ASDF is a framework for training anti-spoofing neural networks for automatic speaker verification.
- Multi-GPU training with DistributedDataParallel
- Multi-GPU scoring of audio files of variable lengths
- Clone ASDF repository
- Navigate to a folder, where you want ASDF to be placed to
git clone https://github.com/vvestman/asdf.gitcd asdf
Use an environment such as the following:
conda create -n asdf python=3.12conda activate asdfpip install -r requirements.txt
- See instructions from asdf/recipes/wav2vec_aasist/README.md
- For more information on how to execute and configure experiments, see asdf/src/settings/README.md
- To train neural networks by using multiple GPUs in parallel, see multigpu_readme.md
- To create custom network architectures, see custom_architectures_readme.md