I just wouldn't be comfortable using a voice interface in public, even for innocent stuff like walking down the street and asking it where's the nearest McDonald's. See, now I have to deal with people's opinion of McDonalds and their silent judging stares. Go to hell, I like McDonald's.
But whoever is trying to engineer useful voice recognition, I hope they succeed, because it's utter crap now. Google Now on Jelly Bean is comically useless (except for a few hundred hardcoded queries), and from what I've heard, Siri is even worse. What they need is some context awareness and learning abilities to improve recognition. If you start a query with "What is the capital of-", it should know it relates to a country or a state. Amazingly, Google Now doesn't. Also, when the user finally gives up and runs the same search in the device's browser, the voice interface should draw some conclusions from it and incorporate them in subsequent searches.