Notice of Removal: Melody Extraction from Polyphonic Music Signals Using Tandem Filter System

Abstract

Melody extraction is a key problem in music information retrieval. This paper presents a method to estimate human voice fundamental frequency in polyphonic music signal using tandem filter. There are three-stage filters in this method. At first stage, it’s easier to extract human voice fundamental frequency if a singing voice separation preprocessing is obtained. This paper uses the robust principal component analysis to roughly extract the human voice from polyphonic music signal. At the second stage, it is believed that although the human voice spectrogram is reserved and the accompaniment effect is reduced, there is still noise or false fundamental frequency. Thus, a hyper-Fourier Transform is implemented to lower the noise. At the third stage, the STFT spectrogram have the original information of the melody frequency structure. So this paper uses a transverse stripe filter system to eliminate the non-possible fundamental frequency position.