BLOG

AI song checklist before release: 10 sound checks

A ten-point sound checklist for an AI-generated song before you upload it to a distributor, covering loudness, true peak, generator marks, clipping, noise, bass, mono and phone checks, and honest AI disclosure.

Before you upload an AI song, run a short sound check. It does not need a studio. It needs the file, a meter, a phone and about half an hour. This checklist covers ten points, in the order that saves the most time. Each one names a number or a test, and what to do when the result is not right.

1. Keep the best file you have

Start with the lossless file. A WAV or FLAC export keeps the most detail for the later steps. An MP3 or M4A can be used, but its encoding losses cannot be undone. Keep the original file next to every version you make, so you can compare them later.

If you are not sure which download your generator gives you, check its download terms first. For Suno, read Suno download limits (2026).

2. Check integrated loudness

Measure the whole track with the free LUFS checker. The target for streaming is about −14 LUFS integrated. Spotify publishes this as its reference. A track far above it is turned down on playback, and a track far below it is not turned up by every service.

If the song is much louder or quieter, master it before you release. The free AI mastering page explains the target and the report you get.

3. Check the true peak

The true peak should be at or below −1 dBTP. Measure the final WAV, and then measure the MP3 or AAC version if you deliver one. Encoding can raise the peak, so the WAV alone is not enough. The true peak page explains why this matters.

If the true peak is above −1 dBTP, lower the limiter ceiling or push less, then measure again.

4. Listen for generator marks

Solo the cymbals, the hi-hats and the "s" sounds. Listen for a ring at one fixed pitch, a seam where one block meets the next, or a row of peaks you can hear as a tone. Cleanup targets these marks, and it leaves loudness and tone as they are. Run it in the lab with "Clean AI artifacts" only if you hear them. For the details, see AI artifact cleanup.

Cleanup does not change the model's own sheen. If the highs sound glassy in every part, a new take is the fix.

5. Check for clipping

Run the free de-clipper check. It returns the clipped spots, their share of the audio and the clip level. Listen to the loudest hits as well. A crackle on the transients is a common sign.

Short clipped peaks can be repaired as part of mastering. Long flat sections cannot, so a new take may be the better choice.

6. Check for hiss and noise

Listen to the intro, the gaps between phrases and the end of the song on headphones. A steady hiss there is in the file. A grain that moves with the vocal may come from the vocal track or a stem split. Use the lightest fix that helps, and compare with the original at matched loudness. The stem splitter page explains that split stems can keep faint artifacts.

7. Check the bass

Solo the bass and listen for a note that booms or rings. Then listen to the low-mids with the vocal back in. If the words are hard to hear, the bass and the low-mids are crowding each other. Fix the balance before you master, because mastering cannot separate a muddy mix.

8. Run a mono check

Play the song in mono. The vocal should stay clear, and the bass should not vanish. The LUFS checker reads stereo correlation, and a value below zero can cancel on a mono speaker. If a part disappears in mono, check the wide effects and the stereo bass first.

9. Check on a phone speaker and in the car

Play the master on a phone speaker and in a car if you can. These systems show a harsh top end, a thin middle and a boom in a way that studio monitors can hide. Listen to the first chorus, where the density is highest. If the chorus is harsh on the phone, go back to the highs and the peak checks.

10. Answer the AI questions honestly

Distributors and platforms ask about AI use when you upload. Spotify, Deezer and DistroKid each have a way to record it, and some show the answer to listeners. Answer for each part that was AI-generated, such as the lyrics, the composition, the vocals and the instrumental. Give the same answers on every platform. Keep a short note of what came from AI, the generator and your plan, and the date you downloaded the file.

Cleanup does not change where the track came from. Do not treat it as a way to avoid a question. For the full order of a release, read how to release an AI-assisted track honestly in 2026.

We don't hide AI origin — we recommend disclosing it. Cleanup is about how the track sounds.

The checklist in one table

# Check Target or test If it fails
1 File format WAV or FLAC kept Keep the original, ask for a better download
2 Integrated loudness About −14 LUFS Master to target
3 True peak At or below −1 dBTP, also after encoding Lower the ceiling, measure again
4 Generator marks No ring at a fixed pitch, no seams Clean in the lab, then compare
5 Clipping No crackle on hits, spots checked De-clip, or a new take for flat sections
6 Hiss and noise Quiet parts clean Lightest noise fix, compare
7 Bass No boom, clear vocal Balance before mastering
8 Mono Vocal and bass hold Check wide effects and stereo bass
9 Phone and car Chorus sounds right Return to highs and peaks
10 AI disclosure Answers match everywhere Answer honestly at upload

FAQ

Do I have to tell a distributor that my song is AI-generated?

When a platform or distributor asks, yes. Answer the AI questions for each part of the song, and give the same answers everywhere. Cleanup and mastering do not change that answer.

Is −14 LUFS a rule I must follow?

No. It is a reference that many streaming services use. A track at a different level still plays, but a very loud master is turned down on playback. Master to about −14 LUFS, and keep the true peak at or below −1 dBTP.

What is the fastest way to run this checklist?

Start with the LUFS checker for loudness and true peak. Then listen on headphones and a phone speaker for the rest. Most problems show up in the first chorus, so start there.

Run the checklist on your file

Start with the free LUFS checker. It reads loudness and true peak without an account, so you can run the first two checks in a minute.

Master a track free