In this paper we study the suitability and performance of conventional time delay of arrival estimation for humanoid robots. Moving away from simulated environments, we look at the influence of real-world robot's shape on the sound source localization. We present a TDOA/GCC based sound source localization method that successfully addresses this influence by utilizing a pre-measured set of TDOAs. The measuring methodology and important aspects of the implementation are thoroughly presented. Finally, an evaluation is made with the humanoid robot Nao. The experimental results are presented and discussed.