refer to the A highly robust Audio fingerprinting System "
Philips Audio Fingerprint extraction process:
Step One:
resample to 5kHz,down-mix to Mono.
Step Two:
Frame processing, frame length and frame movement see literature.
Step Three:
(1) add hanning window,FFT transform, get the amplitude spectrum of each frame
(2) frequency pre- 300~2000hz, divided into a sub-band, similar to the bark domain
(3) obtain the energy of each sub-band and
Step Four:
Get a child fingerprint
Step Five:
A child fingerprint, as an audio fingerprint.
Simulation results:
The first figure is the first fingerprint of Song 1 .
The second figure is the first fingerprint of Song 1 compressed to 192kbps .
The third figure is the first fingerprint of Song 2 .
The fourth figure is the fingerprint difference between Figure 1 and Figure 2 .
The fifth figure is the fingerprint difference between Figure 1 and Figure 3 .
Audio fingerprint emulation (Philips)