Automatic transcribing is one of the attractive application for computer, and many software have been developed. However, most of these software can not address polyphonic sound. The process of transcription for polyphonic sound is divided into two processes; sound source separation and musical instrument identification. In this paper, we propose a method to combine comb filters for sound source separation and multi-class SVM for musical instrument identification. Experimental results showed that our method could identify musical instrument with 57.6% accuracy.