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。

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

Abstract: 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.

Key words: language identification, anchor model, fast discriminative training, Gaussian Mixture Model(GMM) super vector, Support Vector Machine (SVM), speaker characterization vector

中图分类号: 

• TP391