Auditory attention saliency computation model is one of the fundamental problems in the study of auditory attention model, and the key of this model is the selection of appropriate features. In this paper, an auditory significance calculation model based on sparse dictionary learning is proposed from the view of feature selection. The first step is getting the characteristics of a variety of acoustic signals by the K-SVD dictionary learning algorithm. Then the dictionary set is classified and integrated. Based on a selected feature dictionary, OMP algorithm is used for signal sparse representation. And the sparse coefficients are combined frame by frame to obtain the auditory saliency map. The simulation results show that this auditory saliency map computation model can achieve better correspondence characteristic with the nature attribute of acoustic signal in feature selection. The saliency map based on dictionary of basic characteristics can highlight the structure characteristics of noisy acoustic signal. The saliency map based on dictionary of special characteristics can achieve selective attention for certain signals.