An audio server (record only) that uses librespeaker from seeedstudio (for their respeaker microphone boards) and does hotword detection and then feeds data to snips via MQTT
I'm just wondering why you are not using "AloopOutputNode" to stream the audio to a loopback device. This way, snips-audio-server could be redirected to this loopback (/usr/bin/snips-audio-server --alsa-capture hw:Loopback,1,0) and continue streaming the audio to MQTT (which could be removed from this code) and provide audio output (a feature which seems not yet to be integrated in our code).
I made a small prototype on my system at home and it seems to work fine so far. If you are interessted, I could provide a fork/pull-request....