List your company

Audio Deepfake Detection on In-the-wild

0.8EERXLS-R-1B
-1.79215.70433.250.696Jul 26, 2024Nov 17, 2024Mar 12, 2025Jul 5, 2025Oct 28, 2025Feb 20, 2026Jun 15, 2026

Evaluation results

MethodLinks
XLS-R-1BTraining Strategy=DOSS-Weight, #Params=965M, #Hours=12k2025.120.8———
XLS-R-300MTraining Strategy=DOSS-Weight, #Params=317M, #Hours=12k2025.120.81———
SOTATraining Strategy=Existing Benchmarks2025.121.23———
XLS-R-2BTraining Strategy=Naive Aggregation, #Params=2.2B, #Hours=74k2025.121.23———
XLS-R-1BTraining Strategy=Naive Aggregation, #Params=965M, #Hours=74k2025.121.37———
MMS-1BTraining Strategy=Naive Aggregation, #Params=965M, #Hours=74k2025.121.82———
AFSSVenue=Ours2026.032.7———
MMS-300MTraining Strategy=Naive Aggregation, #Params=317M, #Hours=74k2025.122.9———
SCLVenue=Interspeech ’242026.034.52———
BreathNetPublication=Ours2026.024.7———
MamBo-3-Hydra-N3Params (M)=319.372026.014.97———
MamBo-4-Hydra-N1Params (M)=318.022026.015.17———
Nes2Net-XPublication=TIFS 20252026.025.52———
MamBo-1-Mamba2-N2Params (M)=317.482026.015.57———
HyperPotterParams (M)=317.872026.025.72———
Fake-Mamba (L)Params (M)=319.722026.015.85———
LSR+LSAPublication=ICASSP 20252026.025.92———
SSLVenue=ICASSP ’242026.036.1———
MamBo-2-Hydra-N1Params (M)=316.682026.016.24———
WaveSpecPublication=ICASSP 20252026.026.58———
XLSR+MambaParams (M)=3192026.026.7———
XLSR-MambaParams (M)=319.332026.016.71———
XLSR-MambaPublication=SPL 20252026.026.71———
XLS-R + HierConBackbone=XLS-R 300M, Head=HierCon2026.026.87———
Wav2Vec+LogRegBackbone=Wav2Vec, Head=LogReg2026.027.2———
XLSR+SLSParams (M)=3402026.027.45———
XLSR-SLSParams (M)=341.492026.017.46———
SLSPublication=ACMMM 20242026.027.46———
Wav2Vec2+AASIST(Baseline)Params (M)=317.842026.027.58———
OCKDPublication=ICASSP 20242026.027.68———
XLSR+Conformer+TCMParams (M)=3192026.027.79———
XLSR-Conformer+TCMParams (M)=319.772026.017.79———
OCKD2026.027.86———
XLSR+BiCrossMambaParams (M)=318.212026.027.94———
ConformerVenue=Interspeech ’232026.038.52———
XLS-R + SLSBackbone=XLS-R, Head=SLS2026.028.87———
Wav2Vec2+AASISTParams (M)=317.842026.0211.19———
MoEPublication=ICASSP 20252026.0212.48———
SLIMCategory=Finetuned frontend, #Param (million)=2532024.0712.589.8——
SLIMPublication=NeurIPS 20242026.0212.5———
FullCategory=SLIM variants (ours), #Param (million)=112024.0712.989.5——
Wav2Vec+LinearBackbone=Wav2Vec, Head=Linear2026.0216.17———
SFATNet-4Data Condition=Clean (no codecs)2026.0117.29—89.17—
W2V-AASISTCategory=Finetuned frontend, #Param (million)=3172024.0717.584.7——
W2V-ASPCategory=Finetuned frontend, #Param (million)=3172024.0718.683.6——
SFAT-Net 3Data Condition=Clean (no codecs)2026.0119.7—85.2—
Dual-Granularity Orthogonal DisentanglementFine-tuning status=without fine-tuning, Variant=Full Model2026.0621.58———
WavLM-MLPPre-training (SSL-based)=true, Model Parameters=>300M, Fine-tuning status=without fine-tuning2026.0621.85———
