So, really weird issue. When I run Home Assistant Assist on my watches (Galaxy Watch6 and Galaxy Watch7), and I use the voice recognition on HA (externally offloaded to a wyoming whisper server), and I don’t say anything, it always transcribes “Thanks for watching!” - usually (95% of the time) or “I think we’re in” or something similar - much less often. So it thinks I said that. Always one of these two sentences. If I do say something it does properly recognise what I’m saying, whisper is a really good STT.
Of course the LLM doesn’t really understand what to do with this. And gives weird answers. When I use it on my voice preview edition this doesn’t work, it just knows I’m not saying and cancels the request.
Has anyone else seen this happen? And knows a solution perhaps?
It’s a bit annoying because sometimes the assistant triggers from my headset and I just get a weird response back like “Oh hey what are we watching tonight?”
PS This is exacerbated by the feature request I logged - where the Android app would give a beep when it’s time to speak. If that would happen I would know the assistant had triggered and I could cancel it.
