Why Your AI Track Clips After Export — Intersample Peaks and True Peak Limiting
The Distortion That Doesn't Show on Your Meter
You export a Suno track. The waveform peaks at -0.1 dBFS — technically below 0, so no clipping. You upload to Spotify or play it through a Bluetooth speaker. There's a subtle crunchiness that wasn't there in the DAW preview. Not dramatically distorted, but the grit is real.
This is the intersample peak problem, and it catches AI music creators more often than anyone talks about.
What Is an Intersample Peak?
Digital audio stores sound as a sequence of discrete sample values. When your speakers or DAC (digital-to-analog converter) reconstruct the continuous waveform from those values, the actual analog signal passes through points that were never stored — peaks that occur between samples.
A standard peak meter reads the stored sample values. It has no way to see what happens in the gaps. So a track that measures -0.1 dBFS on your export meter can produce analog peaks above 0 dBFS — sometimes several decibels above — when converted back to audio.
This isn't a flaw in digital audio. It's a predictable consequence of waveform reconstruction that has been documented since the early days of CD mastering. The solution is true peak measurement: the signal is oversampled (typically 4× or 8×) before metering, which reveals what actually happens between the stored samples.
Why AI Music Gets Hit Harder
AI music generators produce audio with specific spectral characteristics that make intersample peaks worse than typical recorded content.
Natural recordings — drums, guitar, voice — are dominated by transients: sharp attacks that decay quickly. These create intersample peaks too, but they're brief and spread across the frequency spectrum.
AI-generated audio tends to produce dense, sustained high-frequency content. The consistent mid-and-treble textures that fill every moment of a generated track contain many closely spaced frequency components. When these components align in phase between sample points, they amplify each other in the reconstructed waveform in ways that transient-heavy sounds don't. The denser and more uniform the high-frequency content, the larger the gap between what your peak meter shows and what the DAC actually outputs.
This is why AI music benefits more from careful true peak control than a typical live recording at the same measured loudness level.
The Standard: -1 dBTP
The broadcast and streaming standard (EBU R128, ITU-R BS.1770-4) specifies a true peak ceiling of -1 dBTP (dB True Peak). This provides enough margin to absorb the oversampled waveform reconstruction without audible distortion.
For streaming distribution, -1 dBTP is the important number — not just for the reconstruction math, but because platforms transcode your WAV to MP3 or AAC. Transcoding slightly raises peak levels, often by 0.5 to 1 dB. A track that lands exactly at 0 dBTP before transcoding can emerge from the encoder above 0 dBFS.
How to Check Your Own Exports
Standard peak meters in most DAWs and export dialogs don't display true peak. You need a true peak-aware meter.
Youlean Loudness Meter (free standalone and plugin) shows both integrated LUFS and dBTP side by side. Logic Pro's built-in Loudness Meter plugin, Reaper's loudness analysis, and Audacity's loudness measurement all display true peak as well.
To check an already-exported file without opening a DAW: Youlean's standalone version reads WAV and FLAC files directly, no session needed.
If your true peak reads above -1 dBTP, your track is at risk of audible distortion on consumer DACs and after streaming transcoding — even if the sample-based peak meter showed a safe reading.
What Mastering Needs to Handle This
A standard brickwall limiter prevents stored sample values from exceeding 0 dBFS. It does not prevent intersample peaks. A limiter without true peak mode can deliver a file that measures at -0.1 dBFS on a normal meter but has a true peak of +1 or +2 dBTP.
Proper mastering for AI music uses a true peak limiter: a limiter that operates on an oversampled version of the signal and prevents the reconstructed waveform — not just the stored samples — from exceeding the target ceiling. The result is a file that genuinely stays within the chosen limit after D/A conversion.
This is one of the subtle but audible places where professional mastering of AI music diverges from simply applying a brickwall and calling it done. Loudness (LUFS) and true peak (dBTP) are separate measurements, and both need to be handled correctly for a clean-sounding release.
At antiaimaster.com, true peak control is part of every mastering pass — the ceiling that streaming transcoding requires is applied before the file ever leaves the processor.
The Two Numbers to Check Before Distribution
Before distributing an AI track, verify two values:
- Integrated LUFS — target around -14 LUFS for most streaming platforms (platforms normalize to their own target anyway, but extreme loudness still affects quality)
- True peak — must not exceed -1 dBTP
If your export fails the true peak check, the fix is straightforward: reduce the overall output ceiling by 1–2 dB and re-run the limiter with true peak mode enabled. What you need is genuine headroom in the reconstructed waveform — not just a sample value that stays below 0.
The difference between a track that sounds clean across devices and one that crackles subtly on phone speakers often comes down to this one number that most meters don't even show.