TY  - RPRT
TI  - Optimizing expected word error rate via sampling for speech recognition
AU  - Matt Shannon
PY  - 2017
UR  - https://arxiv.org/abs/1706.02776
ID  - 1706.02776
ER  -