My understanding is that these devices (Siri, Echo) do their speech processing online, so by design everything gets uploaded in order to be interpreted. How hard would it be to have a device like echo that didn't do its speech processing locally and only went online to get data it didn't have? So if I ask for an alarm in an hour, or to play a song I own, or something else that doesn't require data from the "cloud" the command never leaves the house- and, in fact, all conversations stay private and only the "fetch" is external to the system.