Datasets:
Add SSL Feature Gating submission for PyAra (EER=6.152346035336791)
#31
by korallll - opened
submissions/ssl-feature-gating.yaml
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
schema_version: 4
|
| 2 |
+
system:
|
| 3 |
+
name: SSL Feature Gating
|
| 4 |
+
slug: ssl-feature-gating
|
| 5 |
+
description: 'Multi-level SSL Feature Gating for audio deepfake detection: a gating mechanism over XLS-R (wav2vec2-xls-r-300m)
|
| 6 |
+
hidden states as front-end, Multi-kernel gated Convolution (MultiConv) back-end with CKA-enforced feature diversity, and
|
| 7 |
+
multi-head attention pooling. Raw-waveform, FP32, deterministic first-64600-sample (4 s) window. Higher score = more bona
|
| 8 |
+
fide.
|
| 9 |
+
|
| 10 |
+
'
|
| 11 |
+
code: https://github.com/hoanmyTran/dissimilarity_deepfake_detection
|
| 12 |
+
checkpoint: https://huggingface.co/SpeechAntiSpoofingBenchmarks/SSL-Feature-Gating/blob/main/acmmm25_best_model.ckpt
|
| 13 |
+
params_millions: 318.08
|
| 14 |
+
paper:
|
| 15 |
+
arxiv_id: 10.1145/3746027.3754568
|
| 16 |
+
url: https://dl.acm.org/doi/10.1145/3746027.3754568
|
| 17 |
+
bibtex: "@inproceedings{tran2025ssl,\n title = {Multi-level SSL Feature Gating for Audio Deepfake Detection},\n \
|
| 18 |
+
\ author = {Tran, Hoan My and Lolive, Damien and Sini, Aghilas and Delhay, Arnaud and Marteau, Pierre-Fran\\c{c}ois\
|
| 19 |
+
\ and Guennec, David},\n booktitle = {Proceedings of the 33rd ACM International Conference on Multimedia (MM '25)},\n\
|
| 20 |
+
\ year = {2025},\n doi = {10.1145/3746027.3754568}\n}\n"
|
| 21 |
+
dataset:
|
| 22 |
+
id: SpeechAntiSpoofingBenchmarks/PyAra
|
| 23 |
+
revision: 40167badd6d8a289f02054761ee5411f49227b87
|
| 24 |
+
split: test
|
| 25 |
+
scores:
|
| 26 |
+
eer_percent: 6.152346035336791
|
| 27 |
+
n_trials: 201778
|
| 28 |
+
n_skipped: 0
|
| 29 |
+
artifact:
|
| 30 |
+
scores_url: https://huggingface.co/SpeechAntiSpoofingBenchmarks/SSL-Feature-Gating/resolve/aca38bfe8c40309fdcb1c4ef84a5b7d0b8bb7de6/.eval_results/SpeechAntiSpoofingBenchmarks/PyAra/scores.txt
|
| 31 |
+
scores_sha256: 8784f0c7cdf3eed185ae333e2c392f8473f89372a86cec1b0db004a6acdd9fcf
|
| 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: ACM MM 2025 checkpoint acmmm25_best_model.ckpt. XLS-R-300M front-end + MultiConv gating. Deterministic first-64600-sample
|
| 43 |
+
window.
|