Add Pellav2 submission for DECRO (EER=1.3569)

submissions/pellav2.yaml

@@ -0,0 +1,50 @@
+schema_version: 4
+system:
+  name: Pellav2
+  slug: pellav2
+  description: '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).'
+  code: https://huggingface.co/Sadanie/pellav2-audio-deepfake-detector/blob/b668eb4dfe4719665692c73c1d70383680afff6b/pellav2_infer.py
+  checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/Pellav2/blob/0de5ef2278e4fb485bd57199a13d15cbf3686954/pellav2_detector.safetensors
+  params_millions: 580.493146
+dataset:
+  id: SpeechAntiSpoofingBenchmarks/DECRO
+  revision: e54c29df5b81bac75e5cc294d1ab0d4db72a5c71
+  split: test
+scores:
+  eer_percent: 1.3569277668314808
+  n_trials: 37314
+  n_skipped: 0
+artifact:
+  scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/Pellav2/resolve/2431837b68eec382d1d1e74c0124be3e7277a9b1/.eval_results/SpeechAntiSpoofingBenchmarks/DECRO/scores.txt
+  scores_sha256: ef7252732ac3b311c0479383e5c527cd79d5a168ca81de8c573bd9f28bdb90cf
+  bench_version: speech-spoof-bench==0.4.1
+reproduction:
+  reproduced_by: SpeechAntiSpoofingBenchmarks
+  reproduced_at: '2026-11-10'
+  reproduced_bench_version: speech-spoof-bench==0.4.1
+  match: inference
+submitter:
+  hf_username: korallll
+  contact: k.n.borodin@mtuci.ru
+submitted_at: '2026-11-10'
+notes: 'Pellav2 = HF Sadanie/pellav2-audio-deepfake-detector @ b668eb4dfe4719665692c73c1d70383680afff6b,
+  pellav2_detector.pt (sha256 d21aedfb52e2f87706c2753715a29665482b0f4a034800c31579df5d458559ac; 776 fp32
+  tensors, loaded with weights_only=True, strict, all used) re-hosted unchanged as pellav2_detector.safetensors
+  (sha256 050234557deabc95d9697a2ede42a5152ed06ba500a4488a42542ab793af428f). MIT per the model card (no
+  LICENSE file). Model class and preprocessing vendored from the repo''s pellav2_infer.py (sha256 bf13cc97...,
+  read, never imported from the Hub cache); the w2v-BERT 2.0 backbone is built from the facebook/w2v-bert-2.0
+  config (@ da985ba0) and every weight comes from the checkpoint. pyloudnorm''s integrated loudness is
+  re-implemented (same filters, gating and op order; not installed here) with the script''s RMS fallback
+  for clips < 0.4 s. Verified against the authors'' Detector + load_clip on 19 J-SPAW_LA-derived clips:
+  identical preprocessing, max |d score| 2.8e-6. The authors'' own window is a centre 4 s crop with zero-padding,
+  so it is kept (no Arena first-64600/repeat-pad deviation). Training data is undisclosed (no paper, card
+  silent) -> no trained_on entries, but leakage cannot be ruled out: J-SPAW_LA EER is 0.0% with saturated,
+  perfectly separated scores. The batch shortlist cites a DFADD EER of 1.9% (not on the pinned card).
+  No paper -> unpublished tier.'