计算机与现代化 ›› 2013, Vol. 1 ›› Issue (4): 5-9.doi: 10.3969/j.issn.1006-2475.2013.04.002

• 人工智能 • 上一篇    下一篇

一种基于SVDD模型的说话人确认方法研究

陈觉之1,张贵荣2,周宇欢3   

  1. 1.海军指挥学院信息系,江苏 南京 211800;2.中国人民解放军92601部队计量站,广东 湛江 524009;3.解放军理工大学指挥信息系统学院,江苏 南京 210007
  • 收稿日期:2013-03-12 修回日期:1900-01-01 出版日期:2013-04-17 发布日期:2013-04-17

Research on Method of Speaker Verification Based on Support Vector Data Description Model

CHEN Jue-zhi1, ZHANG Gui-rong2, ZHOU Yu-huan3   

  1. 1. Department of Information, Naval Command Academy, Nanjing 211800, China;2. Metering Station, Troop 92601 of PLA, Zhanjiang 524009, China;3. Institute of Command Information System, PLA University of Science and Technology, Nanjing 210007, China
  • Received:2013-03-12 Revised:1900-01-01 Online:2013-04-17 Published:2013-04-17

摘要: 为改善说话人确认的性能,提出一种基于支持向量数据描述(SVDD)模型的说话人确认方法,通过改变SVDD硬判决方式,采用以样本接受率为依据的软判决方式,把似然得分规整到[0,1]之间,简化门限阈值的设定。仿真实验结果表明,与通常基于高斯混合模型(GMM)的说话人确认算法相比,该方法的说话人确认性能有较大提高。

关键词: 说话人识别, 说话人确认, 向量数据描述, 高斯混合模型

Abstract: With the purpose of improving the performance of speaker verification, a novel speaker verification method based on support vector data description (SVDD) model is proposed. The traditional hard decision method of SVDD is changed to a novel soft decision method based on the sample acceptance rate, therefore the confidence scores are normalized to the value [0,1] so as to simplify the threshold value setting. Simulation experiments show that the performance of speaker verification based on this novel method is remarkably better than that based on Gaussian mixture model (GMM) customarily.

Key words: speaker recognition, speaker verification, support vector data description (SVDD), Gaussian mixture model (GMM)

中图分类号: