Modified clustering algorithm for the trachea sound featuring

This study attempts to detect wheeze with k-means clustering algorithm. The subjects which included normal and wheeze sounds were evaluated. The algorithm presented a good performance to filter out noise and segment wheeze episode regions in the spectrograms. The results show that the clustering algorithm improved the signal-to-noise-ratio (SNR) from 29.02 ± 8.65 to 30.49 ± 9.01 dB in the cases of normal subjects, and 36.99 ± 9.67 to 38.29 ± 10.10 dB in the ones of wheeze subjects.