Evaluating MCC-PHAT for the LOCATA Challenge - Task 1 and Task 3

TL;DR

This paper evaluates the MCC-PHAT sound source localization method on the LOCATA Challenge tasks, showing it generally outperforms the MUSIC method in accuracy and reliability for static and moving speakers.

Contribution

The study applies MCC-PHAT to LOCATA tasks and compares its performance with MUSIC, demonstrating improved localization accuracy and robustness.

Findings

MCC-PHAT outperforms MUSIC in most localization scenarios.

MCC-PHAT provides more reliable and accurate source estimates.

Evaluation uses OSPA metric for quantitative comparison.

Abstract

This report presents test results for the \mbox{LOCATA} challenge \cite{lollmann2018locata} using the recently developed MCC-PHAT (multichannel cross correlation - phase transform) sound source localization method. The specific tasks addressed are respectively the localization of a single static and a single moving speakers using sound recordings of a variety of static microphone arrays. The test results are compared with those of the MUSIC (multiple signal classification) method. The optimal subpattern assignment (OSPA) metric is used for quantitative performance evaluation. In most cases, the MCC-PHAT method demonstrates more reliable and accurate location estimates, in comparison with those of the MUSIC method.