[reaperducer]

I think it depends on how you define "high quality."

To me, I'll take a restricted set of actions if they work very reliably.

This is what I had back in the 80's with a Covox Voicemaster plugged into to the joystick port of my Commodore 64. It could only understand a few phrases, but I could define those phrases, and it almost always worked.

If you define "high quality" as being able to respond to a seemingly infinite number of queries, but only understanding and replying correctly occasionally, then Siri is closer to what you want.

[dublin]

Totally agree. I too had the Covox voice recognizer on an SX-64 doing voice-controlled x10 (and other) home automation in the '80s. The amazing thing is that despite the fact that an RPi 4 has more power than a Cray had back then, modern voice recognition really isn't much better than it was then. (Although it was pretty speaker-dependent...)

I have a handful of Echo Dots and Shows in places I don't mind the security risk, and they are maddeningly incompetent at doing anything in the real world other than telling the weather and acting as a voice-controlled radio (their main use...)

It would be interesting to go back to the Covox approach and rebuild it for today's tech from the ground up (shouldn't need the hardware anymore...), as it worked surprisingly well on computers that had less resources than many (most?) of today's microcontrollers...

[passerby1]

Probably it's not fully correct, but feels like the motivation behind Echo and similar is not home automation, but a feeding ad networks with the personal details about user.