|
Download implementation-notes.md from SpeechAntiSpoofingBenchmarks/Pellav2: direct link, hf CLI and curl.
- Browser
- Download file 6.82 kB
-
https://huggingface.co/SpeechAntiSpoofingBenchmarks/Pellav2/resolve/main/implementation-notes.md
- Command line
-
hf download hf://SpeechAntiSpoofingBenchmarks/Pellav2/implementation-notes.md
-
curl -L -o implementation-notes.md https://huggingface.co/SpeechAntiSpoofingBenchmarks/Pellav2/resolve/main/implementation-notes.md
6.82 kB
Pellav2 β implementation notes
Plan: speech-spoof-bench/docs/plans/2026-11-09-thirteen-model-batch-arena-submission.md (model #8).
Source
- HF
Sadanie/pellav2-audio-deepfake-detector@b668eb4dfe4719665692c73c1d70383680afff6b(downloaded by_ckpt_dl/dl_batch4.pyinto_ckpt_dl/Sadanie__pellav2-audio-deepfake-detector/). Card:license: mit, no LICENSE file, no paper, training data not described. Files:pellav2_detector.pt(2.32 GB, sha256d21aedfb52e2f87706c2753715a29665482b0f4a034800c31579df5d458559ac),pellav2_infer.py(sha256bf13cc97β¦6a67),config.json({"name": "pellav2"}only), README. - Architecture deviates from the plan row: the plan says "XLS-R + learned weights over all 25 layers"; the
script is w2v-BERT 2.0 (
facebook/w2v-bert-2.0, 24 conformer layers, 580 M params) on 80-bin kaldi fbank features (stacked to 160-d), softmax layer mix over the 25 hidden states, masked mean,Linear(1024,1). Layer mix + 25 layers as in the plan; only the backbone name differs. pellav2_infer.pyread in full (it only imports torch/transformers/torchaudio/scipy/soundfile/pyloudnorm and downloads from the Hub inhub_weights()); vendored into_net.py, never imported from the cache. A byte-identical copy_upstream_pellav2_infer.py(sha256 asserted) is used only bysource_parity.py.- Backbone: the script calls
Wav2Vec2BertModel.from_pretrained("facebook/w2v-bert-2.0")(unpinned) and then overwrites every tensor with the .pt. We buildWav2Vec2BertModel(config)from that repo'sconfig.jsononly (pinned @da985ba0987f70aaeb84a80f2851cfac8c697a7b, downloaded to_ckpt_dl/facebook__w2v-bert-2.0/, vendored asw2vbert2_config.json); no backbone weights are downloaded. - Weights: plain
OrderedDictstate_dict (776 fp32 tensors) βtorch.load(weights_only=True)works without an allow-list.convert_ckpt.pyβpellav2_detector.safetensors(same keys/values, round-trip asserted; sha256050234557deabc95d9697a2ede42a5152ed06ba500a4488a42542ab793af428f), written next to the download, symlinked here.
Wrapper (pellav2.py:Pellav2, _net.py)
- Preprocessing = the script's
load_clipafter decoding (the runner hands 16 kHz mono float32, so the resample branch is a no-op): mean removal β 2nd-order 20 Hz Butterworth high-pass (sosfiltfilt, whole clip, when len > 480) β centre 4 s (64000) crop β gain to β23 LUFS (clipped Β±40 dB) β clip to [β1, 1] β zero-pad to- The script's own window is already a 4 s crop, so it is kept as is (no Arena first-64600/repeat-pad window; that would change both the crop position and the padding the fbank trim relies on).
- Features as the script: fbank on the trailing-zero-trimmed clip (β₯ 400 samples), Γ32768, per-utterance mean/var norm, frame pairs stacked to 160-d, padded batch + attention mask (so a batch = one-at-a-time up to float noise; the backbone masks attention and conformer convs).
