Mirex 2013 “audio Onset Detection” Submission: Monophonic-atic Algorithm

In this extended abstract, we present an onset detection method for monophonic melodies based on hysteresis on the pitch-time curve. This method is especially designed to perform note segmentation in the case of a-capella singing, in which the pitch evolution during the same note can behave very unstable. The selected approach estimates the regions in which the chroma is stable. Then, a note segmentation stage based on pitch intervals of the sung signal is carried out. To this end, we perform an average of the pitch values of each new note as a representative value of its global pitch, and then we measure the instantaneous pitch deviations with respect to such average. When a sustained / large deviation is detected, a note change is considered, an onset mark is placed and the process restarts.