Searching for music: from melodies in mind to the resources on the web

The paper describes existing solutions that suit various scenarios of music search including audio fingerprint and query-by-humming approaches. Pitfalls of the Earth Mover's Distance algorithm (EMD) are analyzed as well as situations where the quality of comparing music patterns is not good enough. We propose the adaptation of the EMD algorithm that allows to take into account characteristics of note significance in the given music fragment. We examine the ways to formalize human music perception factors with the help of knowledge bases and statistical information gathered in process of music search.