Pellav2: Speech Anti-Spoofing Arena wrapper
Packaged for the Speech Anti-Spoofing Arena from the authors' code.
No paper: the model sits in the Arena's unpublished tier (shown, not ranked).
Pellav2 (Sadanie, August 2026): an open-weights audio deepfake detector. A w2v-BERT 2.0 conformer encoder (24 layers, fine-tuned) on 80-bin kaldi log-mel fbank features (x32768, per-utterance mean/variance normalised, frame pairs stacked to 160-d, trailing digital silence trimmed), a softmax-weighted mix of all 25 hidden states, masked mean pooling and Linear(1024,1) -> logit(fake). Authors' released checkpoint; training data undisclosed. Input as the authors' pellav2_infer.py: 16 kHz, mean removal, 20 Hz high-pass (zero-phase 2nd-order Butterworth), centre 4 s (64000-sample) crop, ITU-R BS.1770 loudness normalisation to -23 LUFS (pyloudnorm semantics), shorter clips zero-padded. score = -logit(fake), higher = bona fide. FP32 (TF32 off).
- Weights: re-hosted here under the upstream license (pellav2_detector.safetensors, commit-pinned); upstream: https://huggingface.co/Sadanie/pellav2-audio-deepfake-detector/blob/b668eb4dfe4719665692c73c1d70383680afff6b/pellav2_infer.py
- Params: 580.49 M. Trained on undisclosed data (no paper; the card is silent), so no Arena cell is marked n/a*, but leakage cannot be ruled out (J-SPAW_LA scores are perfectly separated).
- Score: higher = bona fide. J-SPAW_LA was scored with the PyTorch wrapper; most other datasets with the TensorRT drop-in class (FP32 on the RTX 4070 Ti SUPER; FP32 on the RTX 5060 Ti), which matched PyTorch on the full J-SPAW_LA set (|dEER| <= 0.25, Spearman >= 0.999); the RTX 4060 Ti cards ran PyTorch.
Arena results
EER% (lower is better); EmoSpoofTTS and LRLspoof are spoof-only and use 1-SRR% at the model's
DeepVoice EER threshold. Per-dataset scores are under .eval_results/.
- Downloads last month
- 218