Add Forensics-XLSR-Wild submission for LRLspoof (1-SRR=8.1384)

submissions/forensics-xlsr-wild.yaml

@@ -0,0 +1,48 @@
+schema_version: 4
+system:
+  name: Forensics-XLSR-Wild
+  slug: forensics-xlsr-wild
+  description: 'Forensics 0.3B XLS-R "wild" speech deepfake detector (Eliya Segev, 2026): a fully fine-tuned
+    wav2vec 2.0 XLS-R 300M front-end whose 24 transformer hidden states are pooled by an AASIST-style
+    spectro-temporal graph-attention pool (layer x time graphs, GAT dims 64/32), followed by a 3-block
+    residual MLP classifier (160-d) and a single logit (label 1 = real). Authors'' checkpoint_epoch_4
+    as released. Input as the authors'' inference.py: peak-normalised waveform, 5 s (80000 samples) centre
+    crop, shorter clips repeat-padded. score = the logit (sigmoid = P(real)). Unpublished model (HF model
+    card only, no paper).'
+  code: https://huggingface.co/eliya/forensics_0.3B_xlsr_wild_deepfake_classifier/tree/0cbc87b3fbd683ff8c8c961bcf244a8d5f1bcf02
+  checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/Forensics-XLSR-Wild/blob/76bb2759e1615f5f83300d4fe2ea16d81ecea686/checkpoint_epoch_4.safetensors
+  params_millions: 315.926375
+dataset:
+  id: lab260/LRLspoof
+  revision: ac76f0e5472fdb68a6726e1fa94a148316e76916
+  split: test
+scores:
+  srr_complement: 8.138439113335771
+  n_trials: 1304455
+  n_skipped: 286
+calibration:
+  source_dataset: SpeechAntiSpoofingBenchmarks/DeepVoice
+  threshold: 1.158977990960452
+artifact:
+  scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/Forensics-XLSR-Wild/resolve/ba404b502f00268e03033806d68659df976627fe/.eval_results/lab260/LRLspoof/scores.txt
+  scores_sha256: 48433bb9c422a6ceb0e81607ccc8b61f2317e8d4c91ae6f3df5c7bf48f7e791e
+  bench_version: speech-spoof-bench==0.4.1
+reproduction: {}
+submitter:
+  hf_username: korallll
+  contact: k.n.borodin@mtuci.ru
+submitted_at: '2026-11-04'
+notes: 'Forensics-XLSR-Wild = HF eliya/forensics_0.3B_xlsr_wild_deepfake_classifier @ 0cbc87b3fbd683ff8c8c961bcf244a8d5f1bcf02
+  (formerly eliya/forensics_0.3B_xlsr_wild), checkpoint_epoch_4.safetensors (sha256 39e37fa5...dc21),
+  CC-BY-NC-4.0, re-hosted unchanged. Architecture = the repo''s model.py; the XLS-R 300M backbone is built
+  from the facebook/wav2vec2-xls-r-300m config and every weight is overwritten by the checkpoint (strict
+  load; the 4 training-only projection.* tensors, unused by model.py, are dropped via an asserted allow-list).
+  Preprocessing = inference.py: peak normalisation wav/(max|wav|+1e-8) over the whole clip, then a centred
+  80000-sample (5 s) crop or repeat-padding; no resampling; FP32. Score = the single output logit (higher
+  = bona fide); verified bit-identical to the authors'' model.py + inference.py on sample clips. Training
+  data: the current model card says 5M+ samples from a mix of proprietary (undisclosed) and public data;
+  the first card revision (b2a5756930) listed SpeechFake, MD-CommonVoice, DFADD, CodecFake, ASVspoof2021-DF,
+  ASVspoof2019-LA (later ones add CommonVoice, Aishell); splits are not stated. The weights were uploaded
+  in commits titled "Upload itw_winner_xlsr", i.e. the checkpoint was apparently selected on In-the-Wild,
+  so InTheWild, DFADD, ASVspoof2021_DF and ASVspoof2019_LA are marked trained_on (n/a*). Self-reported
+  In-the-Wild EER 1.03%. No paper -> Unpublished tier.'