Audio Coding Based on Long Temporal Contexts