Not nearly as fast as cloud recognition has. Especially with all three of the big companies having serious cloud-based recognition offerings and focusing on them.
And seriously, Cortana would be useless without the cloud aspect. Half the things it does revolve around connecting your digital life together by accessing various things about you. Without the cloud, it's literally just Windows 8 search.
How is that? Desktop systems respond quickly, cloud systems often respond after a second - it takes time for data to do a round-trip over crappy mobile connection. Mobile latency is a big thing.
I agree wrt. Cortana (and Siri, Google Now) - most of the things you use them for will require Internet access. But there are still a lot of things that you could do with voice that shouldn't require a network connection, and we're missing the ability now. Not to mention you have zero customization options for cloud-based recognition. I could make good ol' MS Speech API recognize pretty much anything I wanted it to. No problem making it recognize a limited subset of two languages at the same time. With cloud-based systems, if the voice recognition doesn't like my accent, I'm out of luck.
I was referring to the improving part. joosters said desktop recognition has been improving, I meant that cloud has been improving faster. Which is true, all the big companies are focusing on cloud-based recognition. The improvements might eventually come to desktop recognition systems, but even that's not a guarantee given where the focus seems to be.
And seriously, Cortana would be useless without the cloud aspect. Half the things it does revolve around connecting your digital life together by accessing various things about you. Without the cloud, it's literally just Windows 8 search.