基于CV音节的高效语种识别方法  

An Efficient Language Identification Method Based on CV-syllables

作　　者：胡月[1] 郭英[1] 张东伟[1] 侯文林[1] 李红光[1] 

机构地区：[1]空军工程大学信息与导航学院,陕西西安710077

出　　处：《微电子学与计算机》2015年第10期26-30,共5页Microelectronics & Computer

基　　金：国家自然科学基金(61172148);通信网信息传输与分发技术重点实验室基金(ITD-U12003/K1260009)

摘　　要：为了快速有效地识别语种,提出了基于元音起始点(Vowel Onset Points,VOP)检测的CV音节划分法,并据此研究了一种新的基于CV音节的语种识别方法.首先,给出一种能有效避免语音结束点错判的双边双门限端点检测法提取语音段;然后采用线性预测残差(Linear Prediction Residue Error,LPRE)检测语音段中的VOP,从而划分出CV音节;最后,提取各CV音节的特征矢量并利用支持向量机(Support Vector Machine,SVM)模型实现语种识别.通过对英语、汉语普通话及粤语三种语言的识别实验表明,所提VOP检测法可确保CV音节的精确划分;新方法识别率高,且识别结果对CV音节长度不敏感,模型训练时间短,可实现语种的高效识别.In order to identify language rapidly and effectively, a method for CV-syllable extraction based on the Vowel Onset Points （VOP） detection is proposed, on this basis, a new method for language identification based on CV-syllables is researched. First, a double-threshold endpoints detection from both sides is given to get speech segment, which could avoid great risk of wrong decision. Second, VOP are detected by Linear Prediction Residue Error （LPRE） and the CV-syllables from the speech segment are obtained exactly. At last, feature vectors for each CV-syllable are extracted. The Support Vector Machine （SVM） is adopted to realize language identification. The simulation experiment for English, Mandarin and Cantonese shows that VOP detection makes sure the precision of each CV-syllable extraetion The new method has the high correct response rate. The change of CV-syllable length almost has no effect on identification results. The training time for model is so short that language identification could complete efficiently.

关 键 词：语种识别 CV音节 元音起始点检测 线性预测残差 支持向量机 

分 类 号：TN912.34[电子电信—通信与信息系统]