The key here is intonation -- the infinitely varying way in which we vary the speed and pitch of our voices in response to another human being. It's very subtle and something Amaazon, Google and Siri have been unable to capture. And it makes interacting with these interfaces awkward and annoying to me. Either it needs to be perfect or it gets relegated to a "that's neat" category and promptly shut off.
When I talk to one of these things, it's downright painful. It's like speaking to a someone who's (at least interacting in english, I'm sure its true of other languages) primary language isn't english and they are just saying a series of words without any meaning behind them. You tolerate this (although its still annoying) because you're interacting with a human being on the other end, not some white or beige box.
The issue behind this of course is these devices will likely never be able to master intonation, because doing so would be a feat that would require clear understanding of subtle context, far greater than what current NLP technology can do, and may even require a general intelligence. It might be possible, I'm not up to date on the current state of the art but certainly wouldn't be computationally feasible in a consumer product any time soon.
Until then, I'll never buy or use one of these stupid voice interfaces. I can get the same task done on my phone or whatever UI without the annoyance.