Xiangyu Zeng 2016-05-02
1.repeat experiments on the wsj dataset, but met a problem that KALDI_ASSERT: at nnet-train-frmshuff-adammax-last:Eval:nnet-loss.cc:73, failed: KALDI_ISFINITE(net_out.Sum()). I guess that is caused by copied data, and the net train the same utterance continuous.
2.revised the paper of online learning, and change the dataset settings.
1.fix the bug of 1, and add NG experiments into paper.