Song Identification through HMM-based Modeling of the Main Melody

This paper describes a methodology for the identification of pop and rock songs based on the statistical modeling of the leading voice. The identification is based on the use of hidden Markov models (HMM), which are automatically built from digital music scores. States of the HMMs are labeled by the notes of the leading voice, and the transition and observation probabilities are directly computed from the information on the score. The methodology has been experimentally evaluated on a collection of pop and rock songs, with encouraging results.