• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2010, Vol. 32 ›› Issue (4): 136-138.

• 论文 • 上一篇    下一篇

基于复合能量和自适应阈值的语音端点检测

姜占才,孙燕,王得芳   

  1. (1.青海师范大学物理系,青海 西宁 810008;2.青海民族学院计算机科学与技术系,青海 西宁 810007)
  • 收稿日期:2008-11-15 修回日期:2009-02-09 出版日期:2010-03-28 发布日期:2010-03-28
  • 通讯作者: 姜占才 E-mail:243740714@qq.com
  • 作者简介:姜占才(1958),男,青海西宁人,教授,研究方向为语音与图像处理、数字信号处理等。

JIANG Zhancai,SUN Yan,WANG Defang   

  1. (1.Department of Physics,Qinghai Normal University,Xining 810008; 2.Department of Computer Science and Technology,Qinghai University for Nationalities,Xining 810007,China)
  • Received:2008-11-15 Revised:2009-02-09 Online:2010-03-28 Published:2010-03-28
  • Contact: JIANG Zhancai E-mail:243740714@qq.com

摘要: 针对低信噪比的加性噪声,本文提出了一种新的语音断点检测算法。采用Teager能量对短时能量的加权归一化——复合能量作特征参数,用噪声的归一化复合能量的均值作初始阈值,引入阈值调整系数,使阈值对噪声能量具有良好的跟踪性能。 实验结果表明,对计算机模拟噪声,在低信噪比(10dB~3dB)下仍能保证较高的检测准确率。

关键词: Teager能量, 复合能量, 语音端点检测, 自适应阈值, 信噪比

Abstract: This paper presents a new speech endpoint detection algorithm to additive noise of low signal to noise ratio, the algorithm uses the Teager energy for normalized weighted short energy named Composite Energy as the characteristic parameter, and the normalized composite energy of noise as the initial threshold value, and imports an adjustment coefficient of threshold to make the threshold have good tracking performance to the energy of noise. To do computer simulation experiments on the algorithm by using MATLAB , and make a comparision between the results of experiments and the results of short energy, the results show that it can still ensure a high accuracy rate in the simulation of noises between the negative 10 dB and the negative 3 dB.

Key words: Teager energy;composite energy;speech endpoint detection;adaptive threshold;signaltonoise ratio