Voice command classification - On Prem

I want to do a voice command classification without using speech to text or cloud technologies. This is a followup of question spoken-utterance-classification-on-ravdess-using-mfcc which did not get much attention.

Using cloud speech to text is easy but dont want to stay offline. Current on prem speech to text solutions do not have decent performance.

Would anyone like to suggest some implementations which actually are working.