it uses snowboy for hotword detection and google streaming speech reco for asr.
we support snowboy custom models, and use annyang for intent handling.
as its getting harder and harder to get snowboy running, and my experience with Home Assistant suggests maybe docker can come to the rescue
I have some test code that connects to the internal mqtt server and can see the hotword and asr responses.
but like previous years, asr results are terrible. event tho this is a pretty small model set of text, it is not exposed to the google speech reco, and we still get great results in fabulous response time.
so, I’d like to build the asr docker container for google streaming speech reco. and add it to the configurable pipeline. we take 16k frames from arecord today, so I think the mqtt frames will be similar…
today we get back partial results and final, but only process final.
I don’t see a section on creating another asr engine, anyone have pointers?