Optimizing expected word error rate via sampling for speech recognition

Re-assign community