Audio fingerprinting
A compact summary of a recording that lets a system identify the song from a few seconds of noisy audio.
An ordinary hash won't do: a song playing in a bar, captured on a phone, shares not a single bit with the original file. In 2003 Avery Wang described how Shazam got around this. The algorithm picks out the strongest peaks in the spectrogram, pairs them up and hashes the two frequencies plus the time between them. Peaks survive noise, compression and bad speakers, and the right song is the one where many hashes agree on the same time offset.
Shazam launched in London in 2002 as a phone service: you dialled a short number and held your handset up to the speaker. Apple bought the company in 2018. The same principle powers copyright filters on video platforms, and the open AcoustID service, with its Chromaprint library, is used to tag music collections automatically.