End-to-end anti-spoofing with RawNet2

This repository contains our implementation of the paper accepted to ICASSP 2021, "End-to-end anti-spoofing with RawNet2". This work demonstrates the effectivness of end-to-end approaches that utilise automatic feature learning to improve performance, both for seen spoofing attack types as well as for worst-case (A17) unseen attack. Paper link here Updated code

Installation

First, clone the repository locally, create and activate a conda environment, and install the requirements :

$ git clone https://github.com/TakHemlata/RawNet_anti_spoofing.git
$ conda create --name rawnet_anti_spoofing python=3.6.10
$ conda activate rawnet_anti_spoofing
$ conda install pytorch=1.4.0 -c pytorch
$ pip install -r requirements.txt

Experiments

Dataset

Our model for the logical access (LA) track is trained on the LA train partition of the ASVspoof 2019 dataset, which can can be downloaded from here.

Training

To train the model run:

python main.py --track=logical --loss=CCE   --lr=0.0001 --batch_size=32

Testing

To test your own model on the ASVspoof 2019 LA evaluation set:

python main.py --track=logical --loss=CCE --is_eval --eval --model_path='/path/to/your/your_best_model.pth' --eval_output='eval_CM_scores.txt'

If you would like to compute scores on the development set of ASVspoof 2019 simply run:

python main.py --track=logical --loss=CCE --eval --model_path='/path/to/your/best_model.pth' --eval_output='dev_CM_scores.txt'

Compute the min t-DCF and EER(%) on development and evaluation dataset

python tDCF_python/evaluate_tDCF_asvspoof19_eval_LA.py dev 'dev_CM_scores.txt'

python tDCF_python/evaluate_tDCF_asvspoof19_eval_LA.py Eval 'eval_CM_scores.txt'

Fusion

Fusion experiments performed using the "Support Vector Machine (SVM)" based fusion approach. We trained SVM on development scores of all three RawNet2 systems with high-spectral resolution LFCC-GMM baseline system from here and tested on evaluation scores of all the systems.

SVM fusion script (matlab) is available in 'SVM_fusion/' directory

Acknowledgement

This work is supported by the ExTENSoR project funded by the French Agence Nationale de la Recherche (ANR) and the VoicePersonae project funded by ANR and the Japan Science and Technology Agency.

Contact

For any query regarding this repository, please contact:

• Hemlata Tak: tak[at]eurecom[dot]fr

Citation

If you use this code in your research please use the following citation:

@INPROCEEDINGS{9414234,
  author={Tak, Hemlata and Patino, Jose and Todisco, Massimiliano and Nautsch, Andreas and Evans, Nicholas and Larcher, Anthony},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={End-to-End anti-spoofing with RawNet2}, 
  year={2021},
  pages={6369-6373}
}