Cough is one of the most common symptoms of many respiratory diseases; the characteristics of intensity and frequency of cough sound offer important clinical messages. When using these messages, we have need to differentiate the cough sound from the other sounds such as speech voice, throat clearing sound and nose clearing sound. In this paper, based on Empirical Mode Decomposition (EMD) and Hidden Markov Model (HMM), we proposed a novel method to analyze and detect cough sound. Employing the property of adaptive dyadic filter banks of EMD, we gained the mean energy distribution in the frequency domain of the signals in order to analyze the statistical characteristics of cough sound and of other sounds not accompanied by cough, and then we found the optimal characteristics for the recognition using HMM. The experiments on clinical date showed that this optimal characteristic method effectively improved the detective rate of cough sound.
Cough ; diagnosis ; physiopathology ; Diagnosis, Computer-Assisted ; methods ; Humans ; Markov Chains ; Monitoring, Physiologic ; methods ; Pattern Recognition, Automated ; methods ; Sound ; Sound Spectrography ; methods