Wirebook › Glossary › Audio fingerprint
What is an audio fingerprint (how song recognition works)?
In one sentence: A tiny summary of a recording made from pairs of its loudest notes and the gaps between them, used to recognise a song from a few seconds of sound.
A little more
Song recognition apps don't compare sound with sound. They turn a few seconds of audio into a spectrogram, keep only the strongest peaks, and pair each peak with a few that come just after it. Each pair (one frequency, another frequency, the time gap between them) becomes a short number: a fingerprint. A song produces thousands of them.
The service fingerprints every recording in its library ahead of time and files the numbers in an index. Your phone's fingerprints are looked up, and matches vote for a song and a moment in it. The real song collects many votes at one consistent moment, while chance matches scatter. Because the method matches one exact recording, it copes well with noise but usually fails on humming, live versions or a song played at a different speed.
See it in action
- BonusHow does a phone know a song from hearing it?Hold up a phone in a noisy shop and it names the song in seconds.8 min · interactive
- BonusWhy does a video call go blurry when the connection is slow?Dial a pretend video call's bandwidth thinner, spend a tiny 10-bit budget on eyes vs mouth vs wallpaper to still read a facial expression, then reveal the clear face.7 min · interactive