--- license: apache-2.0 tags: - audio - anti-spoofing - deepfake-detection - speech-anti-spoofing-arena library_name: pytorch --- # SSL-Mamba: Speech Anti-Spoofing Arena wrapper Packaged for the [Speech Anti-Spoofing Arena](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena) from the authors' [code](https://github.com/hoanmyTran/spoofing_ssl_mamba) and [paper](https://www.isca-archive.org/interspeech_2025/tran25b_interspeech.html). SSL-Mamba (Tran et al., Interspeech 2025): a fully fine-tuned wav2vec 2.0 XLS-R 300M front-end whose 25 hidden states are each projected to 128-d (Linear + SiLU) and summed (self-gated layer aggregation), a multi-kernel gated convolution (MultiConv cgMLP, kernels 7/15/23/31) fusion module, 12 unidirectional Mamba (v1 selective SSM) residual blocks with LayerNorm, multi-head attention pooling (4 heads, mean + std) and a 2-way head. Official authors' checkpoint (interspeech2025.pt), trained on ASVspoof2019 LA train. Not to be confused with XLSR-Mamba (Xiao & Das), a different model. - Checkpoint (commit-pinned): [model.safetensors](https://huggingface.co/SpeechAntiSpoofingBenchmarks/SSL-Mamba/blob/3565084c8f0125b48ca79525674a05167638e40a/model.safetensors) - Params: 317.5213 M. Trained on ASVspoof 2019 LA train. - Score: higher = bona fide. J-SPAW_LA was scored with the PyTorch wrapper; the other datasets with the TensorRT drop-in class (FP32 on the RTX 4070 Ti SUPER; FP32 on the RTX 5060 Ti), which matched PyTorch on the full J-SPAW_LA set. ## Arena results [![arena tier](https://img.shields.io/endpoint?url=https://speechantispoofingbenchmarks-speechantispoofingarena.hf.space/badge/ssl-mamba/tier.json)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![arena rank](https://img.shields.io/endpoint?url=https://speechantispoofingbenchmarks-speechantispoofingarena.hf.space/badge/ssl-mamba/rank.json)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) 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/`. [![EER% 0.27 on ASVspoof2019_LA](https://img.shields.io/badge/EER%25%20on%20ASVspoof2019__LA-0.27%25-brightgreen)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 6.36 on ASVspoof2021_LA](https://img.shields.io/badge/EER%25%20on%20ASVspoof2021__LA-6.36%25-yellow)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 3.41 on ASVspoof2021_DF](https://img.shields.io/badge/EER%25%20on%20ASVspoof2021__DF-3.41%25-green)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 14.16 on ASVspoof5](https://img.shields.io/badge/EER%25%20on%20ASVspoof5-14.16%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 5.83 on InTheWild](https://img.shields.io/badge/EER%25%20on%20InTheWild-5.83%25-yellow)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 15.72 on ADD22_eval_31](https://img.shields.io/badge/EER%25%20on%20ADD22__eval__31-15.72%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 24.08 on ADD2023_track12_test_r1](https://img.shields.io/badge/EER%25%20on%20ADD2023__track12__test__r1-24.08%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 14.55 on CFAD](https://img.shields.io/badge/EER%25%20on%20CFAD-14.55%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 16.71 on CD-ADD](https://img.shields.io/badge/EER%25%20on%20CD--ADD-16.71%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 14.43 on CVoiceFake_small](https://img.shields.io/badge/EER%25%20on%20CVoiceFake__small-14.43%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 27.25 on DeepVoice](https://img.shields.io/badge/EER%25%20on%20DeepVoice-27.25%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 2.5 on LibriSeVoc](https://img.shields.io/badge/EER%25%20on%20LibriSeVoc-2.5%25-green)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 31.57 on SONAR](https://img.shields.io/badge/EER%25%20on%20SONAR-31.57%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 22.59 on ArAD](https://img.shields.io/badge/EER%25%20on%20ArAD-22.59%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 15.51 on DECRO](https://img.shields.io/badge/EER%25%20on%20DECRO-15.51%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 5.88 on J-SPAW_LA](https://img.shields.io/badge/EER%25%20on%20J--SPAW__LA-5.88%25-yellow)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 22.74 on ODSS](https://img.shields.io/badge/EER%25%20on%20ODSS-22.74%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 1.56 on HABLA](https://img.shields.io/badge/EER%25%20on%20HABLA-1.56%25-brightgreen)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 7.3 on DFADD](https://img.shields.io/badge/EER%25%20on%20DFADD-7.3%25-yellow)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 5.53 on PyAra](https://img.shields.io/badge/EER%25%20on%20PyAra-5.53%25-yellow)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 16.45 on XMAD](https://img.shields.io/badge/EER%25%20on%20XMAD-16.45%25-lightgrey)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![EER% 0.17 on EmoFake_test](https://img.shields.io/badge/EER%25%20on%20EmoFake__test-0.17%25-brightgreen)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![SRR COMPLEMENT 89.74 on EmoSpoofTTS](https://img.shields.io/badge/SRR%20COMPLEMENT%20on%20EmoSpoofTTS-89.74%25-blue)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) [![SRR COMPLEMENT 23.46 on LRLspoof](https://img.shields.io/badge/SRR%20COMPLEMENT%20on%20LRLspoof-23.46%25-blue)](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=ssl-mamba) ## Citation ```bibtex @inproceedings{tran25b_interspeech, title = {{Leveraging SSL Speech Features and Mamba for Enhanced DeepFake Detection}}, author = {Tran, Hoan My and Lolive, Damien and Guennec, David and Sini, Aghilas and Delhay, Arnaud and Marteau, Pierre-Fran{\c{c}}ois}, booktitle = {Interspeech 2025}, pages = {5323--5327}, year = {2025}, doi = {10.21437/Interspeech.2025-1703} } ```