Mengyuan Zhao 2015-10-26

来自cslt Wiki
2015年10月26日 (一) 01:16Zhaomy讨论 | 贡献的版本

(差异) ←上一版本 | 最后版本 (差异) | 下一版本→ (差异)
跳转至: 导航搜索
  • specialist-dnn
  1. code: fix some bugs
  2. run some experiments on 100h+20h cn_en data. get some results but performance is bad.
  3. try to fix the code, separate the Softmax function for cn/en outputs.
  • lstm-am
  1. test on Aurora4, wsj, tencent datasets. Results show that LSTM is not good as DNN on small dataset. (DaoHangQuan is an exception.)