An Industrial-Strength Audio Search Algorithm

This is the paper behind Shazam: how a phone can name a song from ten noisy seconds in a pub. Read it to see how a track is reduced to a constellation of spectrogram peaks, how pairing peaks into 32-bit hashes makes lookup about ten thousand times faster, and why a match is declared only when the surviving hashes agree on one time offset.

Reading focus: Why spectrogram peaks survive shouting, traffic, and phone-codec compression when almost every other audio feature is destroyed. How pairing an anchor peak with nearby peaks turns 10-bit features into 30-bit hashes, trading about 10 times the storage for about 10,000 times the search speed. Why a histogram of sample-to-track time offsets can confirm a match with only about 1-2% of the fingerprint tokens surviving the noise.

ISMIR 2003. Avery Li-Chun Wang. 30 min read, easy difficulty.