I've seen the data they collect by default, and your itunes collects more data than this thing does. What gets reported back to MS in audio is average ERLE (echo cancellation performance), average recognition success rate, and any software errors that get caught. And that's if you opt in. _Nothing_ gets sent back or stored if you do not opt in. Cloud recognition necessarily sends your voice to the cloud for matching, but most of the system does not use cloud recognition, only conversational recognition tends to, but it's no different than using siri. It only activates when you ask it to, it's not sending a constant stream to the cloud.