Large-scale, sequence-discriminative, joint adaptive training for masking-based robust ASR

   Abstract