A General Framework for Learning Procedural Audio Models of Environmental Sounds