Audio-Text Models Do Not Yet Leverage Natural Language | IEEE Conference Publication | IEEE Xplore