Ilyushin / resnet

ResNet models

This library contains ResNet models, such as ResNet 34, ResNet 50, and functionality for helping train them on VoxCeleb1 for the speaker recognition task. Some parts of the architecture were taken from http://www.robots.ox.ac.uk:5000/~vgg/publications/2019/Xie19a/xie19a.pdf TensorFlow implementation - https://github.com/WeidiXie/VGG-Speaker-Recognition

Installation:

chmod +x ./build_local.sh
./build_local.sh

Execution:

resnet_models -t \
    -p \
    -a resnet_34 \ 
    --input-dev ./vox1/dev/wav/ \
    --input-eval ./vox1/tests/wav/ \
    -p \
    -o ./data/ \
    --save-models ./tests/models/ \
    -b 300

Pre-trained models

1. ResNet 34