A New Automated Approach for Identification of Respiratory Sounds

We suggest a method for automatic identification of respiratory sounds, for example, identifying wheeze from normal breath sounds. Here we apply higher order moments over time and frequency planes. The method is based on the use of efficient fast Gabor spectrogram followed by our recursively measured instantaneous kurtosis and the sample entropy. The input signal is analyzed first by using a fast Gabor time-frequency distribution into time-frequency plane. Then the normalized instantaneous kurtosis and the sample entropy are found for the time-frequency outputs of interest. Finally, the averaged instantaneous kurtosis over time and frequency and the distribution of the sample entropy provide us the useful identification indices. Illustrative results for tracheal breath and various wheeze sounds show the potential of our proposed method.