This repository contains the official implementation of our SPL paper, "One-class Learning Towards Synthetic Voice Spoofing Detection". Paper link here
python==3.6
pytorch==1.1.0
The LFCC features are extracted with the MATLAB implementation provided by the ASVspoof 2019 organizers. Please first run the process_LA_data.m with MATLAB, and then run python3 reload_data.py with python.
Make sure you change the directory path to the path on your machine.
Before running the train.py, please change the path_to_database, path_to_features, path_to_protocol according to the files' location on your machine.
python3 train.py --add_loss ocsoftmax -o ./models/ocsoftmax --gpu 0
You can change the model_dir to the location of the model you would like to test with.
python3 test.py -m ./models/ocsoftmax -l ocsoftmax --gpu 0
@ARTICLE{zhang2021one,
author={Zhang, You and Jiang, Fei and Duan, Zhiyao},
journal={IEEE Signal Processing Letters},
title={One-Class Learning Towards Synthetic Voice Spoofing Detection},
year={2021},
volume={28},
number={},
pages={937-941},
abstract={Human voices can be used to authenticate the identity of the speaker, but the automatic speaker verification (ASV) systems are vulnerable to voice spoofing attacks, such as impersonation, replay, text-to-speech, and voice conversion. Recently, researchers developed anti-spoofing techniques to improve the reliability of ASV systems against spoofing attacks. However, most methods encounter difficulties in detecting unknown attacks in practical use, which often have different statistical distributions from known attacks. Especially, the fast development of synthetic voice spoofing algorithms is generating increasingly powerful attacks, putting the ASV systems at risk of unseen attacks. In this work, we propose an anti-spoofing system to detect unknown synthetic voice spoofing attacks (i.e., text-to-speech or voice conversion) using one-class learning. The key idea is to compact the bona fide speech representation and inject an angular margin to separate the spoofing attacks in the embedding space. Without resorting to any data augmentation methods, our proposed system achieves an equal error rate (EER) of 2.19% on the evaluation set of ASVspoof 2019 Challenge logical access scenario, outperforming all existing single systems (i.e., those without model ensemble).},
keywords={},
doi={10.1109/LSP.2021.3076358},
ISSN={1558-2361},
month={},}