Well, after about ten minutes of work I have installed and configured Rhasspy to listen for a wake word, "grasshopper", convert my spoken phrase "What time is it?" to text, routed it to a script that outputs the time as text, and then converts that to a voice saying the correct time. It runs 100% locally on my laptop (no cloud) and waiting for the wake word barely uses my CPU. Oh, and it's 100% open source.
Here's the tutorial I followed: https://github.com/rhasspy/rhasspy3/blob/master/docs/tutorial.md
I'm digging how hackable the authors made Rhasspy. It's pythonic and the protocols are as human readable as possible. Python makes it pretty easy to find patterns in text and Rhasspy hides ~all of the complexities of voice. I can just write a script that accepts a string of words that were recognized, do whatever (call out to remote services, append the text to a file, ..), and then return text if I want the service to convert it to speech.
@trevorflowers are you going to be publishing your changes anywhere?
@mauve Right now I'm mostly playing with configurations and going through Rhasspy code. When I start making actual things I'll do it in a piblic repo.