REpeating Pattern Extraction Technique (REPET), a Novel and Simple Approach for Separating the Repeating “Background” from the Non-repeating “Foreground” in a Mixture

Abstract

As far as music is concerned the repetition forms the core. The repeating structure forms the characteristic of the musical piece and superimposed by various elements especially in the case of pop music. Proposed work concentrates on the Indian folk music and music voice separation from audio. The current available separation techniques are well suited for some typical music like REPET goes well with pop music as it has a constant repeating. Thus the study tries to present REPET for separating the repeating background from non repeating forebackground. The process starts with identifying the segments that repeats periodically, further comparison of this with the repeating segment derived model and then extraction of the repeating patterns using the time frequency masking. The Linear Predictive Coding (LPC) technique is used by most of the voice recognition devices in order to get input speech signal spectral information. LPC is used to generate observation vectors that are used by voice recognizers. On the basis of the parameters like Signal to Noise Ratio (SNR) the two techniques LPC and REPET are compared in the present work.