Improving DNN Speaker Independence with I-vector Inputs

   Abstract