Improvement of Sound Source Localization Performance using Flat Group Delay Samples

Abstract

As a sound source localization method using two microphones, GRODMIC(GROup Delay between MICrophones)has been developed. GRODMIC estimates the time differences of arrival between microphones based on the group delay characteristics corresponding to a propagation delay characteristic identified by a digital filter. In this process, a lot of ripples often occur in the group delay characteristics because of estimation residuals and other reasons, those ripples often degrade the estimation accuracy. Therefore, it is effective to extract reliable group delay samples. In this study, just high reliability group delay samples of a single speech signal are extracted, and thus the estimation performance improvement is attempted. The effectiveness of the method is shown through several experimental results.