ishiki-labs/ishiki-deepfake-0.5)ishiki-labs/ishiki-deepfake-0.5 is an open-weights audio deepfake and speech anti-spoofing detector developed by Ishiki Labs, initialized from the Wav2Vec2-XLS-R-300M + AASIST (TakHemlata/SSL_Anti-spoofing) (318M parameters, Paper) base checkpoint.
This repository provides two model checkpoints and the forward-pass calibration dataset (distilled_dataset/):
86.8M Compact Checkpoint (354 MB, variant="86.8M", default).318M Full Checkpoint (1.27 GB, variant="318M").distilled_dataset/: The forward-pass manifests and extracted feature/waveform bundles (12.4K main-fit rows + 3.2K calibration rows + 2.4K multi-codec rows + paired acoustic probes).86.8M Compact & 318M Full vs. Base Checkpoint & Published SOTA)Lower Equal Error Rate (EER %) and higher ROC-AUC indicate stronger detection accuracy.
| Dataset & Official Paper | Partition (Trials) | Base (W2V2-AASIST, 318M) EER / AUC |
Published / Leaderboard Reference SOTA (Method, Params, EER / AUC) | Did Reference SOTA Train on This Dataset? | What Our Forward-Pass Set Used (distilled_dataset/) |
Ours (318M Full) EER / AUC |
Ours (86.8M Compact) EER / AUC |
|---|---|---|---|---|---|---|---|
| ASVspoof 2019 LA · [Paper] | Eval (71.2K) |
0.22% / 0.9997 |
LaST-Net (317M): 0.09% · Nes2NetX (318M): 0.12% |
Yes (25.4K train + 24.8K dev) |
3.0K train + 600 dev (0 eval) |
0.12% / 0.99999 |
0.16% / 0.99996 |
| ASVspoof 2021 LA · [Paper] | Eval (148.2K) |
8.11% / 0.9779 |
XLS-R Conformer: 0.69% |
Trained on ASV19 LA train |
0 clips (uses ASV19 + paired PSTN probes) |
1.42% / 0.9964 |
1.51% / 0.9959 |
| ASVspoof 2021 DF · [Paper] | Eval (533.9K) |
6.63% / 8.32% |
VeriSpoof-Base (2.85B): 0.97% · XLSR-Mamba (319M): 1.88% |
Multi-corpus (8 corpora incl. ASV19 & ASV5) |
0 clips (uses ASV19/ASV5 + paired codec probes) |
1.98% / 0.9983 |
2.19% / 0.9980 |
| In-The-Wild (ITW) · [Paper] | Full (31.8K) |
11.22% / 0.9572 |
UTCN-XLSR2B (433M): 0.37% · DF-Arena-1B (1B): 0.91% · XLSR-Mamba (319M): 6.71% |
Multi-corpus (7 corpora + ITW selection) |
0 clips in forward-pass set |
2.51% / 0.9958 |
2.72% / 0.9951 |
| DECRO (EN + ZH) · [Paper] | Eval (37.3K) |
9.57% / 0.9583 |
Spectra-AASIST (316M): 0.17% |
Undisclosed (Arena submission) | 0 DECRO clips (1.28K CFAD Chinese in ancestor) |
1.30% / 0.9986 |
1.41% / 0.9980 |
| ASVspoof 5 (Track 1) · [Paper] | Eval (680.8K) |
16.25% / 0.9003 |
Modulate VELMA-2 (316M, Prop.): 0.38% · NSGA-II Fusion (6.6B, Open): 2.37% |
Yes (full ASV5 train + dev) |
5.0K train + 3.44K unique dev (0 eval) |
5.73% / 0.9856 |
5.80% / 0.9857 |
| DFADD · [Paper] | Test (3.8K) |
11.88% / 0.9436 |
VeriSpoof-Base (2.85B): 0.00% · XLSR+SLS (340M): 7.54% |
Yes (full DFADD train) |
704 train + 350 dev aliases |
