Fake-Mamba — Speech Anti-Spoofing Arena wrapper

Xuan et al., IEEE ASRU 2025 — Fake-Mamba: PN-BiMamba blocks as an alternative to self-attention.

Packaged for the Speech Anti-Spoofing Arena from the authors' code and paper.

Fake-Mamba (PN-BiMamba): a fully fine-tuned wav2vec 2.0 XLS-R 300M front-end (last transformer layer) projected to 144-d (BatchNorm + SELU), followed by pre-norm bidirectional Mamba blocks (one shared Mamba scans the sequence forward and time-reversed, outputs summed, LayerNorm + GELU feed-forward, residual), attention pooling over time and a linear 2-way head. Trained on ASVspoof2019 LA train (RawBoost). Weights: the authors' release Exp65 (12 PN-BiMamba blocks, not the paper's 7-block configuration), 5-best checkpoint average exactly as their eval script computes it, (2*best_1 + best_2 + best_3 + best_4)/5 (a state_dict aliasing bug drops best_0); reproduces the authors' In-the-Wild scores (r = 0.99996, EER 5.83% vs 5.85% reported). The upstream code and weights carry no license, so the weights are linked from the authors' Hugging Face repo, not re-hosted. FP32, deterministic first-66800-sample window (repeat-padded).

• Weights: not re-hosted — the upstream code and weights carry no license. The wrapper downloads the authors' files from the commit-pinned checkpoint folder.
• Params: 321.2901 M. Trained on ASVspoof2019 LA train only.
• Score: logits[:, 1] (higher = bona fide). Input: first 66,800 samples @16 kHz, shorter clips repeat-padded; FP32. J-SPAW_LA was scored with the PyTorch wrapper; the other datasets with the TensorRT drop-in class (FP32), which matched PyTorch on the full J-SPAW_LA set.

Arena results

EER% (lower is better); EmoSpoofTTS and LRLspoof are spoof-only and use 1-SRR% at the model's DeepVoice EER threshold. Per-dataset scores are under .eval_results/.

Citation

@inproceedings{xuan2025fakemamba,
  title     = {Fake-Mamba: Real-Time Speech Deepfake Detection Using Bidirectional Mamba as Self-Attention's Alternative},
  author    = {Xuan, Xi and Zhu, Zimo and Zhang, Wenxin and Lin, Yi-Cheng and Kinnunen, Tomi},
  booktitle = {Proc. IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)},
  year      = {2025},
  eprint    = {2508.09294},
  archivePrefix = {arXiv}
}