ArtifactNet: Detecting AI-Generated Music via Forensic Residual Physics

Researchers introduced ArtifactNet, a compact audio forensics system for detecting AI-generated music.

It spots synthetic tracks by reading codec artifacts, residuals, and HPSS-based features left by neural audio generators.

Evaluated on ArtifactBench, a benchmark of thousands of real and AI tracks, it outperformed prior detectors on unseen data.

Codec-aware training improved robustness across formats such as WAV, MP3, AAC, and Opus.