A supervised uncued classification approach for a class of multicomponent signals

A class of nonstationary signals composed of activities which are multicomponent in nature and "sufficiently" nonoverlapping in time is considered. It is assumed that for each activity, there is some distinct region where its spectral energy predominates the rest of the components as well the background noise. Uncued classification of these activities in the signal is a challenging problem as it requires automatic segmentation. We present sufficient conditions for segmentation of activities. We provide a methodology to achieve supervised uncued classification by classifying activities class by class. This methodology is tested on phonocardiogram signals.