Video explanation
A record’s signal enters a fingerprint machine. Spectral peaks pair into fingerprints that are matched against a drawer-like database.
Spectral peaks
Audio becomes a time-frequency representation, with prominent peaks selected to reduce reliance on overall loudness.
Peak pairs form fingerprints
Pairing peak frequencies and time differences creates compact features that can match even a short excerpt.
Consistent time offsets
A match produces many hits with consistent time offsets, rather than relying on one shared peak. The machine is an algorithm metaphor.