- pyloudnorm is not installed in this environment (not in conda, no wheel in the pip/uv caches). Without it the
script silently falls back to an RMS-based level (its
except: pass), so running the script here would NOT be the authors' intended path._net.integrated_loudnessre-implementspyloudnorm.Meter(16000).integrated_loudness(pyloudnorm 0.1.1: cookbook K-weighting high-shelf G=4 dB Q=1/β2 1500 Hz + high-pass Q=0.5 38 Hz vialfilter, 400 ms blocks with 75 % overlap, β70 LUFS absolute + β10 LU relative gating, same op order). Sanity: a 0 dBFS 997 Hz mono sine gives β3.083 LUFS at 16 kHz (BS.1770's 48 kHz reference β3.01; pyloudnorm's bilinear-warped cookbook filters at 16 kHz account for the 0.07 dB), and with 48 kHz coefficients a β23 dBFS-RMS sine reads β23.04. Equality with the real package is not proven here:test_matches_pyloudnorm_if_installeddoes it (skipped now). To close it:pip install pyloudnorm(pure Python) and run the tests. A residual difference would only be a per-clip gain offset of βͺ 1 dB. - Score = βlogit(fake) (script:
p_fake = sigmoid(logit)). FP32, TF32 off (cuDNN + matmul). - Strict load: all 776 checkpoint tensors map 1:1 onto
Detector.state_dict()(asserted set equality +strict=True);backbone.masked_spec_embed(SpecAugment token) is loaded but unused at inference. - Params: 580,494,170 total (all checkpoint tensors); 580,493,146 on the forward path (minus masked_spec_embed).
Checks
test_pellav2.py: crop/pad, RMS fallback, loudness target, BS.1770 reference tone, strict load, param count, eval mode, batch == single (atol 1e-3), determinism, bona fide > spoof on J-SPAW. 9 passed, 1 skipped (pyloudnorm).source_parity.py(loglogs_source_parity.txt): the authors'Detector+load_clip(from_pretrained swapped for the pinned config, .pt via weights_only strict, pyloudnorm shimmed to_net.integrated_loudness, clips as FLOAT wavs) one clip at a time vs the wrapper batched, on 16 J-SPAW_LA clips + 0.3 s (RMS-fallback branch) + 1.25 s + 8.6 s (centre-crop branch): preprocessing identical (max |Ξ| 0.0), max |Ξscore| = 2.8e-6. The script has no full-utterance mode, so "window off" = "4 s cap" here.
Batch sweep (sweep.py, 64000-sample view, 2026-11-10)
| card | best | notes |
|---|---|---|
| RTX 4070 Ti SUPER (idx 2) | 56.9 utt/s @24 | plateau 16β64 β batch_size = 24 |
| RTX 5060 Ti (idx 1) | 25.1 utt/s @4 | ~18 at bs β₯ 8 β 4 on 5060 |
Sanity (PyTorch, idx 2)
- Full J-SPAW_LA: EER 0.000 % (2397 trials, 0 skipped). Scores are saturated and fully separated (β β7.5 / +7.2). This is not inverted and not random, but a perfect score points to J-SPAW (or the same source) in the undisclosed training data. No trained_on entry, because the training data is unknown. Flagged for the maintainer.
TensorRT (trt_pellav2.py, sidecar trt_pellav2.json, TRT 10.16.1.11)
The engine takes (feats[B,199,160], mask[B,199]). The fbank front-end stays in torch, padded to a fixed 199 frames.
The fp32 ONNX is 2.33 GB, so it uses the dynamo exporter with external data (pellav2.onnx + .data). FP16 build
parity on 64 InTheWild clips gave Spearman 0.99954 (5060) and 0.99977 (4070), below 0.9999, so both cards fall back
to FP32 (TF32 cleared).
| card | precision | best batch | utt/s | build Spearman / max Ξ | J-SPAW EER | J-SPAW Spearman vs PT / max Ξ |
|---|---|---|---|---|---|---|
| RTX 4070 Ti SUPER (idx 2) | FP32 | 12 | 68.4 | 1.0 / 7.6e-6 | 0.000 (= PT) | 1.000000 / 1.0e-5 |
| RTX 5060 Ti (idx 1) | FP32 | 16 | 49.7 | 1.0 / 2.9e-6 | 0.000 (= PT) | 1.000000 / 1.4e-5 |
Both cards are accepted. Scripts: run_pt.sh, `run_trt_export.sh <2 |
1>` (run idx 2 first, it writes the ONNX), | |||||
| `run_trt_jspaw.sh <2 | 1>, cmp_scores.py`. |