Music Structure Discovery Based on Normalized Cross Correlation

Alexander Wankhammer; I. Vaughan L. Clarkson; Andrew P. Bradley
DAFx-2010 - Graz
Music Structure Discovery (MSD) for popular music is a well known task in Music Information Retrieval (MIR). The proposed approach tries to find the basic musical structure of a piece of music, by applying a template matching algorithm on a modified, bar level Self Distance Matrix (SDM). Mel frequency cepstral coefficients (MFCC) are used to represent timbral qualities of the audio material while chroma vectors are selected to incorporate pitch and harmonic content. The new idea of template matching instead of trying to find explicit blocks or off-diagonal lines is independent of any specific characteristics of the underlying SDM and can therefore be used on a wide range of different songs.