1. Separate musical decisions from technical defects
Before processing, listen once without touching a control. Write down musical issues such as balance, arrangement and vocal placement separately from signal issues such as clipping, pumping, brittle highs or unstable stereo. Mastering can often improve the second list. It rarely fixes the first without side effects.
If you can return to the generation, stem or mix stage, do that for missing notes, garbled lyrics, timing mistakes or an instrument that is fundamentally too loud. A stereo master is the least flexible place to make those repairs.
2. Build a clean, reversible repair pass
Work from broad to narrow. Remove DC offset or isolated clicks if present, control only clearly excessive resonances, and use gentle dynamic equalization when a harsh band appears intermittently. Heavy denoising can replace one artifact with swirling or watery texture, so compare every repair against the unprocessed file.
For unstable or smeared high frequencies, a modest high-shelf reduction or controlled roll-off can sound more natural than aggressive restoration. Preserve consonants, cymbal attacks and ambience; those details carry clarity even when the top octave is imperfect.
3. Shape tone and dynamics after repair
Once distractions are under control, make the same mastering choices you would for any mix: tonal balance, low-frequency focus, stereo stability, crest factor and final level. Use a relevant commercial reference for direction, not as a target to copy exactly.
Saturation or tape-inspired movement can make a sterile source feel more cohesive, but it also creates harmonics and changes peaks. Add it because it helps the song at matched loudness, not because analog processing proves human involvement.
4. Verify the deliverable, then document it honestly
Check the final file on headphones, speakers and mono playback. Listen to the encoded version when a distributor will transcode the master. Confirm duration, channel layout, sample rate and peak behaviour, and keep a lossless archive before creating convenience copies.
An AI detector produces a probabilistic classification based on patterns in its training and processing pipeline. Research shows that detection performance can fall on unseen generators or after ordinary transformations. A changed score is not evidence that authorship changed.
METHOD
How this guide was prepared
This guide combines established digital-audio measurement practice with published research on generated-music artifacts and detector generalization. It deliberately avoids claiming that every generator creates the same artifact.
This is general educational information, not a guarantee of platform acceptance, pressing approval, authorship, or detector outcome. Requirements and services can change; check the receiving party before delivery.
PRIMARY SOURCES
Read the source material
- ITU-R BS.1770-5: loudness and true-peak measurement (opens in a new tab)
- Audio Engineering Society: loudness and true-peak primer (opens in a new tab)
- Transactions of ISMIR: The AI Music Arms Race (opens in a new tab)
- Afchar et al.: A Fourier Explanation of AI-music Artifacts (opens in a new tab)
- Apple Digital Masters and mastering tools (opens in a new tab)