I don't get your suggestion. How does the database tie into speech to text?