Add XLSR-Conformer-TCM-DF submission for LRLspoof (1-SRR=24.3559)

submissions/xlsr-conformer-tcm-df.yaml

@@ -0,0 +1,60 @@
+schema_version: 4
+system:
+  name: XLSR-Conformer-TCM-DF
+  slug: xlsr-conformer-tcm-df
+  description: 'XLSR-Conformer + TCM (Truong et al., Interspeech 2024), DF-track model: a fully fine-tuned
+    wav2vec 2.0 XLS-R 300M front-end projected to 144-d (Linear + BatchNorm + SELU), followed by 4 Conformer
+    blocks whose multi-head self-attention carries the Temporal-Channel Modeling (TCM) module (per-head
+    channel/head tokens, classification-token enrichment with the mean temporal and mean head tokens;
+    4 heads, depthwise conv kernel 31) and a class token feeding a linear 2-way head. Trained on ASVspoof2019
+    LA train with RawBoost (the paper''s DF-track setting is stationary signal-independent additive noise).
+    Checkpoint = the authors'' released fixed-length DF model avg_5_best.pth (their own main.py average
+    of the 5 best-validation checkpoints), unchanged. FP32, deterministic first-66800-sample window (shorter
+    clips repeat-tiled).'
+  code: https://github.com/ductuantruong/tcm_add
+  checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/XLSR-Conformer-TCM-DF/blob/6638adede36355501770785711f0db30f8189d9f/model.safetensors
+  params_millions: 321.2132
+  paper:
+    arxiv_id: 10.21437/Interspeech.2024-659
+    url: https://www.isca-archive.org/interspeech_2024/truong24b_interspeech.html
+    bibtex: "@inproceedings{truong24b_interspeech,\n  title     = {{Temporal-Channel Modeling in Multi-head\
+      \ Self-Attention for Synthetic Speech Detection}},\n  author    = {Truong, Duc-Tuan and Tao, Ruijie\
+      \ and Nguyen, Tuan and Luong, Hieu-Thi and Lee, Kong Aik and Chng, Eng Siong},\n  booktitle = {Interspeech\
+      \ 2024},\n  pages     = {537--541},\n  year      = {2024},\n  doi       = {10.21437/Interspeech.2024-659}\n\
+      }\n"
+dataset:
+  id: lab260/LRLspoof
+  revision: ac76f0e5472fdb68a6726e1fa94a148316e76916
+  split: test
+scores:
+  srr_complement: 24.355892526198673
+  n_trials: 1304455
+  n_skipped: 286
+calibration:
+  source_dataset: SpeechAntiSpoofingBenchmarks/DeepVoice
+  threshold: -2.0625798694267514
+artifact:
+  scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/XLSR-Conformer-TCM-DF/resolve/be1601cbbf65364a9a9e9914b16cde13e610fdfd/.eval_results/lab260/LRLspoof/scores.txt
+  scores_sha256: f116840abd67bbdd446fa48a5b5c35a529e27f596b6edc78813d63206bf6e1f6
+  bench_version: speech-spoof-bench==0.4.1
+reproduction:
+  reproduced_by: SpeechAntiSpoofingBenchmarks
+  reproduced_at: '2026-11-04'
+  reproduced_bench_version: speech-spoof-bench==0.4.1
+  match: inference
+submitter:
+  hf_username: korallll
+  contact: k.n.borodin@mtuci.ru
+submitted_at: '2026-11-04'
+notes: 'XLSR-Conformer + TCM (Truong et al., Interspeech 2024, doi 10.21437/Interspeech.2024-659); code
+  github ductuantruong/tcm_add @ 726387eb731223a21d551476349f8f8f680a3375 (MIT); weights from the authors''
+  NTU SharePoint (linked in the tcm_add README), model_checkpoints/interspeech24_tcm/DF/fixed_length/avg_5_best.pth
+  (1,285,107,735 B), the authors'' own main.py 5-checkpoint average, re-saved unchanged as model.safetensors
+  (convert.py). Strict load, 321,213,220 parameters (incl. the frozen 10000x144 sinusoidal positional
+  table, 1.44 M; the paper''s 319.77 M excludes it). Deterministic first-66800-sample window (4.175 s,
+  source data_utils self.cut; the paper text says 64,600), shorter clips repeat-tiled (source utils.pad);
+  no resampling or normalisation; score = logits[:, 1] (bona fide = class 1; source produce_evaluation_file);
+  higher = more bona fide. FP32 with cuDNN TF32 off (TF32 convs made scores batch-dependent by ~2e-4).
+  Deviations: XLS-R built from its config (the checkpoint holds the fine-tuned front-end); the source
+  extract_feat train() call removed. Same network as DPDA-GA-TCM. Paper EER (fixed-length input): 21DF
+  2.06% (variable-length 2.25%). Trained on ASVspoof2019 LA train (dev for model selection) only.'