I was trying to get it to pause Pi-hole on request. I’m using Home Assistant Cloud (Nabu Casa) for speech to text, and have also got OpenAI plugged into it for if it doesn’t recognise a command. The screen shot is from the debug logs that I eventually found after struggling to work out why it wasn’t running my automation.
I’m using the new Home Assistant Voice Preview. Don’t get me wrong, overall very happy with it for the price point, but for some reason the cloud speech recognition (I believe powered by Google) is very good at understanding me until I start trying to talk about ad blocking.
Use a local, open source voice recognition and the problems with understanding ad blocking related phrases should disappear. 😁
It’s running on a Pi 4, they highly recommend cloud in this case because it’s not very powerful.
But I also speak with a New Zealand accent, and could not get it to understand a single word I said until I connected it to their cloud option.
The OpenAI integration coupled with cloud text to speech/speech to text has actually been fantastic. It works a lot better than I was expecting based on their warnings of it being very early days. The main problem is the data from home assistant is fed in as a prompt, and so to keep it short it only sends the states. Data stored as attributes is not accessible, so I’m making helpers, scripts, and automations for the assistant to trigger in order to output more data.
You can get the same from a local whisperer model to be clear.