Hi all,
I’m Mike, creator of the Rhasspy voice assistant. I recently joined Mycroft to help them get the Mark II over the finish line.
I’ve gotten fully offline speech to text and text to speech (for English) working on the Mark II already with HA 
Take a look:
My goal between now and when the device ships (September 2022) is to have Mycroft automatically communicating with HA to get entity names/device classes, and tuning the speech recognizer for useful commands. And to support more than just English, of course!
If anyone has questions or ideas on how this should work, I’d be interested to hear them 
P.S. For the curious: the speech to text is using Coqui STT (formally DeepSpeech) with a tuned language model, and the text to speech is a follow-up to my own Larynx TTS.