Perceptual coding
Lossy audio compression that throws away what the ear would not have noticed anyway, rather than what is mathematically least significant.
The encoder splits the audio into frequency components, usually with an MDCT, while a psychoacoustic model works out a masking threshold for every band and time slice. Bits are then allocated so that quantisation noise stays below that threshold. Where a loud sound is drowning everything around it, a couple of bits will do; where it is quiet and exposed, it takes more.
The approach underpins MP3 (standardised in 1993), Sony's ATRAC on MiniDisc, AAC and Opus. Karlheinz Brandenburg at Fraunhofer tuned MP3 by listening over and over to Suzanne Vega's a cappella "Tom's Diner", which exposed every artefact. The weak spot is that the threshold assumes one human ear and one pass — transcode a few times and the errors stack up.