Add AASIST submission

#6
Files changed (1) hide show
  1. submissions/aasist.yaml +43 -0
submissions/aasist.yaml ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ schema_version: 4
2
+ system:
3
+ name: AASIST
4
+ slug: aasist
5
+ description: 'AASIST: audio anti-spoofing using integrated spectro-temporal graph attention networks.
6
+ Sinc-convolution front-end, RawNet2-style residual encoder, and heterogeneous stacking graph attention
7
+ over spectral and temporal sub-graphs with a learnable readout. Official clovaai/aasist ASVspoof2019
8
+ LA pretrained checkpoint, FP32, deterministic first-64600-sample window (no random crop).
9
+
10
+ '
11
+ code: https://github.com/clovaai/aasist
12
+ checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/AASIST/blob/e842653505c2832ac9f46bbf56173b0f54ef82a7/AASIST.pth
13
+ params_millions: 0.297866
14
+ paper:
15
+ arxiv_id: '2110.01200'
16
+ url: https://arxiv.org/abs/2110.01200
17
+ bibtex: "@inproceedings{jung2022aasist,\n title={{AASIST}: Audio Anti-Spoofing Using Integrated Spectro-Temporal\
18
+ \ Graph Attention Networks},\n author={Jung, Jee-weon and Heo, Hee-Soo and Tak, Hemlata and Shim,\
19
+ \ Hye-jin and Chung, Joon Son and Lee, Bong-Jin and Yu, Ha-Jin and Evans, Nicholas},\n booktitle={ICASSP\
20
+ \ 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},\n\
21
+ \ pages={6367--6371},\n year={2022},\n organization={IEEE}\n}\n"
22
+ dataset:
23
+ id: SpeechAntiSpoofingBenchmarks/CD-ADD
24
+ revision: c2de87d49b268b624e6af7440dc2890703098965
25
+ split: test
26
+ scores:
27
+ eer_percent: 51.05162523900574
28
+ n_trials: 20786
29
+ n_skipped: 0
30
+ artifact:
31
+ scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/AASIST/resolve/e842653505c2832ac9f46bbf56173b0f54ef82a7/.eval_results/SpeechAntiSpoofingBenchmarks/CD-ADD/scores.txt
32
+ scores_sha256: 3fa7f52a3233ef1c04b8a6951ee7c688025aef5ab312ea8599b3ab6ac0cd3cde
33
+ bench_version: speech-spoof-bench==0.3.0
34
+ reproduction:
35
+ reproduced_by: SpeechAntiSpoofingBenchmarks
36
+ reproduced_at: '2026-06-02'
37
+ reproduced_bench_version: speech-spoof-bench==0.3.0
38
+ match: scoring
39
+ submitter:
40
+ hf_username: korallll
41
+ contact: k.n.borodin@mtuci.ru
42
+ submitted_at: '2026-06-02'
43
+ notes: Out-of-domain; the 2019 LA checkpoint does not generalize well here.