On The Compression Of Recurrent Neural Networks With An Application To LVCSR Acoustic Modeling For Embedded Speech Recognition

  Abstract