Model card with Arena badges; final meta.yaml
Browse files
README.md
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- audio
|
| 5 |
+
- anti-spoofing
|
| 6 |
+
- deepfake-detection
|
| 7 |
+
- speech-anti-spoofing-arena
|
| 8 |
+
library_name: pytorch
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# DeepFense-CodecFake-W2V2-MLP: Speech Anti-Spoofing Arena wrapper
|
| 12 |
+
|
| 13 |
+
Packaged for the [Speech Anti-Spoofing Arena](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena) from the authors' [code](https://github.com/Yaselley/deepfense-framework/tree/3f2fe1146c48ea4bacc480dd96d806b6c72c09cf) and [paper](https://arxiv.org/abs/2604.08450).
|
| 14 |
+
|
| 15 |
+
DeepFense (Kheir et al., 2026) CodecFake recipe: a fully fine-tuned wav2vec 2.0 XLS-R 300M front-end (last transformer layer, features_only) followed by an MLP back-end (Linear 1024->128, LayerNorm, ReLU, mean pooling over time) and a 2-way linear CrossEntropy head. Trained on the Codecfake corpus (Xie et al., AISHELL-3/VCTK-based codec re-synthesis) without data augmentation, seed 42; the released best_model.pth (epoch 2) as is. Input as the DeepFense eval pipeline: raw float waveform (no normalisation), first 64000 samples (4 s), shorter clips repeat-padded. score = logit(bonafide) - logit(spoof); FP32.
|
| 16 |
+
|
| 17 |
+
- Weights: **not re-hosted** (the authors grant no license for them). Get the authors' [checkpoint](https://huggingface.co/DeepFense/CodecFake_Wav2Vec2_MLP_NoAug_Seed42/blob/c08bd2125c83e6c8fdc64566bfbfda18dff4dfb4/best_model.pth); its sha256 is in the `meta.yaml` notes.
|
| 18 |
+
- Params: 317.522306 M. Trained on Codecfake (Xie et al.) train, no augmentation.
|
| 19 |
+
- Score: higher = bona fide. J-SPAW_LA was scored with the PyTorch wrapper; the other datasets with the
|
| 20 |
+
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.
|
| 21 |
+
|
| 22 |
+
## Arena results
|
| 23 |
+
|
| 24 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 25 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 26 |
+
|
| 27 |
+
EER% (lower is better); EmoSpoofTTS and LRLspoof are spoof-only and use 1-SRR% at the model's
|
| 28 |
+
DeepVoice EER threshold. Per-dataset scores are under `.eval_results/`.
|
| 29 |
+
|
| 30 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 31 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 32 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 33 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 34 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 35 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 36 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 37 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 38 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 39 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 40 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 41 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 42 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 43 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 44 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 45 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 46 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 47 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 48 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 49 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 50 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 51 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 52 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 53 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=deepfense-codecfake-w2v2-mlp)
|
| 54 |
+
|
| 55 |
+
## Citation
|
| 56 |
+
|
| 57 |
+
```bibtex
|
| 58 |
+
@article{kheir2026deepfense,
|
| 59 |
+
title={DeepFense: A Unified, Modular, and Extensible Framework for Robust Deepfake Audio Detection},
|
| 60 |
+
author={Kheir, Yassine El and Das, Arnab and Xiao, Yixuan and Wang, Xin and Kallel, Feidi and Erdogan, Enes Erdem and Vu, Ngoc Thang and Polzehl, Tim and Moeller, Sebastian},
|
| 61 |
+
journal={arXiv preprint arXiv:2604.08450},
|
| 62 |
+
year={2026}
|
| 63 |
+
}
|
| 64 |
+
```
|