Development of HMM-based Malay Text-to-Speech System

This paper presents the development of a hidden Markov model (HMM)-based Malay text-to-speech (TTS) system. To our knowledge, this is the first report on the development of the HMM-based speech synthesis system for the Malay language. In this paper, We first discuss the Malay speech characteristics, specifically, on Malay phonological system and syllable structure. In the Malay phonological system, 37 phonemes are adopted as the phonemic representations. Then, we describe a HMMbased TTS framework and language specific knowledge such as phonological, linguistic information, and utterance structure, which is used in context dependent continuous HMM and treebased clustering. After that, we report the development of Malay TTS corpora. Finally, a male and a female HMM-based Malay TTS systems are developed and evaluated. We further conduct listening test based on the Mean Opinion Score (MOS), and the results show that the developed HMM-based Malay TTS system can generate speech with acceptable quality in terms of naturalness and intelligibility.