Re: the irony, the irony
I played with that on the early 43P desktop systems with Ultimedia Services installed in about 1995,and I also had access to ThinkPad Power 850's when they became available, as we used them as Out-of-Hours laptops for the on-call specialists in the IBM AIX Suppport Centre (although the modem cards kept failing, we ended up using external modems for the dial up).
It really wasn't that sophisticated, although I have to admit that I was only using the voice control aspects of CDE, but I wasn't that impressed (down, down, click). I had seen similar technology over the preceding 10 years, using various hardware assists on things as humble as the BBC Micro (although that was using a custom Fast Fourier Analysis piece of kit for the hashing of the words, which the BEEB then looked up in a simple database).
Shortly after voice control of AIX systems, there were multiple voice recognition systems for PC's running Windows. I remember Dragon Naturally Speaking and IBM's own Via Voice, but there were others.
But it took until Apple, Amazon, Google and Microsoft built natural speech recognition into consumer devices for this technology to really take off. That was a very long 10 years (something that seemed to always be the expected timeline) from the early 1980s to actually come to fruition.
I would have liked to see the technology for local speech recognition to have happened, however, rather than having it shipped out to the cloud for processing. Even an Amazon Echo Dot has more processing power than the hardware that started doing reliable voice recognition in the 1980/1990s. Maybe it was only the data harvesting capabilities that served as the final enabler for this technology.