Zhiyuan Tang 2015-09-28

Last week:

1. when changing the halving_factor from 0.5 to 0.7 for end-to-end ASR, the results seem not better;

2. nnet learning from i-vector worked not well;

3. some coding for nnet with metric loss function, not realised.

This week:

more about nnet learning from i-vector and nnet with metric loss function, CNN should be considered.