Last week:
1. With Mengyuan, conducted experiments for small model with big data, initial results here;
2. got the conlusion that mpe improves chain, results here.
This week:
1. transfer learning(dark knowledge) for nnet3;
2. speech & speaker recognition joint training.
