---
title: ArtifactNet
emoji: 🎵
colorFrom: yellow
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
python_version: "3.10"
pinned: true
license: other
short_description: AI-generated music detection (v9.4 Forensic CNN + HPSS)
hardware: cpu-basic
models:
  - intrect/artifactnet-models
---

# ArtifactNet — AI Music Forensic Detector

Upload a track (WAV / MP3 / FLAC, ≤100 MB, ≤5 min). ArtifactNet analyses
spectral + harmonic-percussive forensic features and returns a per-segment
P(AI) distribution.

- **Backbone**: STFT → U-Net artifact residual → HPSS → 7-channel features → CNN
- **Verdict**: energy-weighted median across 4-second segments
- **Runtime**: ONNX Runtime on HF Space CPU (~30–60 s per 4-minute track)

## Paper

ArtifactNet: Forensic Detection of AI-Generated Music via HPSS and Residual
Analysis — [arXiv:2604.16254](https://arxiv.org/abs/2604.16254).

## Links

- Production dashboard: [dash.intrect.io](https://dash.intrect.io)
- Pricing / API: [intrect.io](https://intrect.io)

## Notes

- Short files (<60 s) have fewer segments and lower confidence.
- Mono input disables stereo phase features.
- Heavily processed audio (bitcrushing, vinyl rips) may affect results.
- YouTube / URL intake is disabled on this Space — use the dashboard for batch
  processing.
- Only the ONNX graphs (inference-only, no training metadata) are published;
  the original PyTorch checkpoints remain private.