A Music Similarity Function Based On Time-Frequency Analysis and Pyramid Kernel

Abstract

In this paper, content-based music similarity function from acoustic music signal in an efficient way is considered. This function is to determine similarities among songs, particularly, a piece of input music signal compared with storage music song’s signal into the database and retrieve a whole music song according to the input query. Representing the music signal having sparse nature is accomplished by Matching Pursuit with timefrequency dictionaries. In order to match a candidate segment with the query segment, the music signal similarity measure is performed by Spatial Pyramid Matching. Evaluation results on music similarity illustrate that our music similarity function is better than such previous approaches