有效特征参数分类正常与病理语音  

Classification of normal and pathological speech by effective feature parameters

作　　者：郭乐乐 曹辉[1] 李涛 GUO Le-le;CAO Hui;LI Tao(School of Physics and Information Technology, Shaanxi Normal University, Xi’an 710100, Shaanxi, China)

机构地区：[1]陕西师范大学物理学与信息技术学院

出　　处：《声学技术》2019年第5期554-559,共6页Technical Acoustics

基　　金：国家自然科学基金资助项目(11074159、11374199)

摘　　要：采用残差信号的特征参数——基音幅值(Pitch Amplitude, PA)和频谱平坦度(Spectral Flatness of the Residue Signal,SFR)与语音信号倒谱域特征参数——倒谱峰值突出(Cepstral Peak Prominence, CPP)来区分正常与病理语音,在萨尔布吕肯语音数据库中选择自然音调的正常与病理语音/a/进行仿真实验。统计结果表明,与正常语音相比,病理语音的PA较小,SFR 更接近零,CPP 也较小。结合其他传统特征参数分析对比,证明SFR、PA和CPP 更能有效分类正常与病理语音。通过不同分类算法比较,得出支持向量机的分类准确率相对更高。The feature parameters PA (pitch amplitude) and SFR (spectral flatness of the residue signal) and the vowel cepstrum domain feature parameter CPP (cepstral peak prominence) are used to distinguish between normal and pathological speeches. In the Saarbruecken Voice Database, 216 normal and 216 pathological natural tones /a/ are se- lected for experiments. The statistical results show that compared with normal speech, the PA value of pathological speech is smaller, the SFR value is close to zero, and the CPP value is also smaller. Combined with other features analysis and comparison, it is proved that SFR, PA, and CPP are excellent and stable feature parameters for normal and patho- logical speech classification. The classification accuracy obtained by support vector machine is relatively higher by the comparison of different classification algorithms.

关 键 词：残差信号 基音幅值 频谱平坦度 倒谱峰值突出 支持向量机 

分 类 号：H107[语言文字—汉语]