Sound separation of polyphonic music using instrument prints

Decomposing a polyphonic musical recording to separate instrument tracks or notes has always been a challenge. Such a signal is the superposition of many separate tracks, and it is theoretically impossible to extract the component tracks without the information that was lost at the superposition. This paper introduces a new way of sound separation of mono-aural digital recordings. The proposed algorithm inputs the lost information by using a model of real instruments in order to make the separation of individual musical notes possible. The separation method mainly targets the processing and correction of musical recordings that cannot be re-recorded.