131202 Fanhu Bie
1. Extract all the features from the used NIST SRE05 and 06 data
2. Train the UBM with 05 data
3. Done the experiment with the medium score, given the worse result.
1. Try the same idea with previous score.
2. Take the threshold for the score of the feature as 0.
3. Read the paper on PLDA and done the experiment on ESR for PLDA.