马来语语音合成系统的设计与实现

The Design and Implementation of a Malay Speech Synthesis System

摘要: 马来语广泛使用于马来西亚、新加坡等东南亚国家，目前使用人数约有2亿多人。本文研究马来语语音合成系统的前端文本分析与处理方法、以及基于HMM的后端语音合成方法。在前端文本分析与处理环节，研究并实现了马来语语料的收集与挑选、文本归一化、以及音节自动划分；在后端语音合成环节，研究并实现了马来语音子列表确定、文本标注、上下文属性和问题集设计、HMM声学模型训练、以及语音波形产生。实验结果表明：本文提出并实现的前端文本分析与处理方法可满足后端语音合成的要求，采用本文构建的后端语音合成系统可合成出完整的马来语语句。

Abstract: Malay is widely used in Malaysia, Singapore and other Southeast Asian countries. Currently, there are about 200 million people using Malay. This paper studies the front-end text analysis method of Malay speech synthesis system, and the back-end speech synthesis method based on HMM. In front-end text analysis and processing, the collection and selection of Malay language data, text normalization, and automatic syllable division were researched and implemented; In the back-end speech synthesis section, the Malay Phonetic list determination, text annotation, context attributes and problem set design, HMM acoustic model training, and speech waveform generation were studied and implemented. Experimental results show that the front-end text analysis and processing method proposed and implemented in this paper can fulfil the requirements of back-end speech synthesis. The back-end speech synthesis system constructed in this paper can synthesize a complete Malay sentence.