Recently I managed to get the ChromeOS local STT engine (SODA (Speech On-Device API)) to work on an RPi4 with very impressive results. The RPi4 has plenty of power to do this, the transcription is real-time with very low WER and it has support for dozens of languages. I'd like to make an integration with HA and looked how it is done in https://github.com/rhasspy/wyoming-faster-whisper, but I could definitely use some input form people who have experience with this. A command-line client PoC is here: https://github.com/biemster/gasr.
Anyone interested in helping out?