Speech recognition underwent a sea change in 2010, when a team of researchers at a Microsoft research lab outside Seattle built a new kind of speech recognition system using a method called “deep learning.”