IT lexicon › Programming › ADPCM

ADPCM

Programming På svenska → Updated: 2026-10-28

Audio compression that stores the difference from a prediction rather than the sample itself, with a step size that adapts to the signal.

Audio rarely changes much from one sample to the next, so storing the error in a guess is cheaper than storing the whole value. That much is DPCM. The adaptive part is that the quantizer step grows when the signal moves fast and shrinks when it is calm, as Bell Labs researchers Cummiskey, Jayant and Flanagan described in 1973. Typically 16 bits become 4 — a 4:1 squeeze with hardly any maths involved.

That made ADPCM the workhorse of 1990s game audio: sound cards, arcade boards and consoles like the Super Nintendo, whose SPC700 played its own variant. Telephony's G.726 runs on the same idea, and DECT cordless phones still use it. It sounds noticeably worse than MP3 at the same bitrate, but the decoder fits in a handful of instructions.

← Back to the lexicon