This is just not really how pipeline voice ai works, the speech model doesn't know the semantic feeling of its response