List your company
ResearchFormerly WizWand

Audio Deepfake Detection on ITW In-the-Wild

4.78EERTran et al.
3.3921212.7603122.128531.49669Sep 25, 2025Nov 10, 2025Dec 26, 2025Feb 10, 2026Mar 28, 2026May 13, 2026Jun 29, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Tran et al.Backbone=XLS-R, Backbone status=Fine-tuned, Layers used=25 (gated), Trainable params=318M§, Fusion=Gated2026.064.78
Probing-Guided Layer SelectionBackbone=XLS-R, Backbone status=Frozen, Layers used=4, Trainable params=1.34M, Fusion=Concat2026.064.94
XLSR-Conformer-TCMLoss Function=QAMO2025.095.21
XLSR-AASISTTraining Strategy=DPDA training + PCGrad2025.095.42
XLSR-AASISTTraining Strategy=DPDA training2025.096.2
XLSR-AASIST-SAM2025.096.34
XLSR-MambaTraining Strategy=DPDA training + PCGrad2025.096.43
XLSR-Conformer-TCMTraining Strategy=DPDA training + PCGrad2025.096.48
XLSR-Conformer-NACLLoss Function=WCE2025.096.6
XLSR-Nes2NetXLoss Function=WCE, experimental_setting=Original2025.096.6
XLSR-Nes2Net-X2025.096.6
W2V2-layer7Layer count=72026.066.6
XLSR-MambaLoss Function=WCE2025.096.7
XLSR-MambaTraining Strategy=Baseline2025.096.7
XLSR-Conformer-TCMLoss Function=OC-Softmax2025.096.72
MLDG-LoRABackbone=W2V 2.0, Backbone status=LoRA, Layers used=All, Trainable params=3.59M, Fusion=Feature2026.066.81
Wav2DF-TSL2025.096.83
Xiao & VuBackbone=XLS-R, Backbone status=Frozen, Layers used=25, Trainable params=~25×cls, Fusion=Decision (sum)2026.066.9
XLSR-Conformer-TCMLoss Function=WCE, experimental_setting=Reproduced2025.097.13
XLSR-SLS2025.097.46
XLSR-MambaTraining Strategy=DPDA training2025.097.62
W2V2-layer24Layer count=242026.067.7
W2V2-TCMParams=319M, Training Protocol=Protocol 1 (ASV19 train)2026.037.79
XLSR-Conformer-TCMLoss Function=WCE, experimental_setting=Original2025.097.79
XLSR-Conformer-TCMTraining Strategy=Baseline2025.097.79
XLSR-Conformer-TCMTraining Strategy=DPDA training2025.097.97
XLSR-Nes2NetXLoss Function=OC-Softmax2025.098.23
XLSR-ConformerLoss Function=WCE2025.098.42
XLSR-Nes2NetXLoss Function=QAMO2025.098.83
HuBERT-BaseParams=100M, Training Protocol=Protocol 1 (ASV19 train)2026.039.06
RACL Diffusionreconstruction_method=SemantiCodec, aggregation=multi-layer, regularization=RACL2026.049.155
XLSR-MoE2025.099.17
WavLM-Base+Params=100M, Training Protocol=Protocol 1 (ASV19 train)2026.039.27
XLSR-Nes2NetXLoss Function=WCE, experimental_setting=Reproduced2025.099.76
mHuBERT-FinalParams=100M, Training Protocol=Protocol 1 (ASV19 train)2026.0310.08
XLSR-AASISTTraining Strategy=Baseline2025.0910.46
Agg Diffusionreconstruction_method=SemantiCodec, aggregation=multi-layer2026.0410.679
WavLM-BaseParams=100M, Training Protocol=Protocol 1 (ASV19 train)2026.0311.12
W2V2-AASISTParams=317M, Training Protocol=Protocol 1 (ASV19 train)2026.0311.19
AASIST2026.0611.2
mHuBERT-Iter2Params=100M, Training Protocol=Protocol 1 (ASV19 train)2026.0312.24
mHuBERT-Iter1Params=100M, Training Protocol=Protocol 1 (ASV19 train)2026.0313.58
Baselineimplementation=authors' implementation2026.0417.949
Diffusionreconstruction_method=SemantiCodec2026.0418.159
Encodecreconstruction_method=Encodec2026.0422.964
Baseline*source=CodecFake paper [18]2026.0423.713
HiFi-GANreconstruction_method=HiFi-GAN2026.0423.779
RawNet32026.0629.7
RawNet22026.0633.7
RawGAT2026.0633.8
DACreconstruction_method=DAC2026.0439.477