Shazam identifies songs by converting audio into a spectrogram, a 3D graph of frequency over time, then reducing the data to only the highest-energy peaks forming a constellation of points. It uses combinational hashing, pairing anchor points with nearby target-zone points to generate unique hashes based on frequency and time offset. Background noise is ignored because random noise points don't match the geometric pattern of the original track. These hashes are sent to a massive database where a cluster match returns a result in under two seconds, making it an elegant solution to a large-scale data search problem.
•1m watch time
Sort: