可训练语音合成在侗语合成中的应用

The Application of Hmm-based Trainable Speech Synthesis to Dong language

侗语自1958年创造了拉丁字母的文字方案后,形成了一种拼音化的文字形式。本文通过HMM的语音合成方法,结合侗语自身的语言特点,确定了语音语料的选取、录音方式、音素列表、标注以及上下文问题集和属性集的设计,实现了对于侗语语音的合成。通过在HTS上搭建语音合成平台进行测试,测试结果表明合成语音的可懂度较高。

The Latin alphabetic orthography of Dong language was invented in 1958, and from then on Dong language has its own written language in form of Pinyin. Dong language has its special language features and it is comparatively easy to determine language material selecting, recording, phonemic listing, labeling and to design contextual problem sets and attribute sets. This paper discusses the implementation of synthesizing Dong language sounds via HMM speech synthesis. We conduct a test through HTS speech synthesis and the results show that the synthesized speech has high intelligibility.