Research on text analysis for Tibetan statistical parametric speech synthesis

Abstract

Text analysis is the front-end of a TTS system, which has a great influence on the naturalness of the back-end speech synthesis. Statistical parametric speech synthesis is being commonly applied into speech synthesis now, and gradually becoming an important method of the current speech synthesis, however, the research of front-end text analysis is often overlooked in the process of current Tibetan speech synthesis, so the research of Tibetan text analysis is still staying in an initial stage. The research of this paper is faced on the Tibetan statistical parametric speech synthesis system, the mainly work of us is to have a text analysis for the input Tibetan text, and the aim is to acquire the mono-phone labeling information and context-dependent labeling information, which is needed by the back-end speech synthesis. Finally, we have a speech synthesis to the Tibetan text with the statistical parametric information that we have obtained through the process of text analysis, then the quality of synthetic speech is evaluated by MOS evaluation, we choose randomly 50 sentences to apply to the process of MOS evaluation, through the experiment, the score of MOS evaluation can reach 4.0 points, so it indicated that the naturalness and intelligibility of synthetic speech is good, and the method of this paper is effective.