摘要
提出一种新的基于语音分析的肝豆状核变性(WD)言语障碍诊断算法.实验从提取的所有特征中选择传统特征和非传统特征,分别采用人工神经网络、支持向量机等算法对健康人和运动性构音障碍患者进行分类.研究数据由31个样本(包括23位WD患者和8位健康受试者)的一系列生物医学语音信号组成.要求测试对象对汉字“包”持续发音3秒.从信号中提取22个线性和非线性特征,其中对语音信号有主要影响因素的有15个特征,它们建立在基频F0、基频微扰、振幅微扰和谐噪比的基础上.从这些影响因素中选择最优特征,使用不同的分类器进行比较.实验结果表明,采用不同传递函数的神经网络分类器进行分类,各性能参数各有优劣;支持向量分类器是最佳分类器,准确率达95.92%.
A new diagnostic algorithm for speech disorders in hepatolenticular degeneration(WD)based on speech analysis was presented.Traditional and non⁃traditional features were selected from all the extracted features,and then the algorithm of artificial neural network and support vector machine were used to classify healthy people and patients with motor dysarthria.The data set was collected from a series of biomedical speech signals of 31 individuals,included 23 WD patients and 8 healthy subjects.The test subjects were asked to pronounce the Chinese character“bao”for three seconds.22 linear and nonlinear features were ex⁃tracted from the signal,among which 15 features had the main influence on the speech signal.They were based on fundamental frequency F0,the fundamental frequency perturbation,the amplitude perturbation and the harmonic noise ratio.The optimal fea⁃tures were selected from these factors and compared by different classifiers.The experimental results show that the neural network classifier with different transfer functions has its own advantages and disadvantages;support vector classifiers is the best one with an accuracy of 95.92%.
作者
马春
李芳芳
MA Chun;LI Fangfang(School of Medical Information Engineering,Anhui University of Chinese Medicine,Hefei,Anhui 230012,China)
出处
《宜宾学院学报》
2021年第6期46-50,80,共6页
Journal of Yibin University
基金
国家自然基金面上项目(61672035)
安徽省安徽省高校自然研究重点项目(KJ2018A0285)
安徽省安徽省高校自然研究重点项目(KJ2019A0437)
安徽省高校人文研究重点项目(SK2019A0243)。
关键词
肝豆状核变性
构音障碍
声学分析
特征分类
hepatolenticular degeneration
dysarthria
speech analysis
feature classifier
