Pellav2: Speech Anti-Spoofing Arena wrapper

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

No paper: the model sits in the Arena's unpublished tier (shown, not ranked).

Pellav2 (Sadanie, August 2026): an open-weights audio deepfake detector. A w2v-BERT 2.0 conformer encoder (24 layers, fine-tuned) on 80-bin kaldi log-mel fbank features (x32768, per-utterance mean/variance normalised, frame pairs stacked to 160-d, trailing digital silence trimmed), a softmax-weighted mix of all 25 hidden states, masked mean pooling and Linear(1024,1) -> logit(fake). Authors' released checkpoint; training data undisclosed. Input as the authors' pellav2_infer.py: 16 kHz, mean removal, 20 Hz high-pass (zero-phase 2nd-order Butterworth), centre 4 s (64000-sample) crop, ITU-R BS.1770 loudness normalisation to -23 LUFS (pyloudnorm semantics), shorter clips zero-padded. score = -logit(fake), higher = bona fide. FP32 (TF32 off).

Arena results

arena tier arena rank

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% 1.1 on ASVspoof2019_LA EER% 3.17 on ASVspoof2021_LA EER% 1.18 on ASVspoof2021_DF EER% 17.64 on ASVspoof5 EER% 1.57 on InTheWild EER% 1.13 on ADD22_eval_31 EER% 3.61 on ADD2023_track12_test_r1 EER% 7.3 on CFAD EER% 0 on CD-ADD EER% 3.76 on CVoiceFake_small EER% 2.58 on DeepVoice EER% 0.49 on LibriSeVoc EER% 0.44 on SONAR EER% 0.62 on ArAD EER% 1.36 on DECRO EER% 0 on J-SPAW_LA EER% 1.33 on ODSS EER% 0.02 on HABLA EER% 0.03 on DFADD EER% 1.91 on PyAra EER% 0.2 on XMAD EER% 0.17 on EmoFake_test SRR COMPLEMENT 0.13 on EmoSpoofTTS SRR COMPLEMENT 4.09 on LRLspoof

Downloads last month
218
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support