MG Training (Maximum Gaussianality Training)

A Pytorch implementations of MG training, This is a general Gaussian distribution training method and can be used in any task that requires Gaussian distribution in latent space.

Datasets

trainingset:Voxceleb 
testset: SITW, CNCeleb

Following this link to download the dataset (extraction code：8xwe)

Run DNF with ML training (original DNF model )

./run_ML.sh

Run DNF with MG training

./run_MG.sh

The evaluation and scoring will be performed automatically during the training process.

Other instructions

score.py is a python implementations of the standard kaldi consine scoring, you can also use kaldi to do the plda scoring
tsne.py can be used to draw the distribution of latent space