Indeed. I wonder how much hardware is required to run a speech-to-text ML model locally, and then proceed with whatever NLP you wish?