计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (21): 206-208.DOI: 10.3778/j.issn.1002-8331.2008.21.056

• 机器学习 • 上一篇    下一篇

基于多项式拟合的中性-情感模型转换算法

单振宇,杨莹春   

  1. 浙江大学 计算机科学与技术学院,杭州 310027
  • 收稿日期:2008-04-30 修回日期:2008-05-30 出版日期:2008-07-21 发布日期:2008-07-21
  • 通讯作者: 单振宇

Neutral-emotion model transformation algorithm based on polynomial function fitting

SHAN Zhen-yu,YANG Ying-chun   

  1. Department of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China
  • Received:2008-04-30 Revised:2008-05-30 Online:2008-07-21 Published:2008-07-21
  • Contact: SHAN Zhen-yu

摘要: 情绪变化问题是说话人识别技术面临的一个难题。为了解决该问题,提出了基于多项式方程拟合的中性-情感模型转换算法。该算法建立了中性模型和情感模型之间的函数关系,只需要说话人的中性语音就能训练其各种情感类型的说话人模型。在普通话情感语音库上的实验表明,采用该方法后识别算法的等错误率由16.06%降低到10.31%,提高了系统性能。

关键词: 说话人识别, 高斯混合模型, 情感语音

Abstract: One of the largest challenges in speaker recognition is dealing with speaker-emotion variability problem.A neutral-emotion model transformation algorithm is presented to overcome this limitation,which builds a relationship between emotion and neutral models.In this method,only neutral speech is needed in training emotion models.The experiments on MASC show that the EER is reduced to 10.31% from the 16.06%,and the recognition performance can be improved by this algorithm.

Key words: speaker recognition, gaussian mixture model, emotion speech