Comment by skribanto
maybe it’s better now, but at last time I checked (~2 years ago) humans are still better in jargon heavy contexts - legal as you mentioned, tech/medical/etc. also accents, muffled audio, noise in general, humans are (were?) much better at accounting for that
Also: A STT algorithm can't be held accountable for any errors.