A study on the song title query by melody humming

Abstract

The emergence of various audio data types, especially on the internet, is increasingly requiring more robust and easier information retrieval methods suitable for the specific properties of the target audio data. An effective and natural way of querying a musical audio database is by humming the tune of the wanted music. But the music query by humming is a challenging problem since the humming signal inevitably contains much more variation and inaccuracy in its pitch, duration, and energy than normal speech signal. In this paper, we suggest two systems for querying a wanted song from music database by humming its melody. Systems proposed in this paper are named as the normalized note length selection system and the approximated string matching system according to their characteristics.These systems for querying an audio database by humming are described along with a scheme for representing the melodic information of a song as relative pitch changes. And in order to accept or to adapt the people``s inaccurate humming, a new melody representation technique is also proposed. Our systems is basically a pitch information-based one and perform fairly well. Our performance evaluation produced 83.5% and 81% of correct query rate of the song with the normalized note length selection and the approximated string matching system, respectively, for the top 3 matches when tested with 20 songs.