Detection and Discrimination of Adventitious Frames of Respiratory Sounds*

This paper describes the automatic recognition method which classifies every short frame of the respiratory sounds. Discrimination of every short frame enables not only detection of adventitious sound, but also separation and classification of respiratory sounds which contain different sounds at different times. Discrimination consists of two stages. First by means of the mean square of residual error, respiratory ‘sounds are classified into two classes, continuous sounds and discontinuous ones. Secondly each class is subdivided by means of feature parameters based on AIC (Akaike Information Criterion) which are transformed from PARCOR (Partial Autocorrelation) coefficients. The standard category sets are defined by R. Murphy's medical training tape, and the effectiveness of this method is shown by the discrimination experiment.