计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (2): 161-164.

• 模式识别与人工智能 • 上一篇    下一篇

基于频段互相关系数的咳嗽识别新方法

朱春媚1,2,黎  萍1   

  1. 1.电子科技大学 中山学院 自动化系,广东 中山 528403
    2.华南理工大学 自动化科学与工程学院,广州 510641
  • 出版日期:2016-01-15 发布日期:2016-01-28

Novel method for cough recognition based on band cross correlation coefficient

ZHU Chunmei1,2, LI Ping1   

  1. 1.Department of Automation, Zhongshan Institute, University of Electronic Science and Technology, Zhongshan, Guangdong 528403, China
    2.College?of?Automation?Science?and?Engineering, South?China?University?of?Technology, Guangzhou 510641, China
  • Online:2016-01-15 Published:2016-01-28

摘要: 在咳嗽识别中,语音是影响识别准确率的主要因素。分析咳嗽与语音相邻帧频谱的相似性特征,发现咳嗽相邻帧的频段互相关系数明显小于语音,因此频段互相关系数可以作为区分咳嗽与语音的动态特征。在相同实验条件下,以MFCC为静态特征,比较了以频段互相关系数和一阶MFCC作为动态特征参数的咳嗽识别性能。多组录音的咳嗽识别实验结果表明:采用频段互相关系数作为动态特征参数咳嗽识别的平均准确率为90.27%,其识别能力优于一阶MFCC。

关键词: 计算机辅助诊断, 咳嗽识别, 互相关系数, 动态特征

Abstract: Speech is the main factor affecting the results in cough recognition. By analyzing the spectral similarity of adjacent frames, it is found that the band cross correlation coefficient of cough is obviously smaller than that of speech, which indicates that band cross correlation coefficient can be used as dynamic feature to discriminate cough and speech. Using MFCC as static feature, cough recognition performances respectively using band cross correlation coefficient and first-order MFCC as dynamic feature are compared in the same experiment conditions. Recognition results of various groups of recordings show that, the average recognition rate reaches 90.27% by using band cross correlation coefficient as dynamic feature, and outperforms the counterparts of first-order MFCC.

Key words: computer-aided diagnosis, cough recognition, cross correlation coefficient, dynamic feature