Monday, February 15, 2010

Mel-frequency cepstrum coefficients (mfcc) in audio

The word 'cepstrum' reverses the first 4 letters of 'spectrum', it takes Fourier transform to the decibel of spectrum. The process is:
signal -> FFT -> abs()&square -> log -> FFT -> abs()&square -> spectrum

MFCC is a nonlinear "spectrum-of-a-spectrum". It represents the short-term power spectrum of a sound.

The MFCC procedure is:
audio signal -> FFT -> map to mel scale -> log -> DCT -> MFCC (the amplitudes of the result spectrum)

The matlab code of MFCC is in following links: link0, link1, link2, Auditory Toolbox

No comments:

Post a Comment