Integrating thumbnail features for speech recognition using conditional exponential models | IEEE Conference Publication | IEEE Xplore