Separation of vocals from monaural music recordings using diagonal median filters and practical time-frequency parameters

In this paper we present a novel method for using median filters to separate vocals from single channel music recordings that contain harmonic and percussive sounds. The method incorporates diagonal median filters in addition to the horizontal and vertical ones to better match the characteristics of the vocal track. Further improvements are obtained through proper choices of the filters' lengths using time and frequency units. The effectiveness of the proposed method is demonstrated using a test set of songs by the Beach Boys.