Add AMSDF submission for DECRO (EER=14.989404810587754)

#19
by korallll - opened
Files changed (1) hide show
  1. submissions/amsdf.yaml +43 -0
submissions/amsdf.yaml ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 4
2
+ system:
3
+ name: AMSDF
4
+ slug: amsdf
5
+ description: 'AMSDF (Audio Multi-view Spoofing Detection Framework): captures intra- and inter-view spoofing cues from audio-emotion-text
6
+ correlations. XLS-R (wav2vec2-xls-r-300m) text view + ACRNN emotion view (logfbank+delta+delta2) + raw-audio graph view,
7
+ fused by heterogeneous graph attention. Raw-waveform (+ derived emotion spectrogram), FP32, first-64600-sample (4 s) window.
8
+ Checkpoint M1 (best-EER seed). Higher score = more bona fide.
9
+
10
+ '
11
+ code: https://github.com/ItzJuny/AMSDF
12
+ checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/AMSDF/blob/main/M1.pth
13
+ params_millions: 325.4
14
+ paper:
15
+ arxiv_id: 10.1109/TIFS.2024.3431888
16
+ url: https://ieeexplore.ieee.org/document/10605761
17
+ bibtex: "@article{wu2024audio,\n author = {Wu, Junyan and Yin, Qilin and Sheng, Ziqi and Lu, Wei and Huang, Jiwu and\
18
+ \ Li, Bin},\n journal = {IEEE Transactions on Information Forensics and Security},\n title = {Audio Multi-view Spoofing\
19
+ \ Detection Framework Based on Audio-Text-Emotion Correlations},\n year = {2024},\n volume = {19},\n pages \
20
+ \ = {7133--7146},\n doi = {10.1109/TIFS.2024.3431888}\n}\n"
21
+ dataset:
22
+ id: SpeechAntiSpoofingBenchmarks/DECRO
23
+ revision: e54c29df5b81bac75e5cc294d1ab0d4db72a5c71
24
+ split: test
25
+ scores:
26
+ eer_percent: 14.989404810587754
27
+ n_trials: 37314
28
+ n_skipped: 0
29
+ artifact:
30
+ scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/AMSDF/resolve/3eede3c3aec8c4d32534ad6ab91517847f52b0da/.eval_results/SpeechAntiSpoofingBenchmarks/DECRO/scores.txt
31
+ scores_sha256: f80eba59456090022e17abf77198da42a152d450ececb743237b42aef1d1f251
32
+ bench_version: speech-spoof-bench==0.4.1
33
+ reproduction:
34
+ reproduced_by: SpeechAntiSpoofingBenchmarks
35
+ reproduced_at: '2026-06-26'
36
+ reproduced_bench_version: speech-spoof-bench==0.4.1
37
+ match: scoring
38
+ submitter:
39
+ hf_username: korallll
40
+ contact: k.n.borodin@mtuci.ru
41
+ submitted_at: '2026-06-26'
42
+ notes: AMSDF M1 checkpoint (best-EER seed). Multi-view XLS-R + ACRNN emotion + graph fusion. Score = P(real)-P(fake). First-64600-sample
43
+ window.