I'm not going to spend time answering questions when you didn't read where the author of the Forbes article himself admitted he was in error. I already explained where he did that, but you didn't read it.
For the record I have already confirmed it uses server-side processing. The voice recognition stops the very instant internet connectivity is lost--right in mid-sentence. Very easy to test: unplugged the wifi router's internet connection in the middle of speaking.
A cursory check of the apps's size shows it to be only 12.7 MB. After using for a few days there is another 14 MB in data (likely a cache of previous searches). There is no way--no possible way--this kind of voice recognition can be done in that amount of code. You need upwards of a gigabyte.
Finally, it's wise to not blindly accept an "article" just because it comes from Forbes. Only a little common sense is needed to come to the realization that the author is full of crap.
Michael
Save your keystrokes! In addition to not knowing what you're talking about, it's obvious you can't participate in a civil conversation without resorting to personal attacks. Goodbye