摘要
咳嗽是一种在呼吸疾病中常见的症状。对病人的咳嗽类型进行分类和统计对病人的病理分析将有极大帮助。提出了一种基于MFCC特征和DTW模板匹配的方法来对病人的咳嗽进行自动干湿性分类。通过训练咳嗽样本,使用特征提取算法得到它们的MFCC特征参数从而生成用于比较的参考模板库。然后对需要进行分类的咳嗽信号进行同样的特征提取过程,并将参数和模板库中的进行匹配从而得出咳嗽的干湿性类别。文中对78个未知类型的咳嗽样本进行分类,得出干性咳嗽57个,湿性咳嗽21个,分类错误率为7.69%。经进一步处理,可以将分类错误率减少到仅为2.56%。
Cough is a common symptom of respiratory diseases.The classification and analysis of cough sounds will provide valuable clinical information in the assessment of patients＇diseases.In this paper an automated classification method is proposed based on Mel-Frequency Cepstral Coefficien（tMFCC） and Dynamic Time Warping（DTW） to separate dry cough from wet cough.The features are extracted from training data using the MFCC algorithm and saved as a template database.Then the unknown cough sounds can be classified as dry or wet cough by using the DTW algorithm to compare their features with the features in the template database.The experiment results show a misclassification between＇dry＇and＇wet＇cough sounds for only 7.69% of the all cough samples.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第13期209-212,共4页
Computer Engineering and Applications
