Song recognition

or: a star map of sound for the grocery store song

Your phone records about 10 seconds of noisy room, redraws it as a map of which pitches were loudest at which moment, throws away everything except the brightest peaks so it looks like a chart of the night sky, and turns pairs of those stars into fingerprints¹. Then it checks them against millions of songs, in a database that started with 30 people feeding 100,000 CDs in by hand², and since it launched in 2002 as a text message service you reached by dialing 2580, it has identified more than 100 billion songs³.

And we use it to name the song in the grocery store, then never play it again.

How it works, badly

#
It records a few seconds and draws a map of the loudest pitches.
#
It keeps only the brightest peaks, like a star map.
#
It checks the pattern against millions of songs.
#
Then we name the grocery store song and never play it again.

Wait, is that real?

Yes, your phone really does turn music into constellations.

The star map↩

The app records a snippet of 5 to 10 seconds and keeps only a compact fingerprint. It turns the sound into a spectrogram, a map of how loud each frequency is over time, then filters out all but the loudest peaks, leaving what’s called a constellation map. Each peak is paired with nearby peaks, and every pair becomes a hash built from the two frequencies and the time between them. Shazam’s Avery Wang published the method in 2003.

The database↩

The song with the most matching hashes, lined up in the right timing, wins. Shazam’s database covers millions of songs, and its first million fingerprints were built by 30 workers loading 100,000 CDs by hand over months.

The 100 billion↩

Shazam launched in 2002 as a text message service in the U.K.: you dialed 2580, held your phone up to the music, and got the song’s name by text. Apple says it passed 100 billion song recognitions in November 2024, about 12 for every person on Earth, and has more than 300 million monthly users. Shazam joined Apple in 2018.

The fine print↩

The 10 seconds is the top of the 5-to-10-second range; it often needs less. The CD-loading story comes from an article retelling Shazam’s early days, not from Shazam itself. Apple announced the deal in 2017 and closed it in 2018. Nobody tracks how many identified songs are never played again; we are guessing it’s most of them.

Sources: Towards Data Science: the five-second fingerprint inside Shazam; Apple Newsroom: Shazam hits 100 billion song recognitions

Last fact-checked October 29, 2026.