In a noisy room, a person picks out a desired voice thanks to two ears capturing the difference in sound arrival. A computer with several microphones does the same: it gets spatial cues — where each voice is coming from.
Previously, these cues were used to focus on one direction, like a flashlight beam. But when voices overlapped, the focus blurred and part of the signal was lost — like trying to spot two actors at once in a spotlight beam.
The new method ditched the beam: the algorithm is directly told the speakers' locations. Then something unexpected emerged: with such 'geo-tags', the model recognizes speech better than after the most thorough cleanup. It reduces uncertainty — entropy — and corrects mistakes by linking sound to direction. This resembles spectroscopy, where the spectrum pattern reveals the composition, not just capturing brightness. Even in astronomy it's similar: the right viewpoint helps see faint cosmic dust clouds.
The human ear determines sound direction to within a degree thanks to an interaural delay of just a few microseconds.