Using latent topic features to improve binary classification of spoken documents | IEEE Conference Publication | IEEE Xplore