I'd like to get into voice transcription for note-taking, alongside my recent resurgence of taking written pen-and-paper notes and sketches.

Is there already some decent Linux-friendly thing to manage inbound WAV/MP3/OPUS recordings from a recording device and do transcription over it, or is this something I'll want/need to duct tape some janky nonsense together myself for?

Strict requirement that this be able to run fully offline on machines I control. No cloud services, no subscriptions, etc. I don't mind if it's paid software if it works well. If it uses ML models, it needs to be able to demonstrate the model was trained ethically on data it had permission to access.

(Bonus RFC: anyone have thoughts on the Sony ICD-TX660 as a recording gadget? The form factor being effectively the same size as a vape pen seems absolutely perfect, and it's a nice fully-offline task-built gadget that looks to be built well?)