Masking is the perceptual effect that lossy compression depends on: a loud sound makes nearby quieter sounds inaudible, either in frequency (a strong tone hides softer tones close to it) or in time (a loud transient hides sounds just before and after it). The ear simply doesn't register the masked components, so a codec can quantize them coarsely or discard them entirely without audible loss. Every psychoacoustic model is essentially a masking calculator, estimating a moment-by-moment threshold below which detail can be thrown away. Understanding masking explains why lossy codecs can drop the majority of the data yet sound transparent — they remove what was already inaudible.
