“Mati Staniszewski predicts that high-quality, on-device voice models that do not require cloud connectivity will be deployed in cars within the next two to three years.”