A big data approach to acoustic model training corpus selection

   Abstract