Zhiyuan Tang 2016-05-02

Last week:

1. the enhanced joint model with fisher focusing on spk showed improvement on SRE on sre08, score from Lantian;

2. enhanced the joint model with swbd focusing on speech while keeping the weights of spk unchanged, the resulted model got nearly the same performance as the baseline;

3. enhancing the joint model with both swbd and fisher, while the training is so slow.

This week:

1. speed the training up;

2. trivial things.