pronunciations and a rule inference engine that determines sentence structures are included to improve the quality of Thai TTS. Speech generation by concatenative synthesis is a sequential step that transforms sound symbols to synthetic speech. We have tested our system with magazine and internet articles, together with articles from the experiments of other researchers and we report the results of this informal evaluation. Most syllables from Thai written strings can be converted to phonetic symbols. With a compact Thai unit inventory, the concatenative synthesis system can synthesize synthetic speech covering many Thai syllables.