Pytorch implementation of X-vectors embedding

This repository is using https://github.com/SiddGururani/Pytorch-TDNN repo with few adaptation, including one that allows to stack few TDNN layers sequentially

Files

TDNN layers implementation, Stats pooling and final layers implementation

The pretrained network that was trained on speechs of Trump, Lady Gaga, Macron and Hillary Clinton

Jupyter notebook for the network training

Notebook that extracts X-vectors on new utterances and via LDA reduces thier dimensions and plots them in 2D and 3D

Xvectors 3d scatter of the trained speakers and new unseen speaker - Andrew_ng

Name		Name	Last commit message	Last commit date
Latest commit History 12 Commits
.ipynb_checkpoints		.ipynb_checkpoints
README.md		README.md
TDNN_gpu.py		TDNN_gpu.py
celebs_xvectors		celebs_xvectors
data_path.txt		data_path.txt
lda-2dim-with-andrew-lda-pretrained.png		lda-2dim-with-andrew-lda-pretrained.png
lda-2dim.png		lda-2dim.png
lda-3dim-with-andrew-lda-pretrained.png		lda-3dim-with-andrew-lda-pretrained.png
lda-3dim-with-andrew.png		lda-3dim-with-andrew.png
lda-3dim.png		lda-3dim.png
total_model		total_model
utils.py		utils.py
xvector - gpu.ipynb		xvector - gpu.ipynb
xvector_ext_and_lda_scatter.ipynb		xvector_ext_and_lda_scatter.ipynb