The USA Today has an interesting article on how Google, Microsoft and other voice search engines work, how they focus in on your voice while blocking out all the background noises.
Edward C. Baig from USA Today interviewed Google’s Travis Trekell amongst others at Google and other companies. In the picture below, you can see how Ed and Travis were testing out voice search within Google’s anechoic chamber lab. The anechoic chamber lab isolates the voice of the person talking, removing all echos and other noises from the room. Then what they do is overlay typical noises you’d hear in coffee shops, trains, and other busy areas, so Google can then build algorithms that remove those external noises and focus in on the searchers voice.
Google told USA Today that their error rate on voice search is only 8%, which is down from about 25% just a couple of years ago.
Between languages, accents and just very noisy places, ensuring voice search works when searchers need it is key in users adopting the feature more.