摘要： 咳嗽中包含丰富的病理信息，可以为临床诊断提供重要支持。自动咳嗽检测方法有助于提高检测结果的可靠性，并减少人为工作量。但在自然记录的语音信号中，非咳嗽信号的数量远多于咳嗽，语音流中咳嗽信号的自动检测是个典型的类别不均衡问题。针对该问题，提出一种基于偏最小二乘分类法的咳嗽信号检测模型APLSCX。利用非对称偏最小二乘分类器处理类别不均衡数据的能力，对归一化的特征向量进行特征抽取，同时基于低维数据的方差调整分类平面。实验结果显示，与LCM、SVM等主流模型相比，APLSCX兼顾了小类的召回率和精度指标，具有较高的检出率和较低的误警率，更适用于自然语流中咳嗽信号的检测。

关键词: 咳嗽信号检测, 类别不均衡, 偏最小二乘法, APLSCX模型, 端点检测

Abstract: Cough contains a lot of pathological information, which is helpful for clinical diagnosis. Automatic cough detection method is helpful to improve the efficiency and reliability of this task, and it can reduce the artificial workload. When implementing the method, the amount of cough signals is much less than that of other signals in the collected corpus. Therefore, automatic cough detection in audio recordings is a typical class imbalance problem. Aiming at this problem, this paper proposes a novel imbalance classification method named APLSCX for the detection of cough signals. It uses the ability of asymmetric Partial Least Squares(PLS) classifier for processing class imbalance data, to extract feature of the normalized feature vector. At the same time, it adjusts classification plane based on the variance of low dimensional data. Experimental results show that APLSCX can increase the detection rate of cough while keeping the false alarm rate at a low level. Compared with Leicester Cough Monitor(LCM) and Support Vector Machine(SVM) method, it has higher detection rate and lower false alarm rate, and is more suitable for detecting cough signals in audio recording.

Key words: cough signal detection, class imbalance, Partial Least Squares(PLS) method, APLSCX model, endpoint detection

中图分类号: 

• TP18