Crest Factor — The One Number That Tells You If Your AI Mix Is Alive or Crushed
Short answer: Crest factor is peak level minus RMS (average) level, in dB — the single number that tells you whether a mix is punchy or crushed. Roughly: below ~6 dB is over-limited and fatiguing, ~6–12 dB is the healthy modern zone, and above ~12 dB is under-mastered and small-sounding. AI tracks from Suno and Udio often arrive already compressed with a low crest factor, and stacking a hard limiter on top collapses it further — the track gets louder but feels smaller. Since streaming normalizes loudness anyway, crushing dynamics is a losing trade. These ranges are working guidelines, not hard rules.
When mastering a track from Suno or Udio, a lot of people focus only on whether it's "loud enough." But the real reason a song feels dull and fatigues your ears fast is usually dynamics, not loudness. The number that shows the health of those dynamics is the crest factor — and once you know how to read it, you can tell in one glance whether you've crushed a track or simply haven't finished it.
What crest factor is
It's simple: peak (the loudest instant) minus RMS (the average level), measured in dB.
- A wide gap → the punchy transients are alive and the track breathes.
- A narrow gap → loud and quiet parts barely differ, so the whole thing is one squashed block.
In other words, crest factor tells you whether a track is breathing or being crushed. It's not a quality score by itself — it's a diagnostic.
What's a healthy range
There's no single right answer across genres, but the rough working ranges are:
| Crest factor | State | What it sounds like |
|---|---|---|
| Below ~6 dB | Over-limited | Punch gone, fatigue sets in fast |
| ~6–12 dB | Healthy modern zone | Loud, but still breathing |
| Above ~12 dB | Under-mastered | Small and flat next to commercial releases |
The point isn't to hit a target number — it's to use where you land as a clue: have you crushed the track, or not finished mastering it? A sparse ballad and a dense EDM drop will sit in different places, and that's fine.
Why it collapses on AI music
AI-generated tracks often come out already heavily compressed. Instead of each instrument having distinct dynamics like a careful human recording, the whole mix arrives somewhat flattened — that's just how these models tend to render a full arrangement.
Push a hard limiter on top of that — thinking "louder is better" — and the already-narrow crest factor collapses further. The result is a paradox: the volume goes up, but the song feels smaller. Loud and full are not the same thing. This is closely tied to why AI music needs a different mastering approach than a well-tracked human recording.
The most common misconception: "louder is better"
Loud doesn't mean alive.
A crushed 3 dB master actually sounds smaller and duller than a healthy 9 dB one at the same LUFS. And streaming services normalize loudness on playback — so the volume you forced gets turned back down anyway, while the dynamics you crushed don't come back. That's why "as loud as possible" is an especially losing strategy for AI music. If your track already sounds quiet on streaming, the fix is usually loudness and tone, not more limiting — see why Suno songs sound quiet on Spotify and LUFS targets for streaming.
How to check and restore it
- Measure first. Most DAW meters or free loudness meters show peak and RMS (or short-term LUFS) together. The gap between them is your crest factor.
- Suspect the limiter. If your master-bus limiter's gain reduction is excessive, dynamics die there first. Backing it off slightly often lets the track breathe again.
- Keep glue compression light. Bus compression that holds the track together needs only 1–2 dB of gain reduction. More than that is crushing, not gluing.
- A/B at matched volume. When comparing to bypass, always match levels — it removes the "louder just sounds better" illusion.
Common mistakes
- Watching only LUFS. Two tracks at -14 LUFS can feel completely different — the one with more crest factor sounds bigger.
- Maximizing loudness before streaming normalization turns it back down. You lose the dynamics permanently and gain nothing on playback.
- Judging A/B comparisons without level-matching. The louder version always "wins" — which tells you nothing.
- Treating a range like a target. Aiming for exactly 9 dB on every song ignores genre and arrangement. Use it to diagnose, not to chase.
- Blaming the export. A flat AI export can be revived; over-limiting it can't be undone.
FAQ
What is a good crest factor for streaming? As a working guideline, roughly 6–12 dB keeps a track loud but still breathing. Sparse acoustic material sits higher; dense electronic material sits lower. Treat it as a diagnostic range, not a fixed target.
Why does my AI track sound dull even though it's loud? Almost always crushed dynamics — a low crest factor. Loudness and life are different things; streaming turns your forced volume down but won't restore the punch you limited away.
Does crest factor change my LUFS? No — you can have the same integrated LUFS with very different crest factors. That's exactly why loudness alone doesn't tell you whether a master is healthy.
How do I raise a crest factor that's too low? You can't fully "un-crush" an over-limited master — the transients are gone. Start from the cleaner export and limit less, keeping bus compression to 1–2 dB of gain reduction.
Do I need special tools to measure it? No. Any meter showing peak and RMS (or short-term LUFS) side by side gives you the gap. Some mastering tools report crest factor directly.
If you only watch "how loud," you'll miss the song's life. Check the crest factor once and you'll see at a glance whether your track is crushed or simply not finished — streaming handles loudness for you, but dynamics, once lost, don't come back. To master an AI track to a streaming-friendly loudness while keeping healthy dynamics, you can preview it free (no account) at antiaimaster.com/studio. For the bigger picture, see what mastering actually does to AI music.