Empirical analysis of content-based music retrieval for music identification

Over the past few years, digitized music in forms like MP3 has made a great impact on the way of acquiring and listening to music. Due to the advanced communication tools, the consumers may search and purchase their favorite music online without going to physical music stores. Consider that a user occasionally gets an unknown and preferred music episode, but she/he has no idea on how to identify the query terms to retrieve the music using traditional textual-based search engines. Thereupon content-based music retrieval for music identification serves as an adequate solution for the users to search the targeted music effectively and conveniently. In this paper, we present several methods and similarity functions designed to achieve the effective content-based music identification. In particular, we compare the performances of various similarity functions with different musical features under real environments. The experimental results on real music datasets reveal that the Hamming Distance can bring out very robust performance for content-based music identification in terms of accuracy. In addition to music identification, this paper with detailed empirical analysis also provides the researchers with insightful ideas in other real applications such as audio monitoring, musical copyright, etc.