Re: They're all crap
The problem is that we've spent a heck of a lot of time on natural speech recognition, which overcomplicates matters ridiculously.
With our inefficient GUIs we've trained people out of the idea of learning "commands", and when it boils down to it, every speech recognition interface has some kind of assumptions coded into it that mean we really are dealing with a command set, but a fuzzy one, and one that isn't written down anywhere.
OK, so Google will give you a rough overview of the command structure if you make a mistake, but only if you make a mistake that it recognises (such as it working out that Keiran was trying to make a call) and that it recognises as a mistake (it didn't realise it was phoning the wrong person, for example).
I genuinely think people are open to a formalised compromise.
OK Google, Call [[person]] using [[application]]
seems straightforward, and quick. Getting rid of the "I would like to ..." part sets everything up for a straightforward paradigm.
OK Google, Switch on the [[device]] in [[room]]
And sticking with a fairly simple formula, you can get a reasonable degree of sophistication:
OK Google, open The Register in Chrome and display on the TV in the living room.
It's a strict subset of English, so easily learnable, and if you include a few clear variations (e.g. OK Google, use Chrome to open The Register on the living room TV), no-one's really going to notice that they're not able to say it every which way, because they'll get what they want done easily and quickly enough.