Depends what you consider vocal models 'more text'. Already we're seeing multi-modal AI have a correct answer when executing across image/text or voice/text interpretation.