基于词图的音素识别及在语种识别中的应用  

Lattice-based PPRLM and Its Application in Language Identification

作　　者：王士进[1] 郑榕[1] 徐波[1] 

机构地区：[1]中国科学院自动化研究所,北京100080

出　　处：《中文信息学报》2008年第2期124-128,共5页Journal of Chinese Information Processing

基　　金：国家973规划资助项目(2004CB318105);国家863计划资助项目(2006AA010103)

摘　　要：本文介绍了一种基于词图的并行音素识别方法的自动语种识别系统,基于词图的并行音素识别方法是并行音素识别方法的一个扩展,它用识别产生的词图来描述声学候选结果空间,比并行音素识别方法中用最佳路径音子序列包含更丰富的信息。通过真实环境广播语音测试表明,该方法比并行音素识别方法识别性能提升了约6%,在每个语种约4小时的训练数据下,跟其他的几种语种识别方法也有可比的性能。The paper presents an automatic language identification （LID） system which is based on the lattice-based PPRLM method. As an extension of the original PPRLM, the lattice-based PPRLM method uses lattice to generate the acoustic hypothesis space, which contains more information than that of 1-best phoneme sequence in the original PPRLM. Evaluations from the broadcasting speech in real environments show that the lattice-based PPRLM improves the accuracy rate by 6%. Results are also comparable with other approaches within different languages, while four-hour training set is given for each language.

关 键 词：计算机应用 中文信息处理 语种识别 基于词图的并行音素识别方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]