Music Similarity Analysis through Repetitions and Instantaneous Frequency Spectrum

Abstract

This paper presents a new approach for tunes similarity calculation based on repetitions. Information about repetitions in tunes is important since repetitions make very significant impression to listener. We are providing a way to describe tunes in descriptors which contain frequency information about repetitions of tunes. Frequency information is retrieved by means of a relatively new signal processing approach called instantaneous frequency spectrum (IFS). Further tunes comparison takes this repetitions frequencies information from one tune descriptor and compares to second tune's descriptor. As a result we obtain the similarity between compared tunes. We show that proposed approach gives meaningful information about music pieces similarity and it can successfully be used in music signal processing tasks. 