Direct and latent modeling techniques for computing spoken document similarity | IEEE Conference Publication | IEEE Xplore