Evaluation of MFCC estimation techniques for music similarity
Jesper Højvang Jensen, Mads Græsbøll Christensen, Manohar N. Murthi, Søren Holdt Jensen · 2006
Spectral envelope parameters in the form of mel-frequency cepstral coefficients are often used for capturing timbral in-formation of music signals in connection with genre classifi-cation applications. In this paper, we evaluate mel-frequency cepstral coefficient (MFCC) estimation techniques, namely the classical FFT and linear prediction based implemen-tations and an implementation based on the more recent MVDR spectral estimator. The performance of these meth-ods are evaluated in genre classification using a probabilis-tic classifier based on Gaussian Mixture models. MFCCs based on fixed order, signal independent linear prediction and MVDR spectral estimators did not exhibit any statisti-cally significant improvement over MFCCs based on the sim-pler FFT. 1.