Content-based retrieval of polyphonic music objects using pitch contour

This paper investigates the retrieval of content-based polyphonic music objects in Wav and MP3 format. The system allows user to find an intended song by humming or singing a section of it. In this paper we introduce the baseline system and describe the key components including the pitch extraction in humming/singing clip, the vocal/non-vocal music segmentation, the pitch tracking in polyphonic music, and the DTW based matching algorithm. We conducted evaluations on the system. The experimental results demonstrate the feasibility of retrieving polyphonic music objects by humming/singing.