0.27% / 0.99995 |
0.53% / 0.99991 |
| TIMIT-TTS (Clean) · [Paper] | Full (20.2K) |
2.09% / 0.9954 |
Standard MoE: 0.93% (0.9512 AUC) |
No (unseen test corpus) | 0 clips (completely unseen) |
0.01% / 0.999998 |
0.03% / 0.999997 |
| TIMIT-TTS (Time-Warped DTW) · [Paper] | Full (20.2K) |
10.51% / 0.9535 |
RawNet2 Baseline (17.6M): 0.7800 AUC |
No (ASV19-trained) |
0 clips (completely unseen) |
0.69% / 0.9995 |
0.93% / 0.9997 |
| TIMIT-TTS (Augmented) · [Paper] | Orig Real (20.2K) |
2.14% / 0.9946 |
RawNet2 Baseline (17.6M): 0.9300 AUC |
No (ASV19-trained) |
0 clips (completely unseen) |
0.46% / 0.9996 |
1.17% / 0.9994 |
| TIMIT-TTS (DTW + Augmented) · [Paper] | Orig Real (20.2K) |
12.84% / 0.9381 |
XLS-R 2B + LogReg (2B): 3.60% |
No (ASV19 train + dev) |
0 clips (completely unseen) |
1.85% / 0.9972 |
1.86% / 0.9983 |
| MULTISpoof · [Paper] | Test (43.3K) |
14.62% / 0.9214 |
Asilomar 2026 Paper (no public test table) | N/A | 1.33K val clips in ancestor (0 test) |
2.94% / 0.9964 |
3.32% / 0.9955 |
Fake-or-Real (for-norm) · [Paper] |
Testing (4.6K) |
7.42% / 0.9816 |
VeriSpoof-Plus (2.85B): 0.09% · XLSR+SLS (340M): 5.08% |
Yes (full for-norm train) |
0 clips (completely unseen) |
3.13% / 0.9953 |
4.29% / 0.9912 |
| MLAAD · [Paper] | Eval (64.0K) |
15.40% / 0.9120 |
AASIST Voiced/Unvoiced: 5.82% · WavLM-LDA: 10.00% |
Yes (custom MLAAD splits / CV) |
664 train + 200 dev aliases |
6.84% / 0.9785 |
7.51% / 0.9730 |
| DiffSSD · [Paper] | Test (54.6K) |
18.90% / 0.8840 |
Wiener-Hopf 2D CNN (0.42M): 2.95% · Wav2Vec2 Baseline: 3.00% |
Yes (full ~70K DiffSSD train) |
755 train + 249 val aliases |
8.92% / 0.9742 |
9.77% / 0.9691 |
86.8M Compact & 318M Full Variants)import sys
import numpy as np
from huggingface_hub import snapshot_download
repo_dir = snapshot_download("ishiki-labs/ishiki-deepfake-0.5")
sys.path.insert(0, repo_dir)
from inference import create_predictor
# Load either "86.8M" (default compact model) or "318M" (full model)
predict_86m = create_predictor(repo_dir, device="cuda:0", batch_size=32, variant="86.8M")
predict_318m = create_predictor(repo_dir, device="cuda:0", batch_size=32, variant="318M")
wavs_16khz = np.random.randn(4, 64600).astype(np.float32) * 0.2
print("86.8M Compact scores:", predict_86m(wavs_16khz))
print("318M Full scores :", predict_318m(wavs_16khz))
distilled_dataset/s86_dev_paired_mc.npz 2.7 GBmodel_318m_full.safetensors 1.3 GBdistilled_dataset/f12_7s_pk1_dev_and_nuis.npz 696 MBdistilled_dataset/asv5_multicodec_2400_pk1.npz 642 MBdistilled_dataset/paired_real_and_spoof_ffmpeg_encodec.npz 635 MBdistilled_dataset/paired_awgn_real_and_spoof.npz 473 MBmodel.safetensors 354 MBdistilled_dataset/cc3_pk1_dev_and_nuis.npz 283 MBfacebook/wav2vec2-xls-r-300m