In order to achieve a more human-like soccer game, the robots should not only recognise a whistle but, more importantly, decide which whistle event is relevant and which is ‘noise’ from visitors or games from adjacent fields. Furthermore, each robot should be able to determine the sound source independently, without wireless messages between the robots, as this will allow a fair comparison with human teams in the future. For this task, we propose a neural network approach to evaluate the time-of-arrival difference and a second approach to estimate the distance of the sound source using the direct-to-reverberation ratio (DRR) and spectral analysis. We were able to achieve accuracies of a few degrees for direction estimation, in particular, with distance estimation also delivering reasonable results. Furthermore, we were able to implement the entire system efficiently on the NAO robot’s integrated hardware.