140708 - Fanhu Bie
1. Modify the ivector to use the DNN result as the weight to plus to the F0 statistics.
2. Take out the 1`s experiment, go on to modify the Loading matrix training code.
3. Code for the supervised UBM training, which will be used in the DNN-based ivector training code.
2. Modify the re-weighted T matrix training code and take out the experiment.
3. Train the supervised UBM and modify the code to make a unit.