Peak tracking and partial formation of music signals

In this paper we provide a method for detection of peaks in spectral representations of music signals for the purpose of partial tracking. The basic idea is to detect local maxima in any digitized signal and use statistical techniques for rejecting spurious peaks. Detected peaks are then connected to each other to form partial tracks. The performance of our algorithm is investigated in two levels. First the output of peak detection algorithm is compared with another method. Second, these two sets of peaks are fed into our partial tracking system and the results are compared, which confirms the superiority of our strategy. This superiority is in consistency of our method in detecting the valid tracks and preventing spurious peaks from forming false tracks.