Spectral Hash
Spectral hashing hashes the frequency-domain features of a signal (FFT/DCT coefficients), used for audio fingerprinting and spectral similarity comparison.
Purpose
Provides audio-recognition fingerprints robust to time shifts and compression — the basis of services like Shazam.
Mechanism
FFT/DCT is applied to the signal; coefficients of energy-concentrated bands are selected and quantized into a bit string. Frequency-domain features are invariant to time shifts (only phase changes), hence robust to temporal offsets.
Use cases
- Audio fingerprint recognition
- Spectral similarity
- Music retrieval