The Method of Main Vocal Melody Extraction Based on Harmonic Structure Analysis from Popular Song

Abstract

In this paper, we propose the method of main vocal melody extraction based on harmonic structure analysis technique from polyphonic music signal. It is the most important part of contents based music retrieval method which has mainly three parts. The first part is pitch estimation from humming signal, the second one is the melody extraction from polyphonic music signal and the last one is the matching engine which measure the distance between two vectors. The accuracy of melody extraction affects the overall system performance rather than any other parts. Human vocal track makes the harmonics like most musical instruments. This is one of the most important things that we have considered to utilize. So, we might extract the main vocal melody from the complicated mixed signal with musical instruments. We utilize harmonic structure analysis and track pitch sequence during three frames include current frame. The proposed method contains three major blocks named preprocessing, multi-pitch extraction with peak picking, fundamental frequency detection and the last part with pitch tracking, predominant melody detection. We have started this project with aiming for supporting commercial service for music portal provider, KARAOKE system and mobile devices.