@bugsounet
I imagine its complexity, but I know your skills and I know you would still do a great job!
The advantage of this solution would be that the voice recognition part would be done only by Google Home (which is much more powerful than our makeshift microphones connected to the RPI) which would launch various commands to MM (YouTube, Wikipedia searches, weather, etc …).
In this way you could having two modules:
1- Google Assistant with voice recognition by means of a microphone connected to the RPI
2- Google Assistant without the voice recognition part that receives commands from Google Home (or Alexa …).
Think about it … .I offer myself as a tester!