Investigating audition with a generative model of impact sounds