This git repository contains the training and analysis code used in our paper. The training script for the neural network is not included.
The paper is accepted to Interspeech 2020. The arXiV preprint can be found here
The extracted features and the raw dataset can be downloaded from this website.
After successful feature extraction, you need to create a virtual environment with the packages required. This is contained in the requirements file. Using pip
python3 -m venv venv_name source/venv/bin/activate pip install -r requirements.txt
This should ensure that every package is installed.
Finally, in order to reproduce a tab separated file containing the results in the paper you need to:
./run.sh
If you want to also reproduce the figures, then you need to call
python3 analysis.py
- EER evaluation code is from the ASVspoof 2019 baseline
- American English phonetic posterior processing is from this paper