A Recommender System for Music Less Singing Voice Signals

摘要

With widespread access to internet and huge availability of music players like iPhone, Smart phone songs are now available anytime and anywhere to any people. The problem now is not the availability of songs, but to find right song for right person. Song recommendations are grouped by different aspects like melody, rhythm etc. This paper proposes a recommender system by melodic similarity of songs. Fundamental frequency of song has been extracted and a dynamic programming approach named dynamic time warping has been used to find total fundamental frequency deviation of two songs and this process is continued for all song tracks in a playlist and finally recommendations are given as ascending order of overall fundamental frequency deviation percentage. This system also can give a rhythmic rating based on how a target song deviates with respect to a reference song.