Automatic training set segmentation for multi-pass speech recognition | IEEE Conference Publication | IEEE Xplore