【作者】 李海峰； 房春英； 马琳； 张满彩； 孙佳音；

【Author】 LI Haifeng;FANG Chunying;MA Lin;ZHANG Mancai;SUN Jiayin;Harbin Institute of Technology School of Computer Science and Technology;Heilongjiang Institute of Science and Technology School of Computer and Information Engineering;

【机构】 哈尔滨工业大学计算机学院； 黑龙江科技大学计算机与信息工程学院；

【摘要】 病理语音具有强烈的非平稳性和突变性特点。S变换具有良好的时频分辨率和时频定位能力。本文在S变换基础上结合人耳听觉的Mel特性,提出一种能够突出发声器官病变的病理语音特征MSCC(Mel S-Transform Cepstrum Coefficients)。在NCSC语料库上,通过与经典语音倒谱特征MFCC和当前常用声学特征的对比,看到MSCC特征对语音中动态、快变的病理信息具有更强的刻画能力。此外,选用F-Score方法对特征进行评价和采用粒子群方法进行特征筛选,MSCC表现出了更好的分类性能。可见,MSCC特征为病理语音的高精准分析与临床诊断提供了理论基础。更多还原

【Abstract】 Pathological speech is non-stationary and mutation, the S transform has good time-frequency resolution and time-frequency position capability. In this paper, S transform is combined with the human auditory Mel characteristics, MSCC(Mel S-Transform Cepstrum Coefficients) is proposed which highlight the vocal organs pathological lesions. MSCC is compared with the classical MFCC and current commonly acoustic characteristics in NCSC corpus, MSCC has a stronger ability to portray the dynamic and quickly pathological speech information. In addition, MSCC has also better classification performance by F-Score method to evaluate and particle swarm optimization algorithm to feature selection. Therefore, MSCC provides high precision analysis for pathological speech and theoretical basis in clinical diagnosis.更多还原