I wonder if this is a better use case for “smart” eyeglasses? Audio as the input at first, have the audio files sync wirelessly to your phone, and apply the ML transcription and prompt keys locally.