基于子带能量变换改进MFCC的咳嗽识别  

Method for cough recognition based on MFCC improved by sub-band energy transformation

作　　者：朱春媚[1,2] 黎萍[1] 

机构地区：[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510640

出　　处：《计算机工程与应用》2016年第11期148-151,157,共5页Computer Engineering and Applications

基　　金：广东省高等学校学科与专业建设专项资金(No.2013LYM0103);中山市科技计划项目(No.2014A2FC383)

摘　　要：在咳嗽识别中,频谱能量较高是咳嗽区别于大部分非咳嗽声音的主要特征,在分析得到咳嗽各子带频谱能量分布系数的基础上,提出了一种基于子带能量变换改进MFCC的咳嗽识别新方法。该方法在MFCC提取过程中引入子带能量变换,对高能量子带进行增强以降低非咳嗽信号的误识别率和提高轻微咳嗽的识别率,舍弃低能量子带以提高系统的抗干扰能力。在相同实验条件下,比较了采用改进MFCC和传统MFCC的咳嗽识别性能。实验结果表明:改进MFCC有效改善了咳嗽识别的各项评价指标和增强了算法的鲁棒性,平均识别率从89.29%提高到了94.43%。High spectral energy is the main characteristic which distinguishes cough from most of non-cough sounds. By analyzing the sub-band energy distributional characteristic of cough, an improved MFCC extraction method based on sub-band energy transformation is proposed for cough recognition. Sub-band energy transformation is introduced in the MFCC extraction process. This method enhances sub-bands with high-energy to reduce the misrecognition rate of non-cough similar to cough and increase the recognition rate of mild cough, and neglects sub-bands with low-energy to improve anti-interference ability of the system. Recognition performances using improved MFCC and traditional MFCC under the same experiment condition are compared. The experimental results show that, by using improved MFCC, each performance indicator of cough recognition is improved and the robustness of the algorithm is enhanced. The average recognition rate is raised from 89.29% to 94.43%.

关 键 词：辅助诊断 咳嗽识别 子带能量变换 特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]