Improved System for Identification of Live Music Performances by Dynamic Time Warping

Abstract

The identification of a song performed at a concert, called the "live version," is not yet a popular search among users because of the song possibly having a different arrangement or other modifications, unlike audio searches for an exactly matching song. For live song identification, we examined the application of dynamic time warping to chroma feature series extracted from both studio and live versions of a song. In this paper, we especially focused on measuring the linearity of the warping path. It was evaluated using datasets collected from the internet and the results were compared to those of an existing search service, which showed that the proposed method has the highest accuracy.