融合LPCC和MFCC的支持向量机OSAHS鼾声识别  

Support Vector Machine OSAHS Snoring Recognition by Fusing LPCC and MFCC

作　　者：沈侃文 李文钧[1] 岳克强 SHEN Kanwen;LI Wenjun;YUE Keqiang(Key Laboratory of RF Circuits and Systems,Ministry of Education,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学射频电路与系统教育部重点实验室,浙江杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2020年第6期1-5,12,共6页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：浙江省重点研发计划资助项目(2019C03088)。

摘　　要：单纯打鼾者的鼾声与阻塞性睡眠呼吸暂停低通气综合征(Obstructive Sleep Apnea-Hypopnea Syndrome,OSAHS)患者的鼾声在声学特征上有较大的区分度。从鼾声的声学特征出发,提出一种基于支持向量机(Support Vector Machines,SVM)的Fisher准则融合特征鼾声分类算法。首先,通过使用线性预测倒谱系数(Linear Prediction Cepstrum Coefficient,LPCC)和梅尔倒谱系数(Mel-scale Frequency Cepstral Coefficient,MFCC)的特征提取方法来分别提取鼾声的特征,并通过计算得出LPCC和MFCC每一维特征参数的Fisher比;然后,根据Fisher比的大小进行LPCC和MFCC的特征融合;最后,用SVM进行鼾声的特征分类,识别单纯打鼾者和OSAHS患者。实验结果表明,以融合LPCC和MFCC特征参数作为特征参数时,抗噪性能好且具有较高的识别准确率,准确率达到95.8%。There is a significant distinction in acoustic features between snoring of OSAHS patients and snoring of snorer without OSAHS.In this paper,an algorithm based on support vector machine(SVM)for snore classification is proposed.We adopted linear prediction cepstrum coefficient(LPCC)and Mel-scale frequency cepstral coefficient(MFCC)to extract features of snoring separately and fuse with features using Fisher criterion.The Fisher ratio of LPCC and MFCC in each dimension of features is calculated in the proposed algorithm.Then,the feature fusion based on the Fisher ratios is achieved.Finally,features are classified by a SVM.Experimental results show that the proposed algorithm based on fusion of LPCC and MFCC has high robustness and accuracy,the corresponding accuracy rate reaches 95.8%.

关 键 词：线性预测倒谱系数 梅尔倒谱系数 融合LPCC和MFCC特征参数 支持向量机 

分 类 号：TN912.34[电子电信—通信与信息系统]