Deep multi-metric learning for text-independent speaker verification

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

Dependencies

Python 3.6

Pytorch 1.2

librosa

scipy

soundfile

python_speech_features

Download Dataset

Voxceleb 1/2 corpus can be downloaded directly from the official website.

Preprocess data

First convert the .m4a file to a .wav file

cd convert_data
sh convert.sh

Train model

python train.py

Test model

python test.py

Thanks to the Third Party Libs

metric_learning