Hello,
Is there a way to add a custom networked Linux device (like RPi/Beaglebone) with a USB mic to serve as a source of audio into the Voice Assist?
Something like porcupine for keyword detect, some audio streaming input API on HA end, and some kind of method of determining end-of-phrase.
I think Voice Assist is still too fresh to support automatic keyword wakeup like “Hey Google…” or “Alexa…”. Please correct me if otherwise. I’m also looking for this capability.
The guy who is developing the whole voice part of HA previously developed Rhasspy. Rhasspy has remote satellite capabilities for speaker/mic.
So I believe all this will come to HA in due course, or you could use Rhasspy right now.
I think in the near future, HA will support this. Let’s wait together.
Im waiting 
I would love to make these calls through the google home mini/nest directly
Check out my project that uses Porcupine to detect a wake word and send the audio to voice assistant.