In this work, we propose a note detection approach based on harmonic matching pursuit (HMP) and specifically designed to detect simultaneous notes. However, HMP is not able to decompose harmonic sounds in different harmonic atoms when their fundamental frequencies are harmonically-related. To solve this problem, we propose an algorithm, called atomic spectral smoothness (SS), that works over the harmonic atoms obtained by HMP. This algorithm is based on the spectral smoothness principle which supposes that the spectral envelope of a harmonic sound usually forms smooth contours. Our proposal shows promising results for polyphonic musical signals with two harmonically-related note-events.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.