At some point, someone is going to provide a local speech recognition engine with modern, neural network algorithms. Currently the folks who could do it aren't probably interested, because they want to tie you to their online services. Anyone who does it will have to be someone not beholden to that business model. But it's obviously doable at this point. The question is whether anyone thinks that it's worth the investment to do it.

It would clearly take over the automation market in terms of providing voice control, because installers would get out of all of the issues involved in integrating cloud based SR systems to automation systems. It's pretty messy. But, beyond that, I'm not sure if they could find enough applications for it to justify doing.

We support the Echo, and we also have a local SR solution. But our local solution uses previous generation SR technology because that's all that is available for local use right now, at least in any commercially ready sort of way (i.e. not a doctoral thesis experiment.) So the recognition isn't as good for our local solution. The advantage is that it's all local and that, because it's under our control, it automatically can read our room/device configuration and know what devices are in what rooms and build up the appropriate control grammar on the fly and it knows which mics are in which rooms and therefore what devices the commands apply to and so forth, which is a huge advantage. It makes setting up voice control so much easier (and a voice control that adapts easily to system configuration changes.)

If someone would just provide us with a cutting edge, all local SR solution, we could get the best of both worlds. But no one probably cares about us off in our non-commodity automation world. The commodity market will just stick the simplistic control offered by things like the Echo outside of integration into a serious automation system. And the non-commodity market won't provide sufficient numbers to entice a company big enough to do the work to do the work.

I could be wrong, and hopefully I am, but probably not.

BTW, ultimately the REAL game changer (and scary bit) will be when voice recognition comes along and gets really solid. Then, we not only know what was said, but who said it. This opens the door for personalized voice control, and also for allowing administrative type tasks to be done via voice control, which currently isn't possible (or not advisable) because there's no way to limit access to those commands.

OTOH, it takes the spying on your customers potential three or four notches higher. VR would almost certainly imply an ability to guess age and sex pretty accurately, and even more since any such system would likely include the ability to configure "this voice is Sally", "this voice is Bob", and here is some nice information about Sally and Bob that will help you collect actionable data on them to maximum potential.