傣语语音合成中的文本归一化方法

An Approach to Normalization of Dai Text for Speech Synthesis

本文以开发傣语语音合成系统为目的,重点研究傣语文本中的数字归一化和特殊字符归一化问题。数字和特殊字符都属于傣语文本中的非标准词,文本归一化的主要目的是用标准词表示非标准词的发音。归一化处理过程包括:非标准词识别、歧义判断、消歧处理和非标准词转换为标准词4个步骤。本文采用基于规则和上下文关键词相结合的方法识别非标准词,利用正则表达式判断其歧义类型,根据转换规则对非标准词进行消歧并确定其正确的傣文读音。实验结果表明,本文提出的文本归一化方法的正确率达到了94.6%,可以完全满足傣语文语转换系统前端文本分析的需求,并具有良好的自然语言处理应用价值。

With the purpose of developing a Dai speech synthesis system, this paper focuses on the study of Dai numbers and special characters normalization. Both numbers and special characters are the non-standard words in Dai text. The main purpose of the text normalization is to represent the pronunciation of non-standard words with standard words. The normalization process includes non-standard words recognition, ambiguity judgment, disambiguation and non-standard transla-tion. Firstly, the non-standard words are recognized and the ambiguous types of these non-stan- dard words are determined using a method based on rule-based and context-keyword, in this paper. Then, the types of ambiguity are judged on regular expression. Lastly, the correct pronunciation of no-standard words is determined according to the transformation rules. Experimental results show that the correct rate of this normalization is more than 94.6%. This purposed method can fully satisfy the front-end text analysis in Dai text to speech conversion system, and has a good natural language processing application value.