Acoustic Fingerprint
A compact representation used to match audio by perceptual and temporal features rather than bit-for-bit identity.
An acoustic fingerprint is not a cryptographic hash. The goal is to preserve a match when the same recording has been re-encoded, level-shifted, or moderately corrupted.
Matching Model
In the landmark approach associated with Shazam, stable spectrogram peaks are encoded as time-frequency pairs. A useful match is not a single hash collision; many landmarks should agree on a consistent time offset. This structure maps naturally to an inverted index for large reference collections.
Do Not Confuse With
Acoustic fingerprinting identifies content. Speaker embeddings target speaker identity. Perceptual hashing, ASR, and voice biometrics may all derive features from audio, but they optimize for different invariants.
Source
- https://www.ee.columbia.edu/~dpwe/papers/Wang03-shazam.pdf