True peak explained: why masters clip on cheap systems

Your limiter says the master never crosses 0 dBFS. Then you play it on earbuds from a phone and the drop crackles. Nothing is broken — your meter and the real world are measuring two different signals. This is the inter-sample peak problem, and it's why mastering engineers work in dBTP, not dBFS.

Samples are dots. Audio is a curve.

A digital file doesn't store a waveform — it stores snapshots of one, 44,100 or 48,000 times a second. Playback hardware has to reconstruct the continuous curve that passes through those dots, and sampling theory says that curve is unique and exact for any signal below half the sample rate. Here's the catch: the reconstructed curve doesn't have to pass through its highest point at a sample. Between two samples that both read just under full scale, the true waveform can arc above full scale.

A sample-peak meter only looks at the dots, so it happily reports −0.1 dBFS while the actual analog waveform your converter must produce swings past 0 dB. Those overshoots are called inter-sample peaks, and heavily limited program material — exactly what a club master is — generates them constantly, because limiting flattens the samples up against the ceiling while the underlying curve keeps trying to swing through them.

What true peak measures

True peak (written dBTP) estimates the level of that reconstructed waveform instead of the raw samples. The method, standardized in ITU-R BS.1770, is to oversample the signal — commonly 4× — so the meter can see between the original dots, then report the highest value found. A track can read −0.1 dBFS sample peak and +0.8 dBTP true peak at the same time. The second number is the one the physical world cares about.

Where the clipping actually happens

An over-full-scale true peak isn't automatically audible — it becomes distortion at specific choke points:

• Digital-to-analog converters. A well-designed DAC has analog headroom above digital full scale; a cheap one in a phone, dongle, or Bluetooth speaker often doesn't. The reconstruction stage clips, and it clips hardest on the loudest, brightest transients.
• Lossy encoders. This is the big one for producers. When Spotify, YouTube, or Apple transcode your WAV to AAC or Ogg, the codec re-synthesizes the waveform from a frequency-domain approximation. Peaks routinely come out higher than they went in — commonly fractions of a dB, sometimes more on dense material. A master limited to exactly 0 dBFS is over full scale after encoding, and every decode clips a little.
• Sample-rate conversion and DSP. Resampling from 44.1 to 48 kHz, Bluetooth transport, and "enhancement" DSP in consumer gear all re-interpolate the waveform and inherit the same overshoot.

That's why the crackle shows up on the cheap system and not in your studio: your interface has the headroom to swallow the overs, the €20 earbuds' decoder chain doesn't. The distortion was latent in the file all along.

The −1 dBTP convention

The standard defense is simple: leave a margin. Broadcast loudness standards (EBU R128, among others) specify a maximum of −1 dBTP, and streaming platforms recommend the same — Apple's mastering guidelines and Spotify's docs both point at −1 dBTP, with Spotify suggesting −2 dBTP for very loud masters, since hotter program material produces bigger encoding overshoots. For club masters delivered as WAV to a DJ, the codec risk disappears but the cheap-DAC and resampling risks don't, so the convention holds: ceiling at −1.0 dBTP is the safe, standard call for a loud dance master.

True-peak limiting vs. sample-peak limiting

Not every limiter protects you. A classic sample-peak limiter guarantees no sample exceeds the ceiling — which, as above, guarantees nothing about the reconstructed waveform. A true-peak limiter oversamples internally, detects the inter-sample overshoots, and controls gain against those, usually with lookahead so it can turn down smoothly just before the peak instead of clipping at it. The audible difference at normal settings is nil; the difference after an AAC encode is the whole point.

Two practical notes:

• Order matters. The true-peak limiter is the last dynamic stage. Anything after it — even a well-meaning output trim implemented before a dither stage, or a clipper "for glue" — can undo the guarantee.
• Verify on the render. Meter the exported WAV, not the DAW playback. Plugin delay compensation, freeze paths, and render-time oversampling differences can shift peaks slightly between what you monitored and what you bounced.

What this means for your workflow

1. Mix with headroom and no limiter on the bus — that's a premaster preparation question.
2. Master to your loudness target with a true-peak limiter as the final stage, ceiling −1.0 dBTP.
3. Check the rendered file with a BS.1770 true-peak meter. If it reads over the ceiling, something downstream of the limiter is touching gain.

Do that, and the version of your track that survives Bluetooth, phone DACs, and every codec in the chain is the version you actually signed off on.