ArtifactBench — AI-Generated Music Detection Benchmark

ArtifactBench v2 — lineage-aware frozen protocol

ArtifactBench v2 adds a metadata-first, lineage-aware evaluation protocol while preserving the v1 and v1.1 releases below. Its frozen primary cohort contains 828 entries (605 AI-generated and 223 real) across 15 source strata, split by content lineage into calibration, validation, and sealed-test partitions before final model comparison.

The v2 package is under v2/. It contains the path-free frozen manifest, per-model probabilities, structured inference and chunk failures, source-level and paired metrics, uncertainty estimates, provenance records, and checksums. It does not add or replace audio. Existing v1 audio shards and manifests remain unchanged.

On the 562-track common-success sealed-test intersection, ArtifactNet v9.4 obtains AUROC 0.982 and balanced accuracy 0.918; the public Deezer detector obtains 0.761 and 0.776. These values are tied to the v2 cohort, calibration-only threshold policy, and declared coverage rules and should not be compared as if they were measured on the v1 cohort below.

Paper (preprint): ArtifactBench: Lineage-Aware Evaluation of AI-Generated Music Detectors under Distribution Shift.

Try ArtifactNet on your own audio

Try the free live demo → · Scope and limitations

Upload an audio file or choose a built-in example. No account is required within the demo’s free limits. This is the hosted ArtifactNet detector, not a browser-based reproduction of the four-model ArtifactBench evaluation; the live service may differ from the version-pinned research model.

---

ArtifactBench v1 — original public benchmark

A multi-generator evaluation benchmark for AI-generated music forensic detection, covering 22 AI generators and 6 real music sources.

Motivation

Existing benchmarks (SONICS: 5 generators, MoM: 6 generators) only measure in-distribution performance. Models reporting high F1 on these benchmarks fail catastrophically on out-of-distribution generators:

• CLAM (194M params, F1=0.925 on MoM) → F1=0.824 on ArtifactBench
• SpecTTTra (19M params, F1=0.97 on SONICS) → F1=0.766 on ArtifactBench

ArtifactBench evaluates what matters for deployment: generalization across diverse generators.

Sanity Check Protocol

Per-source pass/fail thresholds:

• Real source FPR ≤ 5%
• AI source TPR ≥ 90% (Stable Audio: ≥ 60%)
• Codec invariance: mean Δ ≤ 0.15, max Δ ≤ 0.35

Baseline Results

Model	Params	F1	FAIL	Suno v4 TPR	Real FPR
ArtifactNet v9.4	4.2M	0.983	4/28	98%	1.5%
CLAM (MoM)	194M	0.824	16/28	78%	70.5%
SpecTTTra	19M	0.766	23/28	55%	21.4%

Public Model Comparison — 8-way (v1.1 test partition, 2026-07-04)

Eight publicly-available detectors scored on identical files with the same runner (adapters in artifactbench/models/), τ = 0.5. n = 2,104 (1,388 AI / 716 real — the locally-restored real set; see the provenance caveat in v1.1/RESULTS_8WAY.md).

Rank	Model	Params	F1	Precision	Recall (TPR)	FPR	Sanity FAIL
1	ArtifactNet v9.4 (public ONNX)	4.2M	0.952	0.932	97.3%	13.8%	8/28
2	AI-Music-Detection AST-60s	90.8M	0.840	0.848	83.1%	28.9%	16/28
3	CLAM (MoM)	194.3M	0.787	0.711	88.3%	69.7%	14/28
4	SpecTTTra α-120s	18.7M	0.777	0.880	69.5%	18.4%	22/28
5	Deezer ISMIR fakeprint LR	3.6K	0.754	0.906	64.6%	13.0%	18/28
6	FST (Mippia, arXiv:2601.13647)	174.4M	0.735	0.984	58.7%	1.8%	17/28
7	DeepFense EAT+Nes2Net	—	0.650	0.589	72.4%	97.8%	11/28
8	SpecTTTra β-5s	18.7M	0.563	0.884	41.3%	10.5%	24/28

The ArtifactNet production pipeline (v9.7 / cnn_v95, PyTorch — not the public ONNX export above) measured on the same partition: F1 0.984 / TPR 98.9% / FPR 4.2% (1/28 FAIL). Per-source TPR/FPR tables, ROC summaries, and notes: v1.1/RESULTS_8WAY.md.

