True Peak vs Sample Peak: Why "-1 dB" Still Clips

Two Different Peaks

A digital file stores the waveform as separate points, the samples. Sample peak is simply the biggest of those points. It's what a basic peak meter and "Normalize to -1 dB" use.

But you never hear the points. A converter, or an MP3 decoder, rebuilds a smooth continuous wave through them, and that wave can rise between two samples, above both. That's an inter-sample peak. True peak (dBTP) estimates it by oversampling the signal, usually 4x as in the ITU-R BS.1770 loudness standard, and taking the peak of that finer version.

The overshoot is largest on dense, bright, fast-changing material: transients, distorted synths, clipped masters. In other words, the material electronic music is made of.

What We Measured

Every PhaseForge pack ships an 8-bar demo beat mixed from the pack's own drums. Our first version capped each demo at a sample peak of 0.89, which is -1.01 dBFS. That looks like a safe "-1 dB". Our quality gate measures true peak with 4x oversampling, and five of the twelve demos failed it:

Of the other seven, three hit the same -1.01 dBFS cap and passed by a hair, at -1.002 to -1.007 dBTP. The other four were quieter and never reached the cap. Same setting, same kind of material, and the overshoot still ranged from almost nothing to a quarter of a dB. A fixed sample-peak margin passes some files and fails others, which is exactly why it isn't a safe rule.

This wasn't a one-off. On 4 October 2026, before we fixed it, the same mistake caused every QA failure in three of our older packs: they had been normalised to sample peak, and their inter-sample peaks sat above -1 dBTP.

Why a Few Tenths of a dB Matter

• Lossy encoding adds more. MP3 and AAC change the waveform, and a file that peaks right at the ceiling can clip after encoding. That's why streaming services publish true peak targets, not sample peak targets.
• Samples get stacked. A one-shot that's already at the ceiling leaves no room when it's layered, pitched or saturated in a sampler.
• Cheap converters clip audibly. Phones and budget interfaces don't always have headroom above 0 dBFS for inter-sample overs.

The Fix

Measure and set peaks in dBTP, with an oversampling meter. Then:

• Samples and stems: as the last processing step, apply a gain change so true peak lands at about -1 dBTP. Gain only, no limiter. We use -1.1 dBTP for a small margin, and the five rebuilt demos now measure exactly that.
• Masters: use a true-peak limiter (most modern limiters have an ISP or "true peak" mode) with the ceiling at -1 dBTP. Use -2 dBTP if the master is very loud and headed for lossy streaming.
• Check after export, not just on the master bus. Sample-rate conversion and dithering on export can move the peak.

Normalising to "-1 dB" in an editor that only reads sample peaks is the one habit to drop. It's the exact setting that failed above.

Quick Checklist

• Sample peak ≠ true peak. Inter-sample peaks can sit a few tenths of a dB higher.
• Meter in dBTP with oversampling (4x or more).
• Files: true-peak normalise to about -1 dBTP, gain only, as the very last step.
• Masters: true-peak limiter, -1 dBTP, or -2 dBTP for very loud masters.
• Re-check the exported file, not the session.