Learning Mid-Level Auditory Codes from Natural Sound Statistics