Audio sources are coded by recognizing different classes of audio such as speech and music. The classes are used to select between coding algorithms and to provide object definitions. Objects have abstract and concrete classes which may further rely on parameters produced by linear prediction and subband filters to provide a frame-based bit stream of information. Each object in the bit stream has layers of information such as basic bit rate, coding parameters and enhancement parameters. The layers of information in each object allow altering selected parameters to manipulate audio signals.