To complement Arjan P’s comment, dither is usually applied to help remove quantization effects caused by discrete steps in volume caused by transitions from one bit level to the next at very low volumes (least significant bits, LSB’s). If not applied, the sound can be “grating” or “grainy” for music and other types of audio signals. The smaller the bit rates, the more necessary dither becomes (say from 18 to 16 to 14, etc.). And there are a number of technical studies that have determined that application of dither actually permits listening to sounds below the threshold limits of the playback media! This sounds contradictoraly weird, but psychoacoustic testing proves it out. 16 bit recordings such as CD’s generally have a noise floor around -96dB below Full Scale (FS), however with proper dithering audible sounds can actually be detected below this level up to about -6 dB further below the theoretical media noise floor.
Of course, very few folks ever find a noise floor in a live environment that comes close to being below -70dB FS unless their doing measurements in an anechoic chamber. But as Arjan has noted, reverb tails, and other very quiet passages or “digital black” bands of no signal between tracks, may be quite noticeable for this “grainy sound” if dither is not applied.
What’s also interesting is some of the newest dither techniques apply dither at very high frequencies approaching the brick wall cutoffs at the top of the sampling spectrum. Most of this very high frequency dither is virtually inaudible, but leads to very effective dithering throughout the rest of the audible spectrum of the recording. One can actually see it’s implementation through use of WL’s spectrum analyzers; but most folks would be extremely hard pressed to actually hear the noise effects (similar to tape hiss) of the dither.
Hope this info. proves helpful.