Combining multiple subword representations for open-vocabulary spoken document retrieval | IEEE Conference Publication | IEEE Xplore