Add ProSDD-ASV19 submission for DECRO (EER=26.1450)

submissions/prosdd-asv19.yaml

@@ -0,0 +1,55 @@
+schema_version: 4
+system:
+  name: ProSDD-ASV19
+  slug: prosdd-asv19
+  description: 'ProSDD (Mahapatra et al., 2026): a fully fine-tuned wav2vec 2.0 XLS-R 300M encoder, first
+    trained (Stage I, bona fide LibriSpeech train-clean-100 + dev only) to predict masked frame-level
+    speaker + prosody targets, then (Stage II) jointly fine-tuned with that objective and a binary real/fake
+    head on ASVspoof 2019 LA train/dev (RawBoost series 3). Inference uses only the clean path: XLS-R
+    CNN front-end + feature projection, latent sequence fixed at 200 frames (zero-padded), 24-layer transformer
+    encoder, mean pool, Linear(1024,512)-ReLU-Linear(512,2). Input as the authors'' data_utils_eval.py:
+    16 kHz, no amplitude normalisation, fixed 4 s (64000 samples) centre crop, shorter clips centre zero-padded.
+    FP32; score = logits[:,1] (bona fide class). Authors'' released checkpoint; no upstream license, so
+    the weights are linked from the authors'' Google Drive, not re-hosted.'
+  code: https://github.com/ProSDD/codes/tree/bdb80440cd946368e77f4b51f6ddf26c7443dc33
+  checkpoint: https://drive.google.com/file/d/1RHY-VELyi4XH-L1egeRcgWjHdkdWV2bO/view
+  params_millions: 315.963522
+  paper:
+    arxiv_id: '2604.13229'
+    url: https://arxiv.org/abs/2604.13229
+    bibtex: "@article{mahapatra2026prosdd,\n  title   = {ProSDD: Learning Prosodic Representations for\
+      \ Speech Deepfake Detection against Expressive and Emotional Attacks},\n  author  = {Mahapatra,\
+      \ Aurosweta and Ulgen, Ismail Rasim and Lee, Kong Aik and Andrews, Nicholas and Sisman, Berrak},\n\
+      \  journal = {arXiv preprint arXiv:2604.13229},\n  year    = {2026}\n}\n"
+dataset:
+  id: SpeechAntiSpoofingBenchmarks/DECRO
+  revision: e54c29df5b81bac75e5cc294d1ab0d4db72a5c71
+  split: test
+scores:
+  eer_percent: 26.14498319731157
+  n_trials: 37314
+  n_skipped: 0
+artifact:
+  scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/ProSDD-ASV19/resolve/1538df9e38c0c0212e58387f56fde82d331334a7/.eval_results/SpeechAntiSpoofingBenchmarks/DECRO/scores.txt
+  scores_sha256: 9f04a6280c8dd5e60db282fbfb5d5238fdf8b5c2ea9183538ed6eecad33f6f37
+  bench_version: speech-spoof-bench==0.4.1
+reproduction: {}
+submitter:
+  hf_username: korallll
+  contact: k.n.borodin@mtuci.ru
+submitted_at: '2026-11-07'
+notes: 'ProSDD-ASV19 = github.com/ProSDD/codes @ bdb80440cd946368e77f4b51f6ddf26c7443dc33 (model_stage2realfake.py
+  ProSDDStage2, main_eval.py inference_forward, data_utils_eval.py pad), checkpoint ProSDD_Trained2019
+  from the authors'' Google Drive folder 1h250-Um5qWo-rpeOE6K_Gdeygy46k3xg (file id 1RHY-VELyi4XH-L1egeRcgWjHdkdWV2bO,
+  sha256 d2cb857cf7b14a9caa200cde14100062e72fe2a39865bc435f83f5be74da1c8a; a plain PyTorch state_dict,
+  loaded with weights_only=True). The repo has no LICENSE, so the weights are linked, not re-hosted. The
+  XLS-R 300M backbone is built from the facebook/wav2vec2-xls-r-300m config (HF @ 1a640f3) and every tensor
+  is overwritten by the checkpoint (strict load; the 5 training-only tensors mask_embed, pros_ln.*, final_proj.*
+  (Stage-I 1024->448 speaker+prosody head) are dropped via an asserted allow-list; the source''s own strict=False
+  load reports 0 missing / 0 unexpected). classifier_pool = mean (the checkpoint has no attention-pool
+  tensors). Preprocessing exactly as data_utils_eval.py / main_eval.py: 64000-sample centre crop / centred
+  zero-pad, no normalisation, latent frames 199 -> zero-padded to 200, encoder without attention mask,
+  mean over all 200 frames. Score = logits[:,1]; verified against the authors'' modules (max |d| 1e-5).
+  Training: Stage I LibriSpeech train-clean-100 + dev (bona fide), Stage II ASVspoof 2019 LA train/dev
+  with RawBoost 3. The Arena ASVspoof sets are eval partitions, so no trained_on entries. Self-reported
+  EER% (paper): 19LA 0.42 / 21LA 3.87 / ASVspoof 2024 (ASVspoof5) 16.14 / EmoFake 3.70.'