False, they're doing on device speech recognition. Servers are only used if you request information from the Internet. Look at the demos and read the hands on reports. On device recognition makes it much more usable than Google Voice Actions.