Empirical-Channel-CM

An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems

This repository contains our implementation of the paper, "An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems". [Paper link] [arXiv] [Video presentation]

Cross-Dataset Studies

Existing datasets: ASVspoof2019LA, ASVspoof2015, VCC2020 training data, VCC2020 submissions

Augmented data:

Training + Development: ASVspoof2019LA-Sim v1.0

Evaluation: ASVspoof2019LA-Sim v1.1

Channel Robust Strategies

Run the training code

python3 train.py -o /path/to/output/the/model

The options:

--AUG use the plain augmentation

--MT_AUG use the multitask augmentation

--ADV_AUG use the adversarial augmentation

Run the testing code

python3 test.py -m /path/to/the/trained/model --task ASVsppof2019LA

The options for testing on different dataset:

ASVspoof2019LA, ASVspoof2015, VCC2020, ASVspoof2019LASim

The code is based on our previous work "One-class Learning Towards Synthetic Voice Spoofing Detection" [code link] [paper link]

Citation

@inproceedings{zhang21ea_interspeech,
  author={You Zhang and Ge Zhu and Fei Jiang and Zhiyao Duan},
  title={{An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems}},
  year=2021,
  booktitle={Proc. Interspeech 2021},
  pages={4309--4313},
  doi={10.21437/Interspeech.2021-1820}
}

Please also feel free to check out our follow-up work:

[1] Chen, X., Zhang, Y., Zhu, G., Duan, Z. (2021) UR Channel-Robust Synthetic Speech Detection System for ASVspoof 2021. Proc. 2021 Edition of the Automatic Speaker Verification and Spoofing Countermeasures Challenge, 75-82, doi: 10.21437/ASVSPOOF.2021-12 [link] [code] [video]