The authors develop an understanding of sound beyond shallow analysis paradigms, and relax many of the constraining assumptions. They also present structured models for computational modeling in sound using weak or no supervision. Novel algorithms and a hierarchical framework are provided.