Audio classification on a deaf machine