I am having a difficult time getting intent recognition working on 2.5.5. I have tried Kaldi, Deepspeech, and Pocketsphinx. Each has similar problems. Either there is no recognition, or it is very slow, or both. In every case, after I speak a command, there is ~5 seconds before the feedback beep comes from rhasspy. Pocketsphinx is the only one thus far to be able to process the intent properly, but its hit or miss (correct 1 out 3 times). I am on a 3B+ with matrix voice on top and an english profile. I have tried using my NUC as the master rhasspy instance, but that doesn’t work either. In addition, I have tried adjusting the VAD sensitivity and the silence after settings, but neither helps, only makes it worse
First of all you are actually talking about speech to text and not intent recognition.
Did you test it using a wakeword or did you use the button in the web interface?
What was your problem with the rhasspy master on your NUC?
Did you choose open transcription or mixed as that would explain why it is so slow and not working that well ?
As i use a german profile i dont know what the best option is for english. Kaldi is slower but gives more reliable results for german atleast. I am using a pi4 for rhasspy and i am getting a response in a second or so most of the time.