Introduction

This repository provides the implemetation of the replayed speech detection system used in the papers:

Shih-Kuang Lee, Yu Tsao, and Hsin-Min Wang, “A Study of Using Cepstrogram for Countermeasure Against Replay Attacks,” arXiv preprint arXiv:2204.04333, 2022.

Shih-Kuang Lee, Yu Tsao, and Hsin-Min Wang, “Detecting Replay Attacks Using Single-Channel Audio: The Temporal Autocorrelation of Speech,” in 2022 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (APSIPA ASC 2022), Chiang Mai, Thailand, Nov. 2022.

Shih-Kuang Lee, “Arbitrary Discrete Fourier Analysis and Its Application in Replayed Speech Detection,” arXiv preprint arXiv:2403.01130, 2024.

Results

All systems reported in the papers are trained, validated and evaluated on ASVspoof 2019 physical access (PA) database, scores can be found in the directory CMs.

Results of the baseline system, it is the implementation of the system used in:

Lavrentyeva, G., Novoselov, S., Tseren, A., Volkova, M., Gorlanov, A., Kozlov, A. (2019) STC Antispoofing Systems for the ASVspoof2019 Challenge. Proc. Interspeech 2019, 1033-1037, doi: 10.21437/Interspeech.2019-1768

Results of the single systems:

Results of the fusion systems fused with equal weight (sum of scores):

Clone repository

git clone --recursive https://github.com/shihkuanglee/RD-LCNN.git

Dependencies

pip install -r requirements.txt

Prepare data

sh prepare_PA.sh

Prepare feature

pip install -r TAC/requirements.txt
python prepare_TAC.py --path_data ../ASVspoof2019 --task PA

Training, validation and evaluation

python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section CQT
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section LFCC
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section DCT
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section Spec
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section Spec1724
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section Ceps
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section Ceps1724
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section TAC --dmode_train fixed --dmode___dev fixed --dmode__eval fixed
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section ADFA
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section MDFA
python main.py --cuda_vd 0 --path_data ../ASVspoof2019 --task PA --conifg_section CQA

Citation Information

Shih-Kuang Lee, Yu Tsao, and Hsin-Min Wang, “A Study of Using Cepstrogram for Countermeasure Against Replay Attacks,” arXiv preprint arXiv:2204.04333, 2022.

@article{lee2022study,
  title={{A Study of Using Cepstrogram for Countermeasure Against Replay Attacks}},
  author={Shih-Kuang Lee and Yu Tsao and Hsin-Min Wang},
  journal={arXiv preprint arXiv:2204.04333},
  year={2022}}

Shih-Kuang Lee, Yu Tsao, and Hsin-Min Wang, “Detecting Replay Attacks Using Single-Channel Audio: The Temporal Autocorrelation of Speech,” in 2022 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (APSIPA ASC 2022), Chiang Mai, Thailand, Nov. 2022.

@INPROCEEDINGS{Lee2211:Detecting,
  AUTHOR={Shih-Kuang Lee and Yu Tsao and Hsin-Min Wang},
  TITLE={{Detecting Replay Attacks Using Single-Channel Audio: The Temporal Autocorrelation of Speech}},
  BOOKTITLE={2022 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (APSIPA ASC 2022)},
  ADDRESS={Chiang Mai, Thailand},
  MONTH={nov},
  YEAR={2022}}

Shih-Kuang Lee, “Arbitrary Discrete Fourier Analysis and Its Application in Replayed Speech Detection,” arXiv preprint arXiv:2403.01130, 2024.

@article{lee2024arbitrary,
  title={{Arbitrary Discrete Fourier Analysis and Its Application in Replayed Speech Detection}},
  author={Shih-Kuang Lee},
  journal={arXiv preprint arXiv:2403.01130},
  year={2024}}

Licensing

This repository is licensed under the ISC License.

This repository includes modified codes from Librosa, also ISC licensed.