Artificial intelligence

Abstract: This paper presents a parameter-efficient approach to detecting AI-generated voices under domain shift. The task is formulated as binary classification of bona fide and spoofed speech. A frozen WavLM Base+ backbone is used as the feature encoder, while only a lightweight post-encoder adapter and a linear classifier are trained. The study follows a strict source-only protocol: models are trained on ASVspoof 2019 LA train, selected on ASVspoof 2019 LA development data, and evaluated on ASVspoof 2021 LA and the external In-the-Wild dataset. The main comparison includes no adaptation, an MLP adapter, and a KAN-inspired adapter under nearly identical trainable-parameter budgets. Experimental results show that both learned adapters substantially improve cross-domain performance compared with the no-adaptation baseline. The KAN-inspired adapter achieves the lowest mean EER on ASVspoof 2021 LA, while MLP and KAN-inspired adaptation remain close on In-the-Wild. The results indicate that lightweight post-encoder adaptation is effective for improving transfer under benchmark domain shift, whereas the advantage of KAN-inspired nonlinear adaptation is domain-dependent rather than universal.

Keywords: audio deepfake detection, speech anti-spoofing, domain shift, parameter-efficient adaptation, WavLM, KAN-inspired adapter, equal error rate

References:

1. Wang, X., Yamagishi, J., Todisco, M., Delgado, H., Nautsch, A., Evans, N., Sahidullah, M., Vestman, V., Kinnunen, T., Lee, K. A., et al. (2020). ASVspoof 2019: A large-scale public database of synthesized, converted and replayed speech. Computer Speech & Language, 64, Article 101114. https://doi.org/10.1016/j.csl.2020.101114
2. Yamagishi, J., Wang, X., Todisco, M., Sahidullah, M., Patino, J., Nautsch, A., Liu, X., Lee, K. A., Kinnunen, T., Evans, N., & Delgado, H. (2021). ASVspoof 2021: Accelerating progress in spoofed and deepfake speech detection. arXiv preprint, arXiv:2109.00537. https://doi.org/10.48550/arXiv.2109.00537
3. Delgado, H., Evans, N., Kinnunen, T., Lee, K. A., Liu, X., Nautsch, A., Patino, J., Sahidullah, M., Todisco, M., Wang, X., & Yamagishi, J. (2021). ASVspoof 2021: Automatic speaker verification spoofing and countermeasures challenge evaluation plan. arXiv preprint, arXiv:2109.00535. https://doi.org/10.48550/arXiv.2109.00535
4. Müller, N. M., Czempin, P., Dieckmann, F., Froghyar, A., & Böttinger, K. (2022). Does audio deepfake detection generalize? Proceedings of Interspeech 2022, 2783-2787. https://doi.org/10.21437/Interspeech.2022-108
5. Müller, N. M., Dieckmann, F., Czempin, P., Canals, R., Böttinger, K., & Williams, J. (2021). Speech is silver, silence is golden: What do ASVspoof-trained models really learn? Proceedings of the 2021 Edition of the Automatic Speaker Verification and Spoofing Countermeasures Challenge, 55-60. https://doi.org/10.21437/ASVSPOOF.2021-9
6. Oneata, D., Stan, A., Pascu, O., Oneata, E., & Cucu, H. (2023). Towards generalisable and calibrated synthetic speech detection with self-supervised representations. arXiv preprint, arXiv:2309.05384. https://doi.org/10.48550/arXiv.2309.05384
7. Chen, S., Wang, C., Chen, Z., Wu, Y., Liu, S., Chen, Z., Li, J., Kanda, N., Yoshioka, T., Xiao, X., Wu, J., Zhou, L., Ren, S., Qian, Y., Qian, Y., Wu, J., Zeng, M., Yu, X., & Wei, F. (2022). WavLM: Large-scale self-supervised pre-training for full stack speech processing. IEEE Journal of Selected Topics in Signal Processing, 16(6), 1505-1518. https://doi.org/10.1109/JSTSP.2022.3188113
8. Jung, J., Heo, H., Tak, H., Shim, H., Chung, J. S., Lee, B., Yu, H. J., & Evans, N. (2022). AASIST: Audio anti-spoofing using integrated spectro-temporal graph attention networks. Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, 6367-6371. https://doi.org/10.1109/ICASSP43922.2022.9747766
9. Houlsby, N., Giurgiu, A., Jastrzebski, S., Morrone, B., de Laroussilhe, Q., Gesmundo, A., Attariyan, M., & Gelly, S. (2019). Parameter-efficient transfer learning for NLP. Proceedings of the 36th International Conference on Machine Learning, 97, 2790-2799. https://proceedings.mlr.press/v97/houlsby19a.html
10. Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W. (2021). LoRA: Low-rank adaptation of large language models. arXiv preprint, arXiv:2106.09685. https://doi.org/10.48550/arXiv.2106.09685
11. Liu, Z., Wang, Y., Vaidya, S., Ruehle, F., Halverson, J., Soljačić, M., Hou, T. Y., & Tegmark, M. (2024). KAN: Kolmogorov-Arnold networks. arXiv preprint, arXiv:2404.19756. https://doi.org/10.48550/arXiv.2404.19756