That's way more computationally expensive and only available if the camera is on (and the user's mouth is visible).