MUSIC SHAPELETS FOR FAST COVER SONG RECOGNITION

A cover song is a new performance or recording of a pre-viously recorded music by an artist other than the original one. The automatic identification of cover songs is useful for a wide range of tasks, from fans looking for new ver-sions of their favorite songs to organizations involved in licensing copyrighted songs. This is a difficult task given that a cover may differ from the original song in key, tim-bre, tempo, structure, arrangement and even language of the vocals. Cover song identification has attracted some attention recently. However, most of the state-of-the-art approaches are based on similarity search, which involves a large number of similarity computations to retrieve po-tential cover versions for a query recording. In this pa-per, we adapt the idea of time series shapelets for content-based music retrieval. Our proposal adds a training phase that finds small excerpts of feature vectors that best de-scribe each song. We demonstrate that we can use such small segments to identify cover songs with higher identi-fication rates and more than one order of magnitude faster than methods that use features to describe the whole music. 1