英语字素音素的可视匹配与特征分析  

Exploring the Phonetic Features in English Based on Visual Analysis

作　　者：李昕[1] 李珊[1] 龚文涛 陈晓莹 李华昱[1] LI Xin;LI Shan;GONG Wentao;CHEN Xiaoying;LI Huayu(College of Computer Science and Technology, China University of Petroleum, Qingdao 266580, China;Information Construction Department, China University of Petroleum, Qingdao 266580, China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580 [2]中国石油大学(华东)信息化建设处,山东青岛266580

出　　处：《郑州大学学报（理学版）》2021年第4期44-52,共9页Journal of Zhengzhou University:Natural Science Edition

基　　金：中石油重大科技项目(ZD2019-183-004);中央高校基本科研业务费专项资金(20CX05019A)。

摘　　要：语音识别和语音合成是近年来的热门研究,相关研究主要是在字素音素匹配的基础上进行统计分析,探寻拼写与发音之间的内在规律,达成形音转换。语言自身的混沌性使字素音素匹配非常复杂,提出一种交互式可视化工具,通过从粗到细的动态分类方式完成字素音素增量匹配。在此基础上,通过全局关联性分析揭示英语发音的整体规律,并及时检验和处理匹配错误和特殊发音等异常情况。采用5个著名语料库中的35182个单词进行实验,匹配过程快速精准,并且发现的规律具有高度的可解释性,为相关人员进行语言学习和语音研究提供了坚实的基础。Speech recognition and speech synthesis were hot topics of study.Related study was mainly based on the statistical analysis of the grapheme and phoneme matching,exploring the internal patterns between spelling and pronunciation,and achieving the conversion of form and sound.Due to the chaos of the language itself,the process of grapheme-phoneme matching was very complex.An interactive visualization tool was proposed to complete the incremental matching of graphemes and phonemes through dynamic classification from coarse to fine.On this basis,the overall rules of English pronunciation were revealed through global relevance analysis,and abnormal situations such as mismatches and special pronunciations were checked and dealt with in time.35182 words in five famous corpora were used to conduct experiments.The matching process was fast and accurate,and the discovered rules were highly interpretable,providing a solid foundation for fourther exploration in language learning and phonetic research.

关 键 词：增量匹配 可视分析 动态分类 异常检测 关联分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]