Impact of Distance in Pitch Class Profile Computation.pdfVIP

Impact of Distance in Pitch Class Profile Computation 1 1 2 Giordano Cabral , Jean-Pierre Briot , François Pachet 1Laboratoire d’Informatique de Paris 6 – Université Pierre et Marie Curie 8 Rue du Capitaine Scott 75018 Paris – France 2Sony Computer Science Lab Paris 6 Rue Amyot 75005 Paris – France {Giordano.CABRAL,Jean-Pierre.BRIOT}@lip6.fr, pachet@csl.sony.fr Abstract. Pitch Class Profiles (PCP) [Fujishima 1999] are largely used for all applications involving harmonic content. Although the main steps of the PCP calculation are generally equal over all the scientific literature, some implementation details may vary, specially the impact of the distance between the frequency of the FFT bins and their closest note. This paper compares 6 ways of using this distance to weight the contribution of each FFT bin in the final PCP vector. We present the results of the 6 computed PCPs when used in a chord recognizer, a tonality estimator and a key detector. 1. Introduction Pitch Class Profiles (PCP) are vectors of low-level instantaneous features, representing the intensity of each of the twelve semitones of the tonal scale. They are largely used in all applications involving harmonic content, especially chord recognizers [Yoshioka at al. (2004)], tonality estimators [Gómez and Herrera 2004a] and key detectors [Pauws 2004]. The main advantages of the PCPs are the simplicity of cal ulation, the concision of the harmonic information and the power to unify various dispositions of a single chord class. Although the main steps of the PCP calculation are very similar over all the scientific literature, some implement