面向语音合成的印尼语文本分析与处理

Indonesian Text Analysis and Processing for Speech Synthesis

摘要: 本文以开发印尼语语音合成系统为目的，研究印尼语文本分析与处理方法，主要研究了文本归一化和音节划分方法。采用正则表达及关键字相结合的方法，对文本中数字及特殊字符进行归一化处理；采用基于音节列表及特殊规则相结合的方案实现音节划分。从发音语料库中挑选出500个包含特殊字符的句子进行归一化测试，按特殊字符词数统计的正确率达96.0%。选取词典中的1000个单词进行集内测试，其音节化结果和人工划分结果相比，正确率为98.2%；从文本语料库中任意选取480个句子共计5850个单词进行集外测试，其正确率为97.1%。实验结果表明，上述方法为印尼语音合成系统的开发奠定了良好的基础。

Abstract: This paper focused on the development of Indonesian speech synthesis system, and it studied Indonesian text analysis and processing methods. It mainly studied text normalization and syllable division methods. By using a combination of regular expressions and keywords, the numbers and special symbol in the text are normalized. Furthermore, a combination of syllable lists and special rules are used to achieve syllable segmentation. From the pronunciation corpora, 500 sentences containing special characters were selected for normalization testing. The correct rate of the number of words according to special characters was 96.0%. The inset testing selected 1000 words in the dictionary, and the syllable results were compared with the results of artificial division with the correct rate of 98.2%. From the text corpora, 480 sentences were randomly selected for a total of 5850 words for out-of-set testing, and the correct rate was 97.1%. The above experiments laid a good foundation for the development of the Indonesian speech synthesis system.