Add hubert-xl-antideepfake-nda LRLspoof submission (srr_complement 7.38%)

submissions/hubert-xl-antideepfake-nda.yaml

@@ -0,0 +1,49 @@
+schema_version: 5
+system:
+  name: HuBERT-XL-AntiDeepfake-NDA
+  slug: hubert-xl-antideepfake-nda
+  description: AntiDeepfake detector (HuBERT extra-large self-supervised backbone)
+    post-trained for deepfake speech detection without data augmentation (NDA variant);
+    head = AdaptiveAvgPool1d + Linear(D,2), score = real logit. Released by nii-yamagishilab
+    (arXiv:2506.21090). Evaluated with a deterministic first-4 s window + batching
+    (NOT the paper's whole-utterance protocol).
+  code: https://github.com/nii-yamagishilab/AntiDeepfake
+  checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/HuBERT-XL-AntiDeepfake-NDA/blob/b6ba9945932dabd6a6ffc5c7dc2c7bbb1aaf5180/model.safetensors
+  params_millions: 964.3242
+  paper:
+    arxiv_id: '2506.21090'
+    url: https://arxiv.org/abs/2506.21090
+    bibtex: "@article{ge2025posttraining,\n  title={Post-training for Deepfake Speech\
+      \ Detection},\n  author={Ge, Wanying and Wang, Xin and Yamagishi, Junichi and\
+      \ others},\n  journal={arXiv preprint arXiv:2506.21090},\n  year={2025}\n}\n"
+dataset:
+  id: lab260/LRLspoof
+  revision: 5caf11349cf96aa198527cfb040cdba585cdd49e
+  split: test
+scores:
+  srr_complement: 7.379105008630016
+  n_trials: 1304455
+  n_skipped: 286
+calibration:
+  source_dataset: SpeechAntiSpoofingBenchmarks/DeepVoice
+  threshold: 0.4249582395480226
+artifact:
+  scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/HuBERT-XL-AntiDeepfake-NDA/resolve/880e8dc33a65f7434ea0118ea0f7fa4d2562aa23/.eval_results/lab260/LRLspoof/scores.txt
+  scores_sha256: cf3ad75031093447d5459295017f9981fd2ab98a0b77b6dce941e75a74d65635
+  bench_version: speech-spoof-bench==0.4.1
+reproduction:
+  reproduced_by: SpeechAntiSpoofingBenchmarks
+  reproduced_at: '2026-06-23'
+  reproduced_bench_version: speech-spoof-bench==0.4.1
+  match: scoring
+submitter:
+  hf_username: korallll
+  contact: k.n.borodin@mtuci.ru
+submitted_at: '2026-06-23'
+notes: HuBERT-XL-AntiDeepfake-NDA on LRLspoof (spoof-only, srr_complement = 1-SRR,
+  lower better). TensorRT engine (parity-verified vs PyTorch). Scored 1,304,455 of
+  1,304,455 TTS utterances; 286 source files are empty/zero-length (skipped), 172
+  corrupt-header WAVs recovered via ffmpeg. Audio resampled per-file to 16 kHz; score
+  = net(x)[:,1] (higher = bona fide). t* = this model's own DeepVoice EER operating-point
+  threshold (DeepVoice EER 16.16%), transferred via the calibration block. Verified
+  with reproduce --scoring.