Lantian Li 15-05-11

Weekly Summary

1. Make complementary experiments on discriminative feature training based on DNN / SVM.

Exp. results show that the performance of DNN-based system is better than SVM-based system under most conditions.

Yet, the DNN-based system is greatly influenced by the DNN struture.

2. Help ZiYang research on the effect of speech rate for speaker recognition.

Results show that for normal-rate trained speaker model, the verification performance for test utterance at normal and

faster rates is better than at a slower rate.

3. Make preliminary experiments on Speech watermarking.

Next Week

1. Go on the task 1,2,3.