Abstract
A sound source separation that identifies mixed tones from more than one sources is one of the important task when transcribing from a musical acoustic signal to a musical score. This paper presents a method of sound source separation using GA. In this method, after segmentation in time domain, tone parameters such as notes or volume are estimated by minimizing error between input spectrogram and a spectrogram which is synthesized by some templates using these parameters. The minimization of parameters is done by GA and Hill-climbing. We will show that this method costs lower than normal methods.
