北京大学学报(自然科学版)

多特征组合多分类器的方法用于“文本无关”的说话人辨认

王岚, 陈珂, 迟惠生   

  1. 北京大学信息科学中心,北京,100871
  • 收稿日期:1997-12-05 出版日期:1998-04-20 发布日期:1998-04-20

Text-independent Speaker Identification in Combining Multiple Classifiers with Different Features

WANG Lan, CHEN Ke, CHI Huisheng   

  1. Center for Information Science, Peking University, Beijing 100871, China
  • Received:1997-12-05 Online:1998-04-20 Published:1998-04-20

摘要: 组合多分类器可以看作是一种用于获得较高识别效果的混合系统。重点探索了以不同特征作为输入的组合多分类器方法。实验结果表明:利用多特征组合多分类器的方法可以提高“文本无关”说话人辨认系统的识别率和可靠性。

关键词: “文本无关”说话人辨认, 多特征组合多分类器, 证据推理, 线性组合, 胜者优先

Abstract: Combining Multiple Classifiers can be viewed as a novel hybrid system to achieve high recognition accuracy for Text-Independent Speaker Identification. This article has summarized current methods of combining multiple classifiers, and investigated on embodying different features as input vectors. The experimental results have shown that Combining Multiple Classifiers with different features can result in satisfactory and significant improvement in recognition performance.

Key words: text-independent speaker identification, combining multiple classifiers, evidential reasoning, linear opinion pools, winner-take-all

中图分类号: