Elamur Research⌕
☰
home / papers / 2026 / XLSR-MamBo: Scaling the Hybrid Mamba-Attention Backbone for Audio Deepfake Detection

XLSR-MamBo: Scaling the Hybrid Mamba-Attention Backbone for Audio Deepfake Detection

Paper · 2026 · Kwok-Ho Ng, Tingting Song, Yongdong WU et al.

Summary

Modern TTS and voice-conversion systems can produce fake speech that is hard to tell from real speech. Earlier Mamba-based audio deepfake detection (ADD) models made the causal SSM see both directions by adding hand-built tricks, such as running two scans and fusing them. The authors argue that spoofing traces have two sides: fine local artifacts and global spectral inconsistencies. They suggest pairing SSMs, which compress temporal patterns well, with Attention, which is good at content-based retrieval.

Their framework, XLSR-MamBo, uses a pre-trained XLSR front-end and a hybrid SSM-Attention back-end. They test four block layouts, each combined with Mamba, Mamba2, Hydra or Gated DeltaNet, at several depths. All models are trained only on ASVspoof 2019 LA.

The best setup, MamBo-3-Hydra-N3, is competitive with leading single systems on ASVspoof 2021 LA/DF and In-the-Wild. It also generalizes to diffusion- and flow-matching fakes in DFADD. The authors credit Hydra's built-in bidirectional mixing for the gains, and they find that deeper backbones reduce the instability seen in shallow ones. Code is public.

Key points

  • Hydra uses a quasiseparable-matrix sequence mixer, which gives it native bidirectional processing at linear cost. It replaces the dual-branch "BiMamba" designs used in RawBMamba, XLSR-Mamba and Fake-Mamba.
  • With backbone depth L = 5 and roughly 319M parameters, MamBo-3-Hydra-N3 reaches the following EERs:
    • 0.81% on ASV21LA
    • about 1.70% on ASV21DF
    • 4.97% on ITW
  • Compared with baselines, it improves on XLSR-Mamba by 12.90% (21LA) and 9.57% (21DF) in relative terms. It also improves on Fake-Mamba (L) by 15.04% on ITW.
  • Deeper backbones reduce run-to-run variance. Training converges quickly, usually in under 10 epochs, and the checkpoint with the lowest validation loss is not always the best on unseen data.
  • The authors note several limitations:
    • Training uses only one source, ASV19LA.
    • Fine-tuning data is mostly English, so performance on tonal or otherwise different languages (relevant for Ukrainian) has not been tested.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Jinan UniversityGuangzhou · China5 works · 11 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 works

🧭 Topics 1Research subjects and languages the work is about.Speech & spoken language · 480 works

Cite

Cite as Kwok-Ho Ng, Tingting Song, Yongdong WU and Zhihua Xia. 2026. XLSR-MamBo: Scaling the Hybrid Mamba-Attention Backbone for Audio Deepfake Detection. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1573

@inproceedings{ng2026xlsr,
  title = {{XLSR-MamBo: Scaling the Hybrid Mamba-Attention Backbone for Audio Deepfake Detection}},
  author = {Kwok-Ho Ng, Tingting Song, Yongdong WU and Zhihua Xia},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1573},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1573}
}

References

51 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Simran Arora, Sabri Eyuboglu, Michael Zhang, Aman Timalsina, Silas Alberti, Dylan Zinsley, James Zou, Atri Rudra, and Christopher Ré. 2024. Simple linear attention language models balance the recall-throughput tradeoff. In Proceedings of the 41st International Conference on Machine Learning, volume…
  2. Hussam Azzuni and Abdulmotaleb El Saddik. 2025. Voice cloning: Comprehensive survey. arXiv preprint arXiv:2505.00579. — [arXiv:2505.00579](https://arxiv.org/abs/2505.00579) find
  3. Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick von Platen, Yatharth Saraf, Juan Pino, Alexei Baevski, Alexis Conneau, and Michael Auli. 2022. Xls-r: Self-supervised cross-lingual speech representation learning at scale. In Interspeech 2022…
  4. Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in neural information processing systems, 33:12449–12460.
  5. Sarah Barrington, Emily A Cooper, and Hany Farid. 2025. People are poorly equipped to detect ai-powered voice clones. Scientific Reports, 15(1):11004. find
  6. Aaron Blakeman, Aarti Basant, Abhinav Khattar, Adithya Renduchintala, Akhiad Bercovich, Aleksander Ficek, Alexis Bjorlin, Ali Taghibakhshi, Amala Sanjay Deshmukh, Ameya Sunil Mahabaleshwarkar, and 1 others. 2025. Nemotron-h: A family of accurate and efficient hybrid mamba-transformer models. arXiv p…
  7. Edresson Casanova, Kelly Davis, Eren Gölge, Görkem Göknar, Iulian Gulea, Logan Hart, Aya Aljafari, Joshua Meyer, Reuben Morais, Samuel Olayemi, and Julian Weber. 2024. Xtts: a massively multilingual zero-shot text-to-speech model. In Interspeech 2024, pages 4978–4982. ISCA. \#\# Acknowledgments Sanyua… find
  8. This work is supported by the National Natural Science Foundation of China (Grant No. 6227134), Guangzhou Basic and Applied Basic Research Foundation (Grant No. 2025A04J3585). We use Sirui Chen, Jingji Chen, Siqi Zhu, Ziheng Jiang, Yanghua Peng, and Xuehai Qian. 2025. Mesh-attention: A new communica… find
  9. Yujie Chen, Jiangyan Yi, Jun Xue, Chenglong Wang, Xiaohui Zhang, Shunbo Dong, Siding Zeng, Jianhua Tao, Zhao Lv, and Cunhang Fan. 2024. Rawbmamba: End-to-end bidirectional state space model for audio deepfake detection. In Interspeech 2024, pages 2720– 2724. ISCA. find
  10. Tri Dao, Dan Fu, Stefano Ermon, Atri Rudra, and Christopher Ré. 2022. Flashattention: Fast and memory-efficient exact attention with io-awareness. Advances in neural information processing systems, 35:16344–16359. find
  11. Tri Dao and Albert Gu. 2024. Transformers are SSMs: Generalized models and efficient algorithms through structured state space duality. In Proceedings of the 41st International Conference on Machine Learning, ICML’24. find
  12. Phuong Tuan Dat and Tran Huy Dat. 2025. Xlsr-kanformer: A kan-intergrated model for synthetic speech detection. In 2025 IEEE International Conference on Advanced Visual and Signal-Based Systems (AVSS), pages 1–6. IEEE. find
  13. Jiawei Du, I-Ming Lin, I-Hsiang Chiu, Xuanjun Chen, Haibin Wu, Wenze Ren, Yu Tsao, Hung-yi Lee, and Jyh-Shing Roger Jang. 2024. Dfadd: The diffusion and flow-matching based audio deepfake dataset. In 2024 IEEE Spoken Language Technology Workshop (SLT), pages 921–928. IEEE. find
  14. Zhihao Du, Changfeng Gao, Yuxuan Wang, Fan Yu, Tianyu Zhao, Hao Wang, Xiang Lv, Hui Wang, Chongjia Ni, Xian Shi, and 1 others. 2025. Cosyvoice 3: Towards in-the-wild speech generation via scaling-up and post-training. arXiv preprint arXiv:2505.17589. — [arXiv:2505.17589](https://arxiv.org/abs/2505.17589) find
  15. Joel Frank, Thorsten Eisenhofer, Lea Schönherr, Asja Fischer, Dorothea Kolossa, and Thorsten Holz. 2020. Leveraging frequency analysis for deep fake image recognition. In Proceedings of the 37th International Conference on Machine Learning, ICML’20. JMLR.org. find
  16. Paolo Glorioso, Quentin Anthony, Yury Tokpanov, James Whittington, Jonathan Pilault, Adam Ibrahim, and Beren Millidge. 2024. Zamba: A compact 7b ssm hybrid model. arXiv preprint arXiv:2405.16712. find
  17. Albert Gu and Tri Dao. 2024. Mamba: Linear-time sequence modeling with selective state spaces. arxiv 2023. In The Conference On Language Modeling (COLM). find
  18. Albert Gu, Karan Goel, and Christopher Ré. 2022. Efficiently modeling long sequences with structured state spaces. In The International Conference on Learning Representations (ICLR). find
  19. Anmol Gulati, James Qin, Chung-Cheng Chiu, Niki Parmar, Yu Zhang, Jiahui Yu, Wei Han, Shibo Wang, Zhengdong Zhang, Yonghui Wu, and Ruoming Pang. 2020. Conformer: Convolution-augmented transformer for speech recognition. In Interspeech 2020. ISCA.
  20. Sukjun Hwang, Aakash Lahoti, Ratish Puduppully, Tri Dao, and Albert Gu. 2024. Hydra: Bidirectional state space models through generalized matrix mixers. Advances in Neural Information Processing Systems, 37:110876–110908. find
  21. Gautam Siddharth Kashyap, Zohaib Hasan Siddiqui, Mohammad Anas Azeez, Rafiq Ali, Shantanu Kumar, Navin Kamuni, and Jiechao Gao. 2025. Fooling the forgers: A multi-stage framework for audio deepfake detection. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processin… find
  22. Awais Khan, Khalid Mahmood Malik, James Ryan, and Mikul Saravanan. 2023. Battling voice spoofing: a review, comparative analysis, and generalizability evaluation of state-of-the-art voice spoofing counter measures. Artificial Intelligence Review, 56(Suppl 1):513–566. find
  23. Yassine El Kheir, Tim Polzehl, and Sebastian Möller. 2025. Bicrossmamba-st: Speech deepfake detection with bidirectional mamba spectro-temporal cross-attention. arXiv preprint arXiv:2505.13930. — [arXiv:2505.13930](https://arxiv.org/abs/2505.13930) find
  24. Menglu Li, Yasaman Ahmadiadli, and Xiao-Ping Zhang. 2025. A survey on speech deepfake detection. ACM Computing Surveys, 57(7):1–38. find
  25. Opher Lieber, Barak Lenz, Hofit Bata, Gal Cohen, Jhonathan Osin, Itay Dalmedigos, Erez Safahi, Shaked Meirom, Yonatan Belinkov, Shai Shalev-Shwartz, and 1 others. 2025. Jamba: A hybrid transformer-mamba language model. In The International Conference on Learning Representations (ICLR).
  26. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988.
  27. Xiaohui Liu, Meng Liu, Longbiao Wang, Kong Aik Lee, Hanyi Zhang, and Jianwu Dang. 2023. Leveraging positional-related local-global dependency for synthetic speech detection. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE. find
  28. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In The International Conference on Learning Representations (ICLR). find
  29. Joel R. McConvey. 2025. Deepfakes a ‘now problem’ as eu ai act passes compliance deadline: Reality defender. Last accessed: 2025-12-27. find
  30. Nicolas M Müller, Pavel Czempin, Franziska Dieckmann, Adam Froghyar, and Konstantin Böttinger. 2022. Does audio deepfake detection generalize? In Interspeech 2022. ISCA. find
  31. Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, and 1 others. 2022. In-context learning and induction heads. arXiv preprint arXiv:2209.11895.
  32. Jongho Park, Jaeseung Park, Zheyang Xiong, Nayoung Lee, Jaewoong Cho, Samet Oymak, Kangwook Lee, and Dimitris Papailiopoulos. 2024. Can mamba learn how to learn? a comparative study on in-context learning tasks. In Proceedings of the 41st International Conference on Machine Learning, ICML’24. find
  33. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pages 28492–28518. PMLR.
  34. Liliang Ren, Yang Liu, Yadong Lu, Yelong Shen, Chen Liang, and Weizhu Chen. 2025. Samba: Simple hybrid state space models for efficient unlimited context language modeling. In The International Conference on Learning Representations (ICLR).
  35. Eros Rosello, Alejandro Gómez Alanís, Angel M Gomez, Antonio M Peinado, N Harte, J Carson-Berndsen, and G Jones. 2023. A conformer-based classifier for variable-length utterance processing in anti-spoofing. In Interspeech, volume 2023, pages 5281–5285. find
  36. Noam Shazeer. 2020. Glu variants improve transformer. arXiv preprint arXiv:2002.05202. find
  37. Chengzhe Sun, Shan Jia, Shuwei Hou, and Siwei Lyu. 2023. Ai-synthesized voice detection using neural vocoder artifacts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 904–912. find
  38. Hemlata Tak, Madhu Kamble, Jose Patino, Massimiliano Todisco, and Nicholas Evans. 2022. Rawboost: A raw data boosting and augmentation method applied to automatic speaker verification anti-spoofing. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)… find
  39. Ruibin Xiong, Yunchang Yang, Di He, Kai Zheng, Shuxin Zheng, Chen Xing, Huishuai Zhang, Yanyan Lan, Liwei Wang, and Tieyan Liu. 2020. On layer normalization in the transformer architecture. In International conference on machine learning, pages 10524–10533. PMLR. find
  40. Xi Xuan, Zimo Zhu, Wenxin Zhang, Yi-Cheng Lin, and Tomi Kinnunen. 2025. Fake-mamba: Real-time speech deepfake detection using bidirectional mamba as self-attention’s alternative. In Proceedings of the IEEE ASRU. find
  41. Junichi Yamagishi, Xin Wang, Massimiliano Todisco, Md Sahidullah, Jose Patino, Andreas Nautsch, Xuechen Liu, Kong Aik Lee, Tomi Kinnunen, Nicholas Evans, and 1 others. 2021. find
  42. Asvspoof 2021: accelerating progress in spoofed and deepfake speech detection. arXiv preprint arXiv:2109.00537. — [arXiv:2109.00537](https://arxiv.org/abs/2109.00537) find
  43. Songlin Yang, Jan Kautz, and Ali Hatamizadeh. 2024. Gated delta networks: Improving mamba2 with delta rule. In The International Conference on Learning Representations (ICLR). find
  44. Tianle Yang, Chengzhe Sun, Siwei Lyu, and Phil Rose. 2026. Forensic deepfake audio detection using segmental speech features. Forensic Science International, 379:112768. find
  45. Biao Zhang and Rico Sennrich. 2019. Root mean square layer normalization. In Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc. find
  46. Qishan Zhang, Shuangbing Wen, and Tao Hu. 2024. Audio deepfake detection with self-supervised xls-r and sls classifier. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 6765– 6773. find
  47. Xiangyu Zhang, Qiquan Zhang, Hexin Liu, Tianyi Xiao, Xinyuan Qian, Beena Ahmed, Eliathamby Ambikairajah, Haizhou Li, and Julien Epps. 2025. Mamba in speech: Towards an alternative to self-attention. IEEE Transactions on Audio, Speech and Language Processing. find
  48. Duc-Tuan Truong, Ruijie Tao, Tuan Nguyen, Hieu-Thi Luong, Kong Aik Lee, and Eng Siong Chng. 2024. Temporal-channel modeling in multi-head self-attention for synthetic speech detection. In Interspeech 2024, pages 537–541. ISCA. find
  49. Siyi Zhou, Yiquan Zhou, Yi He, Xun Zhou, Jinchao Wang, Wei Deng, and Jingchen Shu. 2025. Indextts2: A breakthrough in emotionally expressive and duration-controlled auto-regressive zero-shot text-to-speech. arXiv preprint arXiv:2506.21619. — [arXiv:2506.21619](https://arxiv.org/abs/2506.21619) find
  50. Xin Wang, Junichi Yamagishi, Massimiliano Todisco, Héctor Delgado, Andreas Nautsch, Nicholas Evans, Md Sahidullah, Ville Vestman, Tomi Kinnunen, Kong Aik Lee, and 1 others. 2020. find
  51. Asvspoof 2019: A large-scale public database of synthesized, converted and replayed speech. Computer Speech & Language, 64:101114. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ng-2026-xlsr-mambo-scaling-hybrid-mamba/