Wildfire was doing much of this twenty years ago.[1][2] Listen to their demo. That system was way ahead of its time. It was quite popular, but compute intensive enough that the service was expensive. At its peak, Wildfire was a standard offering on Orange cellular service in the UK, but they dropped it in 2005. Somewhere in the 2000s, Microsoft bought the technology and trashed the product, although Cortana is to some extent a successor. Now a small startup has the original technology, and offers it, at an excessive price.
One key to doing this is that the system must consistently respond to the user in about 200-300ms. Long delays, as with Google voice dialing, are unacceptable. That was why it was so expensive in the 1990s; the company had to provision a lot of server capacity per user. Today, it ought to be much cheaper.
$49 a month reflects what Wildfire cost in the 1990s. Today, it should cost about $2/month.
[1] https://www.wildfirevirtualassistant.com/ [2] https://en.wikipedia.org/wiki/Wildfire_Communications