[sliverstorm]

... could easily be done on a local machine.

I think you're going to need to provide some evidence to support that. It is not intuitively obvious in the slightest, especially given how frequently Cortana/Siri/GoogleNow trip up even with a data center backing them.

Seriously, if processing locally was so easy but they still wanted to steal all your data, they would crunch it locally & give results, and silently ship off the recording to an archive in parallel. Local processing is cheaper for them (bandwidth, compute) and faster for you.