Things like that are perfectly possible without any internet-connected services with open source software like CMUSphinx. And as an addition you can get better features like continuous listening (no need to push any button), flexible vocabulary and even speaker identification (only you will be able to open the garage). I really wonder what is cool in Siri doing that.