Melody Information Retrieval System Using Dynamic Time Warping

Abstract

With the increasing popularity of music, there is an intensive need for an efficient and natural way of querying an audio database in order to retrieve the desired songs. This paper presents a system that accepts user input, usually a tune hummed or sung by the user, which accounts for the most natural way of retrieval. A modified form of Dynamic Time Warping technique is used as a similarity metric in order to suit for various humming/singing habits of the user. This technique is employed to compare a given user input to a number of songs stored in the database and performs an exhaustive search to return results as a list of matching songs.