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