By Jiwei Xu, Xinggang Wang, Bin Feng, Wenyu Liu.
This code is a implementation of the experiments on Voxceleb 1 and Voxceleb 2
Our method achieved an EER of 3.48. model-3.48 link
We randomly add some noise signals to the training data during the training process as our data enhancement method. noise link
Python 3.6
Pytorch 1.2
librosa
scipy
soundfile
python_speech_features
Voxceleb 1/2 corpus can be downloaded directly from the official website.
First convert the .m4a file to a .wav file
cd convert_data
sh convert.sh
python train.py
python test.py