Lantian Li 2015-08-31
1. Go on my three deep speaker embedding tasks:
1). large-scale deep speaker vector framework -- hold.
2). knowledge transfer for i-vector -- in process.
Dev.: Fisher; Eva.: SRE08.
The ivector knowledge transfor(IKT) on DNN has been done.
The training processes have been done with different learning rate and network structure.
Besides, frame-level randomization and utter-level randomization both have been done.
Evaluation process is still in process. (SRE08 for female has about 100 utterances which have no features as voiced).
3). derive binary i-vectors using Hamming distance learning (identification efficiency Exp.)
Results have shown in CVSS 373. Identification and verification efficiency experiments are still in proess.
2. Help Pro.Zheng write two patents.
1. Go on the 2) and 3) of the task 1.
2. Go on the task 2.