A Multi-Criteria Approach To Optimization Of Acoustic Feedback Detection

Irritating howling, which is caused by acoustic feedback, is an ubiquitous problem in amplified live-sound situations. In this contribution, we present a multi-criteria approach to optimal acoustic feedback detection. To do so, we consider three commonly used criteria: Peak-To-Average Power Ratio (PAPR); Peak-To-Harmonic Power Ratio (PHPR) and Peak-To-Neighboring Power Ratio (PNPR) and differences between them. In order to reach global optimal solutions, we randomly vary four parameters, which dictate the values of the criteria: the information window size; the number of peaks to select the howling from; the number of window neighbors to check and the percentage, representing the samples in a frame. In order to examine our approach, we implemented the corresponding environment. We explored our approach using three types of music: classical, jazz and rock. We discovered total correlation between the PAPR feature and the PNPR feature that reduced the number of the optimization criteria. It turned out that the global optimal values of the parameters are pretty the same for jazz and rock music. The situation for classical music is a bit different: while the optimal values of three last parameters are almost the same, the optimal value of the information window size is very different. (C) 2021 Elsevier Ltd. All rights reserved.