I hate the HA implementation of voice assist. It’s either “what you are saying has to be recognized as EXACTLY this string” or it’s “try an LLM lol”.
Hie the fuck do you influence if your voice recognition model spits out “To-do, ToDo, todo, To-Do, To Do or to do”? Because if you try to add an item to the ToDo list, then this difference matters to HA.
The answer is: by matching what is recognized against all known entities, and picking (with a threshold) the closest match. But that’s just not possible currently.
Same. Wanted to give some AI a try for my HA and it was fun. But this was very early on when they didn’t have official implementation of intents and such for AI so I sometimes had to say “You didn’t turn on the lights. Please turn the lights on”.
![[Spoilers] Realistic "2001: A Space Odyssey" rule](https://lemmy.world/pictrs/image/b46daf52-01b3-45a0-8067-2429a7086361.jpeg){kind=link}
This is a perfect demonstration of how trying to use ollama with Home Assistant went for me.
I hate the HA implementation of voice assist. It’s either “what you are saying has to be recognized as EXACTLY this string” or it’s “try an LLM lol”.
Hie the fuck do you influence if your voice recognition model spits out “To-do, ToDo, todo, To-Do, To Do or to do”? Because if you try to add an item to the ToDo list, then this difference matters to HA.
The answer is: by matching what is recognized against all known entities, and picking (with a threshold) the closest match. But that’s just not possible currently.
Same. Wanted to give some AI a try for my HA and it was fun. But this was very early on when they didn’t have official implementation of intents and such for AI so I sometimes had to say “You didn’t turn on the lights. Please turn the lights on”.