Mixture densities formulation of a spectrogram segmentation task

This communication addresses the general problem of non-stationary signal interpretation. We show that the Time-Frequency Representation segmentation task can be assimilated to a mixture densities parameter estimation when performed in a statistical Features Space. Two models of mixture are discussed, setting the number of features to be extracted. We propose to resolve moments equations to characterize the energy evolution of the signal components in the Time-Frequency plane. This provides quantitative measures associated to the components which can be used in a decision procedure.