Maybe that'll change now with ChatGPT. The problem I have with voice interfaces is you need something at the level of AGI to make it compelling. I could have an hours-long engaging voice-only conversation with a real person, but what the hell am I going to do with a dumb "smart" speaker that will be even a fraction as engaging as, say, a videogame? 99% of my smart speaker use cases are: Setting timers/alarms, querying the weather, turning on/off the lights, playing songs/podcasts/videos. For anything more interesting/interactive than that, you need more intelligence than existing systems actually have.
Summarize the following weather forecast for tomorrow
12 am, 2° F, clear, 15 mph
1 am, 2° F, clear, 15 mph
2 am, 1° F, clear, 14 mph
Tomorrow's forecast will be cold and clear with temperatures dropping from 2°F to 1°F and winds of around 15mph decreasing to 14mph.
Give it the proper data in a machine reasonable format and as it to make it into a human summary, and that's where its strength appears to be.
The problem is that GPT would be too expensive for Amazon Alexa right now.
The problem was that it was rather difficult to do for a while and the "here's a great ecosystem of games and tools" never materialized for echo devices.
Part of that is that people really don't know how to to do voice interfaces well.
Another part is that it has to run on AWS which has all the AWS sizing and billing problems that need to be solved by programmers (we're programmers - not SREs - and its really easy to rack up a too large bill on AWS).
As such, in order to try to monetize a skill (which wasn't possible at first) you needed to commit to spending some AWS money to try to get it. Discovery of Alexa skills was poor at best. Free skills normally end up costing money (and not just time) for the programmer.
And so, a possible revenue stream for Alexa has never realized.