For starters, this can be an efficient way to set up voice activation. Sound signals are generally temporally compact, and somewhat sparse in Mexican hat basis.