Difference between revisions of "Lantian Li 2015-03-16"

From cslt Wiki
Jump to: navigation, search
(以“Weekly Summary 1. Explore the generalization of d-vector for text-indedenpent Speaker Recognition. The experimental results show that: d-vector performs better tha...”为内容创建页面)
 
 
Line 9: Line 9:
 
The experimental results show that: global CMVN without speaker CMVN performs the best which meets the expectations.
 
The experimental results show that: global CMVN without speaker CMVN performs the best which meets the expectations.
  
3. Train a text-content-based neural networks and extract d-vectors from these networks. But results show this method does not work.
+
3. Train a text-content-based (block-mask) neural networks and extract d-vectors from these networks. But results show this method does not work.
  
 
Next Week
 
Next Week
  
 
1. Go on the task1 and task2.
 
1. Go on the task1 and task2.

Latest revision as of 09:14, 16 March 2015

Weekly Summary

1. Explore the generalization of d-vector for text-indedenpent Speaker Recognition.

The experimental results show that: d-vector performs better than i-vector only under cosine distance. While LDA and PLDA do not work for d-vector.

2. Explore the impact of CMNV on the d-vector for Speaker Recognition.

The experimental results show that: global CMVN without speaker CMVN performs the best which meets the expectations.

3. Train a text-content-based (block-mask) neural networks and extract d-vectors from these networks. But results show this method does not work.

Next Week

1. Go on the task1 and task2.