论文部分内容阅读
病理语音具有强烈的非平稳性和突变性特点,较难分析。S变换具有良好的时频分辨率和时频定位能力。该文将S变换与人耳听觉的Mel特性结合,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(Mel frequency cepstrum coefficients)和当前常用声学特征的对比,表明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算法进行特征筛选,MSCC表现出了更好的分类性能。可见,MSCC特征可以为临床诊断提供病理语音的高精准分析。
Pathological speech has a strong non-stationary and mutation characteristics, more difficult to analyze. S transform has good time-frequency resolution and time-frequency positioning capabilities. In this paper, we combine the S transform with the Mel feature of human ear hearing to propose a Mel-S-transform cepstrum coefficients (MSCC) that can highlight the pathological voice features of vocal organs. On the NCSC corpus, compared with the commonly used MFCC (Mel frequency cepstrum coefficients) and the commonly used acoustic features, MSCC features show stronger characterization of dynamic and fast changing pathological information in speech. In addition, the F-Score method was used to evaluate the features and the particle swarm optimization algorithm was used for feature selection. MSCC showed better classification performance. As can be seen, MSCC features provide high-precision analysis of pathological speech for clinical diagnosis.