Dual-Granularity Orthogonal Disentanglement (Cosine Only)Fine-tuning status=without fine-tuning, Variant=Cosine Only2026.0622.16———
DG-AggFine-tuning status=without fine-tuning2026.0622.73———
Wav2Vec2-AASISTPre-training (SSL-based)=true, Model Parameters=>300M, Fine-tuning status=without fine-tuning2026.0623.17———
Dual-Granularity Orthogonal Disentanglement (No Disentangle)Fine-tuning status=without fine-tuning, Variant=No Disentangle2026.0623.48———
GRL BaselineFine-tuning status=without fine-tuning2026.0624.18———
SSL-FusionCategory=Frozen frontend, #Param (million)=102024.0724.276.5——
W2V-LCNNCategory=Frozen frontend, #Param (million)=N/A2024.0724.5———
W2V-LLGFCategory=Frozen frontend, #Param (million)=102024.0725.175.6——
Enc_style+lingCategory=SLIM variants (ours), #Param (million)=102024.0725.175.7——
WLM-ASPCategory=Frozen frontend, #Param (million)=92024.0725.475.1——
DependencyCategory=SLIM variants (ours), #Param (million)=92024.0725.875——
RawGAT-STFine-tuning status=without fine-tuning2026.0625.83———
Dual-Granularity Orthogonal Disentanglement (Single Branch)Fine-tuning status=without fine-tuning, Variant=Single Branch2026.0625.86———
Res-TSSDNetFine-tuning status=without fine-tuning2026.0626.14———
AASISTFine-tuning status=without fine-tuning2026.0627.41———
W2V+HUBCategory=Frozen frontend, #Param (million)=92024.0727.973.7——
WLM+HUBCategory=Frozen frontend, #Param (million)=92024.0729.272.4——
Enc_styCategory=SLIM variants (ours), #Param (million)=92024.0729.272.4——
PS3DTCategory=Baseline, #Param (million)=N/A2024.0729.7———
HUB-ASPCategory=Frozen frontend, #Param (million)=92024.0729.971.8——
W2V-ASPCategory=Frozen frontend, #Param (million)=92024.0730.270.5——
W2V+WLMCategory=Frozen frontend, #Param (million)=92024.0730.370.4——
Enc_lingCategory=SLIM variants (ours), #Param (million)=92024.0730.470.8——
AASIST-LVenue=ICASSP ’222026.0332.61———
RawNet2Category=Baseline, #Param (million)=42024.0737.860.2——
AASIST-BVenue=ICASSP ’222026.0339.45———
CAFNetZero-shot=true, Trained on=MLADDC T22026.0545.9—56.2253.62
LCNNCategory=Baseline, #Param (million)=42024.0765.637.3——
AASIST-BVenue=ICASSP ’222026.03——65.31—
AASIST-LVenue=ICASSP ’222026.03——73.72—
AFSS2026.03——99.64—
ClovaAI2026.03——74—
ConformerVenue=Interspeech ’232026.03——97.12—
ECAPA-TDNN2026.03——79—
FlowFakeTrain=ASV192026.06———61.71
FlowFakeTrain=FoR2026.06———70.91
FlowFakeTrain=MLAAD v12026.06———62.39
H/ASP2026.03——85—
HyperPotterParams (M)=317.872026.02—95.4——
POI-Forensics2026.03——91—
RawGAT-STTrain=ASV192026.06———50
RawGAT-STTrain=FoR2026.06———49.8
RawGAT-STTrain=MLAAD v12026.06———47.7
SCLVenue=Interspeech ’242026.03——99.1—
SSL W2V2Train=ASV192026.06———79.7
SSL W2V2Train=FoR2026.06———57.8
SSL W2V2Train=MLAAD v12026.06———68
TitaNet2026.03——91—
Wav2Vec2+AASISTParams (M)=317.842026.02—90.9——
Wav2Vec2+AASIST (Baseline)Params (M)=317.842026.02—93.9——
Whisper DFTrain=ASV192026.06———76.5
Whisper DFTrain=FoR2026.06———54.1