Apologies if I’ve missed something in my searches, I just ordered a couple of Voice PE’s. I’m looking to replace Alexa with something way cooler, we basically only use Alexa to control Lifx, alert us to the Ring doorbell and check the weather. Alexa isn’t connected to HASS.
I’m running Home Assistant locally on a NUC.
So is there a good beginners getting started guide that shows the current best way to set this up for good results?
I see references to the voice pipeline, conversion agent, speech-to-text and text-to-speech. That all makes sense.
I want start with being able to check sensor values and perform actuations on stuff managed by HASS, after that I would love to setup getting decent answers to general knowledge questions, not sure I want to pay OpenAI for that though.
EDIT: another thought, say I wanted to turn my living-room Daikin A/C on to cool to 25°C, how do you figure out what to say?
I myself have not found a beginner’s guide. There are various chapters of “Year of the Voice” that might be a place to start, but its a lot to read and some of it may not be applicable anymore.
It sound like you want to do stuff locally for now. Here are some pointers:
It would be good to learn about “Intents” and the associated “Sentences” and their “Responses”. I would start with the Starter Pack.
If your requests to the voice assistant don’t quite work, first try out the debug from the UI’s Voice Assistant that you are using to see what the STT captured.
Maybe the STT didn’t capture your speech correctly. Then you’ll have to play with the models say those provided by Whisper (along with the processing times) to see if there are better models. Or maybe the name of the entity your are trying to act on is too complicated for the STT to get right, in which case you can use an “alias” (even multiple aliases) to simplify the name.
If the STT works ok, you may have to dive into the details of the actual built-in intents here starting with the “sentences” (depends on your language too) to find out why the sentence pattern didn’t match. If you find there isn’t an sentence intent to match your request, you can “customize your own or extend from the built-in sentences” yourself, as well as sentence triggers.