I am using HA Cloud voice recognition, but the accuracy of matching recognized text to the intended command is very low. Often the system transcribes the command as “zapne” instead of “zapni”, “zhasne” instead of “zhasni”, “nastavte” instead of “nastav” etc. That results in no match and command not executed (or handed over to LLM which is a completely different story).
I did some research and discovered that vosk is providing rapidfuzz as an option to try and match the intent even if it wasn’t a 100 % match.
So my question is: would it be possible to somehow integrate the two (HA cloud and RapidFuzz) together?
Thanks, jose