Lantian Li 2016-01-25

来自cslt Wiki
2016年1月25日 (一) 00:55Lilt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
  • Weekly Report
  • Max-margin metric learning task:
All the Switchboard data have been used as the new Dev.set.
Two tricks: i-vector shuffle and speaker selection.
  • Pair-wised speaker vector training.
  • Deep speaker embedding.
  • To Do next week
  • max-margin metric learning --> TRP.
  • Pair-wised speaker vector training --> Finish.
  • Deep speaker embedding --> the CNN structure.