Audio Forensics & Restoration

AI-generated music often suffers artifacts from lossy training data — 16 kHz ceilings, phase incoherence, spectral smearing. Six forensic tells identify it, and a diagnostic plus restoration path was proposed to repair it.

Artifact Index

TellSymptom
HF shelfHard 16/22 kHz cutoff
TransientsSmeared drum attacks
StereoCollapsed or pseudo-width
PhaseIncoherent left/right correlation
VocalsRobotic formants
NoisePumping artifacts

Rehab Pipeline

Historical proposal

This three-step path from AI generation to release-grade fidelity is design documentation, not current executable MixBox behavior.

  1. Diagnose — detect specific issues, e.g. with mastering/scripts/diagnose.py in the mastering pipeline.
  2. Neural super-resolution — break the 16 kHz ceiling; theorized to run before restoration.
  3. Unified restoration — fix dynamic, spatial, and EQ degradations.

SR Tools

ToolNiche
AudioSRMost versatile super-resolution; cloud/GPU preferred, or CPU batch
LavaSRLocal CPU-only processing, ~50x realtime
NovaSRUltra-fast fallback

Historical

Raw AI output sits in a legal grey zone; establishing the human master in a DAW was treated as a requirement for copyright defense and streaming-platform acceptance.