Before You Hit Master: A Source-Quality Gate for AI Music

01 / Understand the chain

Why AI Music Breaks Traditional Mastering Chains

Many generated stereo files arrive already loud. If the highest sample or reconstructed true peak sits near full scale, an EQ boost, saturation stage, or compressor can push the signal over the available digital ceiling before the final limiter even starts. Turning down the clip does not undo distortion printed into the export, but it prevents your own chain from adding another overload.

Start by measuring, then lower the file before the first processor. A true peak around -3 dBTP is a conservative working target for this gate, not a delivery standard. It gives common mastering moves room to operate. If the source still crackles in exactly the same places after lowering clip gain, the crackle is likely part of the file and needs restoration or a new generation.

Low frequencies can spend your limiter budget

Energy below the useful range of your speakers may be difficult to hear while still moving a meter and triggering gain reduction. A sustained sub note can make the limiter pull down the entire mix on every kick. Do not automatically high-pass every track at one frequency. Compare the file on headphones and speakers, inspect the spectrum, and ask whether the lowest octave supports the song or merely makes the rest of it pump.

Wide does not always mean stable

A wide pad or bass texture can sound impressive in stereo and partly cancel when the left and right channels are summed. That loss may become more obvious after limiting changes the balance around it. Use a mono fold-down as a listening test. A correlation meter can support the decision, but the musical question is whether the kick, bass line, lead, and groove keep their identity.

If you cannot name the dominant problem yet, return to the initial audio triage guide. Mastering should begin only after the repair decision is clear.

02 / Make the decision

The 5-Point Mastering Go/No-Go Gate

Run these checks on the exact file you intend to master. Bypass any loudness normalization in the player, keep the monitor level comfortable, and record pass or repair beside each row. The targets are practical starting points for this workflow rather than universal specifications.

A pass does not mean the meter must sit on one exact value. It means the source behaves predictably and does not ask the mastering chain to solve a mix or generation problem. Write down what you heard. “Low end holds in mono” is more useful than a screenshot of a meter with no listening context.

After a clean pass, keep the same file for your broader 3-pass release readiness check. The mastering gate answers whether the source can enter the chain; the release check asks whether the finished song translates outside your workstation.

03 / Export cleanly

Setting Up the Pre-Master Export

Export a 24-bit or 32-bit float WAV. Keep the project sample rate unless a specific delivery workflow requires conversion. Do not create an MP3 for mastering: lossy encoding removes information and can add pre-echo or high-frequency texture that becomes another variable in the chain.

Disable loudness normalization and any “make louder” option. Remove the final loudness limiter if it exists only to raise level, but keep deliberate sound-design processing that defines the mix. If bypassing a bus compressor changes the groove, it is part of the mix; if a limiter is shaving peaks solely to win a loudness comparison, print a clean version without it.

Do not dither every time you export. Dither is relevant when reducing bit depth, and it should normally be applied once at the final delivery conversion. A 24-bit or 32-bit float pre-master gives the mastering stage room to process without making an early 16-bit decision.

Leave a small, intentional amount of room before the first sound and after the final decay. Check both boundaries at normal level and on headphones. A 20 ms fade can remove a cut click, but do not use a fixed fade to amputate a reverb tail. Name the file clearly—for example, song-premaster-24b-v03.wav—and keep the untouched generated export beside it.

04 / Choose the finishing path

AI Mastering Services vs. Manual Limiting

An automated mastering service is fast and consistent. It can provide a useful first version when the source already passes the gate, and it may reveal whether the tonal balance leans too dark, bright, or bass-heavy. What it cannot know is which strange texture you consider intentional. A bright preset may treat metallic shimmer as desirable air; an aggressive low-end profile may spend most of its limiting on sub energy.

Manual limiting gives you more control over timing, true-peak ceiling, and how much gain reduction occurs in each section. It also gives you more ways to make a poor source worse. Begin with a conservative ceiling near -1 dBTP, raise level slowly, and listen to the loudest chorus. The moment the snare loses attack, the bass starts modulating the whole mix, or the vocal hardens, back off and check whether the source gate missed something.

Choose by feedback, not prestige. If an automated result stays balanced and survives your listening checks, it is a valid result. If the song needs section-specific control, manual work is the clearer route. Neither path repairs clipped words, broken pronunciation, unstable stereo bass, or a click at the edit boundary.