This repository provides the overall framework for training and evaluating audio anti-spoofing systems proposed in RawBMamba: End-to-End Bidirectional State Space Model for Audio Deepfake Detection.
To train RawBMamba:
python train.py -o ./save_path/
To test RawBMamba on ASVspoof2019LA:
python ./ASVspoof2019LA_eval/19LA_test.py -o ./model/
To test RawBMamba on ASVspoof2021LA:
python ./ASVspoof2021LA_eval/21LA_test.py -o ./model/ -e ./model/21LA_eval.txt
To test RawBMamba on ASVspoof2021DF:
python ./ASVspoof2021DF_eval/21DF_test.py -o ./model/ -e ./model/21DF_eval.txt
We found that there is variance in model training, which can sometimes result in better outcomes than those reported in the paper. These are our experimental findings.
| Models | 19LA | 21LA | 21DF | ||
|---|---|---|---|---|---|
| EER(%) | t-DCF | EER(%) | t-DCF | EER(%) | |
| ours | 1.19 | 0.0360 | 3.39 | 0.2726 | 15.85 |
We provide pre-trained RawBMamba. Run the following code in the root directory of RawBMamba-main, and remember to modify the file paths accordingly.
To evaluate RawBMamba on ASVspoof2019LA:
python ./ASVspoof2019LA_eval/evaluate.py
To evaluate RawBMamba on ASVspoof2021LA:
bash ./ASVspoof2021LA_eval/evaluate.sh
To evaluate RawBMamba on ASVspoof2021DF:
bash ./ASVspoof2021DF_eval/evaluate.sh
@inproceedings{liu2023leveraging,
title={Leveraging positional-related local-global dependency for synthetic speech detection},
author={Liu, Xiaohui and Liu, Meng and Wang, Longbiao and Lee, Kong Aik and Zhang, Hanyi and Dang, Jianwu},
booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages={1--5},
year={2023},
organization={IEEE}
}
@article{mamba,
title={Mamba: Linear-Time Sequence Modeling with Selective State Spaces},
author={Gu, Albert and Dao, Tri},
journal={arXiv preprint arXiv:2312.00752},
year={2023}
}
If you use this codebase, or otherwise find our work valuable, please cite RawBMamba.