Combining stochastic average gradient and Hessian-free optimization for sequence training of deep neural networks | IEEE Conference Publication | IEEE Xplore