Improving Blind Source Separation Performance By Adaptive Array Geometries For Humanoid Robots

Abstract

In this paper, the concept of an adaptation algorithm is proposed, which can be used to blindly adapt the microphone array geometry of a humanoid robot such that the performance of the underlying signal separation algorithm is improved. As a decisive feature, an online performance measure for blind source separation is introduced which allows a robust and reliable estimation of the instantaneous separation performance based on currently observable data. Experimental results from a simulated environment confirm the efficacy of the concept.