基于C_nV结构的自动语言辨识研究  

Automatic Language Identification Using Structure C_nV

作　　者：戴冠男[1] 王炳锡[1] 屈丹[1] 

机构地区：[1]解放军信息工程大学,郑州450002

出　　处：《计算机工程与应用》2006年第23期173-175,共3页Computer Engineering and Applications

基　　金：国家自然科学基金资助项目(编号:60372038)

摘　　要：与说话人识别、连续语音识别相比,自动语言辨识是一个相对较新的研究,而且是一项较难的课题。与音素配位学相比较韵律是语言辨识的更有希望的一个语言辨识特征。论文介绍了一种基于伪音节结构CnV的自动语言辨识方法,该系统提取了辅音、元音构成的伪音节结构的MFCC和!MFCC特征参数,使用了与语言无关的GMM算法模型化该特征。经过对OGI-TS数据库中的英语、法语、汉语测试表明,元音、辅音特征信息在语言辨识中起到一定作用,伪音节结构模型也是语言辨识的有效模型之一。Compared with other speech technologies in speech processing,automatic language identification is a relatively new yet difficult problem.Besides phonotactics,rhythm is actually one of more promising features to be considered for language identification.In this paper,a language identification algorithm is described that is based on CnV structure.The system extracts MFCC and △MFCC of Pseudo-Syllable.And we study these features using Gaussian Mixture Model.Some experiments are conducted using English,French and Mandarin from OGI-TS telephone speech corpus.Then experiments results show that vowel and consonant are efficient features in language identification.Pseudo-Syllable model is one of efficient model to language identification.

关 键 词：自动语言辩识(LID) 伪音节(PS) 伪音节结构模型(PSM) 高斯混合模型(GMM) 

分 类 号：TN912.34[电子电信—通信与信息系统]