We present a bottom-up method for melody detection in polyphonic musical signals. Our approach is based on the assumption that the melodic line is often salient in terms of note intensity (energy). First, trajectories of the most intense harmonic groups are constructed. Next, note candidates are obtained by trajectory segmentation (in terms of frequency and energy variations). Too short, low-energy and octave-related notes are then eliminated. Finally, the melody is extracted by selecting the most important notes at each time, based on their intensity. We tested our method with excerpts from 12 songs encompassing several genres. In the songs where the solo stands out clearly, most of the melody notes were successfully detected. However, for songs where the melody is not that salient, the algorithm performed poorly. Nevertheless, we could say that the results are encouraging.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.