区分性锚模型应用于语种识别的研究  

Research on Discriminative Anchor Model Applied in Language Identification

作　　者：聂智良[1] 张兴明[1] 杨镇西[1] 张丽[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002

出　　处：《计算机工程》2012年第3期172-175,共4页Computer Engineering

基　　金：国家"863"计划基金资助项目(2008AA011002)

摘　　要：在语种识别领域,语音所含说话人的差异会影响系统识别性能。基于此,对能够实现说话人无关的锚模型进行研究。根据其在语种识别中的应用原理,结合快速区分性训练思想,提出一种语种区分性的锚模型训练算法。实验结果表明,锚模型的引入能提高系统识别性能,加入语种区分性的锚模型能进一步降低系统等错误率。In the area of language identification, the system recognition performance is affected by the speaker variability in the test utterance. In this paper, the anchor model which can realize speaker-independent is studied in the language identification. Based on its application principle, the algorithm of training language discriminative anchor model which combines the fast discriminative training algorithm is proposed. Experimental results indicate that the anchor model can improve the performance of the system and the language discriminative anchor model can further reduce the error of the system.

关 键 词：语种识别 锚模型 快速区分性训练 高斯混合模型超矢量 支持向量机 说话人特征矢量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]