This paper tries to remind to those that develop speech recognition engines the possibility and importance of separating sounds depending on the spatial position of the sources that generate them.