论文部分内容阅读
咳嗽的自动分类在临床上具有重要的辅助诊断作用。传统的Mel频率倒谱系数(MFCC)采用Mel均匀滤波器组,高频段的滤波器分布较稀疏,未能最大程度反映两类咳嗽的特征差别。针对这个问题,本文在分析干性咳嗽和湿性咳嗽频谱能量分布特点的基础上,提出了一种改进的反向MFCC提取方法,采用反向Mel刻度上的均匀滤波器组,并放置在两类咳嗽都具有高频谱能量的频段,使得特征提取集中在两类咳嗽特征信息丰富且差别显著的频段进行。基于隐马尔可夫模型的咳嗽干湿性自动分类实验结果表明,该方法获得了优于传统MFCC的分类性能,总体分类准确率从89.76%提高到了93.66%。
Cough automatic classification has clinically important ancillary diagnostic role. Traditional Mel Frequency Cepstral Coefficients (MFCC) use Mel uniform filter bank, high-frequency filter is sparsely distributed, failed to reflect the maximum difference between the two types of cough characteristics. To solve this problem, based on the analysis of the energy distribution characteristics of the spectrum of dry cough and wet cough, an improved reverse MFCC extraction method is proposed, which adopts the uniform filter set on the reverse Mel scale and placed in two types Coughing has a band of high spectral energy that allows feature extraction to be focused on two types of frequency-rich and significantly different bands of cough characteristics. Experimental results of cough wet-dry classification based on Hidden Markov Model show that the proposed method has better classification performance than traditional MFCC, and the overall classification accuracy is improved from 89.76% to 93.66%.