融合MPEG-7和声门特征的病理嗓音识别方法研究  

Research on the Design of Pathological Voice Recognition System Integrating MPEG-7 and Glottal Features

作　　者：朱欣程 伍远博 赵登煌 张晓俊 陶智 ZHU Xincheng;WU Yuanbo;ZHAO Denghuang;ZHANG Xiaojun;TAO Zhi(School of Optoelectronic Science and Engineering,Soochow University,Suzhou Jiangsu 215006,China)

机构地区：[1]苏州大学光电科学与工程学院,江苏苏州215006

出　　处：《电子器件》2022年第3期587-592,共6页Chinese Journal of Electron Devices

基　　金：国家自然科学基金项目(61271359);教育部光电教指分委教育教学研究项目(gdyljs52);苏州大学高等教育教改研究课题项目(5731503920)。

摘　　要：提出了一种融合多媒体内容描述接口(MPEG-7)和声门特征的病理嗓音识别方法,以更细致地表征病理嗓音与健康嗓音之间的差异度,提高病理嗓音识别率。首先将声门特征与MPEG-7特征进行融合,随后通过贝叶斯网络、BP神经网络、逻辑回归、支持向量机、局部加权线性回归五种机器学习方法进行识别实验。采用DSP芯片TMS320VC5509A为核心实现该方案。采用MEEI数据库中的正常嗓音和病理嗓音进行十折交叉验证实验,实验结果表明,MFCC、LPCC和MPEG-7融合声门特征的平均识别率分别比融合前分别提高了2.87%、1.78%和0.6%。其中,融合MPEG-7和声门特征在支持向量机方法下性能最优,能达到100%的识别率。A pathological voice recognition method that integrates MPEG-7 and glottal features is proposed,in order to characterize the difference between pathological voices and healthy voices and improve the recognition rate of pathological voices.Firstly,the glottal features are fused with MPEG-7 features,and pathological voice recognition experiments are carried out by using five machine learning methods,i.e.,Bayesian network,BP neural network,logistic regression,support vector machine,and local weighted linear regression.The pathological voice recognition system is realized with the DSP chip TMS320 VC5509A as the core component.Ten fold cross validation experiments are performed using normal voice and pathological voice on the MEEI database.The experimental results show that the average recognition rates of MFCC,LPCC and MPEG-7 after the fusion of glottal features are 2.87%,1.78%and 0.6%higher than those before the fusion of glottal features respectively.The recognition rate of MPEG-7 after integrating glottal features can reach 100%by using the support vector machine method.

关 键 词：特征提取 声门逆滤波 融合特征 MPEG-7 病理嗓音 

分 类 号：TN912.34[电子电信—通信与信息系统]