Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -13,6 +13,7 @@ tags:
|
|
| 13 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 14 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 15 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
|
|
|
| 16 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 17 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 18 |
|
|
@@ -69,6 +70,7 @@ crop), so the numbers are exactly reproducible from the pinned score file.
|
|
| 69 |
| ASVspoof2019_LA | test | **1.86** | 71,237 | 0 | in-domain (training data) |
|
| 70 |
| CD-ADD | test | **54.55** | 20,786 | 0 | out-of-domain (modern neural-TTS); does not generalize |
|
| 71 |
| InTheWild | test | **55.92** | 31,779 | 0 | out-of-domain (real-world deepfakes); does not generalize |
|
|
|
|
| 72 |
|
| 73 |
The ASVspoof2019_LA result reproduces near the paper's reported 2.25 % on the LA eval
|
| 74 |
set; the deterministic window (vs. the paper's random crop) accounts for the small
|
|
|
|
| 13 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 14 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 15 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 16 |
+
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 17 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 18 |
[](https://huggingface.co/spaces/SpeechAntiSpoofingBenchmarks/SpeechAntiSpoofingArena?system=rescapsguard)
|
| 19 |
|
|
|
|
| 70 |
| ASVspoof2019_LA | test | **1.86** | 71,237 | 0 | in-domain (training data) |
|
| 71 |
| CD-ADD | test | **54.55** | 20,786 | 0 | out-of-domain (modern neural-TTS); does not generalize |
|
| 72 |
| InTheWild | test | **55.92** | 31,779 | 0 | out-of-domain (real-world deepfakes); does not generalize |
|
| 73 |
+
| ASVspoof2021_LA | test | **18.70** | 181,566 | 0 | cross-dataset generalization |
|
| 74 |
|
| 75 |
The ASVspoof2019_LA result reproduces near the paper's reported 2.25 % on the LA eval
|
| 76 |
set; the deterministic window (vs. the paper's random crop) accounts for the small
|