Usage

from artifactbench.bench import main
# or
# python -m artifactbench.bench --model artifactnet --manifest artifactbench_v1_manifest.json

Per-Source Breakdown (v1.0.1)

Source	Class	Tracks	bench_origin: test	Generator
aime_musicgen_large	AI	200	30	MusicGen Large
aime_musicgen_medium	AI	200	30	MusicGen Medium
aime_musicgen_small	AI	200	30	MusicGen Small
aime_riffusion	AI	200	30	Riffusion
aime_stable_audio_v1	AI	200	50	Stable Audio v1
aime_stable_audio_v2	AI	200	50	Stable Audio v2
aime_suno_v3	AI	200	30	Suno v3
aime_suno_v35	AI	200	30	Suno v3.5
aime_udio	AI	200	30	Udio (AIME)
mom_diffrythm	AI	200	100	DiffRhythm
mom_riffusion	AI	200	100	Riffusion (MoM)
mom_udio	AI	200	100	Udio (MoM)
mom_yue	AI	200	100	Yue
sonics_chirp-v2-xxl-alpha	AI	200	80	Chirp v2
sonics_chirp-v3	AI	200	80	Chirp v3
sonics_chirp-v3.5	AI	200	80	Chirp v3.5
sonics_udio-120s	AI	200	80	Udio 120s
sonics_udio-30s	AI	200	80	Udio 30s
suno_cdn_latest	AI	200	100	Suno CDN (post-freeze)
suno_extra	AI	200	80	Suno extras
udio_cdn_latest	AI	200	35	Udio CDN (post-freeze) — v1.0.1 balanced
udio_extra	AI	200	80	Udio extras
sonics_real	Real	500	300	SONICS real partition
mom_real	Real	400	200	MoM real (mp3 + wav)
fma_hardneg	Real	300	150	FMA mp3 hard-negatives
mom_extra_real	Real	200	110	MoM extra real
mom_real_wav	Real	200	42	MoM real WAV variants
youtube_hardneg	Real	200	73	YouTube curated hard-negatives
TOTAL	—	6,200	2,280	28 sources, 22 AI generators

Real sources are intentionally over-represented (1,800 total) to enable rigorous FPR estimation across diverse codec and production conditions.

Files

• artifactbench_v1_manifest.json — Track manifest with bench_origin tags
• metadata.json — Dataset statistics and generator list

Citation

@article{oh2026artifactnet,
  title        = {ArtifactNet: Detecting AI-Generated Music via Forensic Residual Physics},
  author       = {Oh, Heewon},
  journal      = {arXiv preprint arXiv:2604.16254},
  year         = {2026},
  eprint       = {2604.16254},
  archivePrefix= {arXiv},
  primaryClass = {cs.SD},
  doi          = {10.48550/arXiv.2604.16254},
  url          = {https://arxiv.org/abs/2604.16254}
}

arXiv: 2604.16254 · DOI: 10.48550/arXiv.2604.16254

License

CC BY-NC 4.0

v1.1 (2026-07-03) — integrity-audit purged test partition

An audit of the test partition against the ArtifactNet v9.4 training manifest found 34 overlapping real tracks (all YouTube-derived); 5 further real tracks became unrecoverable. v1.1/ ships the purged partition (n = 2,224), a per-track status CSV, and official v1.1 result bounds. v1 files are unchanged — results computed on v1 remain reproducible. See v1.1/RESULTS_v1.1.md.

8-way public model comparison (2026-07-04)

Extends the v1.1 evaluation with five more publicly-available detectors. The summary table is in Public Model Comparison above; per-source breakdowns and notes are in v1.1/RESULTS_8WAY.md.