Downloads · 30 days
0
voicefence-rawnet2-v3-hien
Arya12367/voicefence-rawnet2-v3-hien
voicefence-rawnet2-v3-hien is a audio classification model from Arya12367. Use it for the audio classification task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as mit.
RawNet2-style audio anti-spoofing checkpoint for Voicefence.
Access
Public
Updated Jul 12, 2026
Repo size
282 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt282 MB · 100%
At a glance
- Task
- Audio Classification
- Library
- pytorch
- License
- mit
- Access
- Public
- Created
- Jul 11, 2026
- Updated
- Jul 12, 2026
- SHA
- 77cf0027
Model card
From the Hugging Face model README
voicefence-rawnet2-v3-hien
RawNet2-style audio anti-spoofing checkpoint for Voicefence.
Trained on ASVspoof 2019 LA + codec augmentation (opus6k / amr475 / mulaw) plus a synthetic Hindi-English code-switch spoof set (parler-tts spoofs + Common Voice Hindi / LibriSpeech English bonafide). Total train 106,320 utts. Best epoch 36, dev EER 0.47%.
Held-out results (EER)
| Test set | EER |
|---|---|
| Synthetic hi-en held-out (overall) | 0.75% |
| — hi-en | 0.50% |
| — hi | 1.00% |
| — en | 1.19% |
| ASVspoof LA eval (clean) | 5.66% |
| Codec sweep (avg of 6) | 6.06% |
Closes the cross-TTS-engine / code-switch gap of the earlier
voicefence-rawnet2-codecaug checkpoint (~32% held-out hi-en EER) while
preserving codec robustness. best.pt is a torch checkpoint dict with keys
model, config, eer_threshold. Score convention: higher = more bonafide.
- Task
- Audio Classification
- Library
- pytorch
- License
- mit
- Created
- Jul 11, 2026
- Updated
- Jul 12, 2026