英文文语转换系统中基于形态规则和机器学习的重音标注算法  

English accent assignment based on morphological rules and machine learning

作　　者：王永生[1] 李梅[2] 

机构地区：[1]同济大学留德预备部,上海200092 [2]同济大学外国语学院,上海200092

出　　处：《计算机应用》2008年第1期88-91,共4页journal of Computer Applications

摘　　要：在英文TTS系统中,未登录词的重音标注是除字音转换外另一个十分重要的环节。由于主重音的重要性要远远大于次重音,且主重音的情况要比次重音的情况简单一些,因而将主重音的标注与次重音的标注分开进行。主重音的标注采用形态规则和机器学习相结合的标注算法;而次重音的标注完全通过机器学习算法来进行。经过10轮交叉验证,主重音的平均标注正确率为94.4%,次重音的平均标注正确率为86.9%,总的标注正确率为83.6%。Accent assignment of out-of-vocabulary is a very important component besides letter-to-phoneme Conversion in English Text-To-Speech （TTS）. Considering that primary accent is much more important and simpler than secondary accent, their assignment was conducted separately. A hybrid algorithm of morphological rules and machine learning was presented to tackle the assignment of primary accent. And a machine learning algorithm was proposed to handle the assignment of secondary accent. After 10-fold cross validation, the average accuracy of primary and secondary accent assignment reached 94.4% and 86.9% respectively, and the total accuracy was 83.6%.

关 键 词：文语转换 未登录词 重音标注 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]