Study of the cepstral coefficient probability density function
Jean‐Yves Tourneret, B. Lacaze, Francis Castanie · 2003
Cepstral coefficients, used in pattern recognition and classification with the k-nearest-neighbor method, give far better results than classification with the centroid distance rule. This paper proposes an analysis of cepstral coefficient probability density which reveals why the k-NN rule is in many instances a necessary tool in this particular representation space.>