The effect of speech denoising algorithms on sound source localization for humanoid robots

Abstract

In this paper we study the suitability and performance of four major speech denoising algorithms as weighting functions for the Generalized Cross-Correlation (GCC) and discuss the role of the microphones' frequency response on the localization estimates. An evaluation is made with the humanoid robot Nao. Moving away from simulated environments, we look at the performance of sound source localization under different in-system and ambient noise conditions. We also present a time delay of arrival (TDOA) / GCC based sound source localization method for real-world humanoid robots which was used for our experiments. This method addresses the robot's shape influence by utilizing a pre-measured set of TDOAs.