Short answer: true peak is the real maximum level your signal reaches once it is turned back into a continuous analog waveform for playback, including the peaks that fall between the digital samples. Those inter-sample peaks can rise higher than any single stored sample, so a track that looks safe on an ordinary meter can still overshoot. True peak is measured in dBTP, decibels true peak.
The numbers below, like a -1 dBTP ceiling, are common starting points rather than strict rules, and they shift a little by genre and by platform. Levels are written in dBFS and dBTP, decibels measured from 0, the point where digital audio clips.
True peak is the real maximum level a signal reaches once it is converted back from digital samples into a continuous analog waveform, the form it takes when it actually plays. That maximum includes the inter-sample peaks: the points between the stored samples where the rebuilt waveform can rise higher than any single sample value.
Digital audio keeps the signal as a row of evenly spaced samples, snapshots of the level taken tens of thousands of times a second. On playback a converter joins those points back into one smooth, continuous wave. That wave does not step flatly from sample to sample, it curves, and between two samples it can bulge above the height of either one. The tallest point of that smooth wave is the true peak. A sample peak reads only the tallest stored sample, while true peak reads the tallest point of the wave those samples produce.
So the two readings answer slightly different questions. Sample peak asks how tall the biggest stored number is. True peak asks how tall the signal really gets once it is reconstructed for your ears. Most of the time they are close, but near the ceiling on loud, busy material the gap can be enough to matter.
dBTP stands for decibels true peak, measured from the same 0 ceiling where digital audio clips. A reading of -1 dBTP means the real, reconstructed peak sits 1 dB below that ceiling. It is written as a number below 0 just like dBFS, but it counts the inter-sample peaks a plain sample-peak reading in dBFS can miss.
They happen because the samples stored in a digital file are only snapshots, and the smooth waveform rebuilt from them on playback can swing higher between two snapshots than either one on its own. The peak lands in the gap the samples never measured.
When your file plays, a converter reconstructs the single continuous curve that passes through all the sample points. That curve can overshoot in the spaces between points, most often where the signal is loud and moving quickly close to the ceiling. Nothing was stored at the exact moment of that overshoot, so it does not show up as a sample. It only appears once the samples are joined back into a wave.
This is why an ordinary sample-peak meter can under-read. That kind of meter looks only at the stored sample points, reports the tallest one, and calls it the maximum. It never sees the taller bulge in between. So a meter that shows a peak sitting right at 0 dBFS can be hiding a true peak that is actually above 0 once the waveform is reconstructed, which is the level your listener's hardware has to deal with.
A clean-looking sample-peak reading is not proof you are safe. The meter measures the dots, not the line drawn through them. A master that a normal meter shows just kissing 0 dBFS can still hold inter-sample peaks above 0, and those are what clip on playback. To see them you need a meter or limiter reading true peak, in dBTP.
It matters because streaming almost always delivers your music in a lossy format like MP3 or AAC, and the encoding step can push inter-sample peaks even higher than they were in your master. If they cross the ceiling on the listener's device, they clip, adding a harsh distortion you never hear in your own session.
A lossy codec throws away detail it judges inaudible and rebuilds an approximation of the sound on playback. That approximation does not match your file sample for sample, and its own inter-sample peaks can end up taller than the ones you started with. So a master sitting right at 0 can come out of the encoder overshooting, and the converter on the listener's phone or laptop clips the overshoot. Because you cannot control which codec or which device is on the other end, the only reliable fix is to leave a little room at your end.
That is why loudness advice for streaming settles on a true-peak ceiling below 0 rather than mastering right up to the top. It is the same reason our guide on how loud a master should be pairs its loudness target with a -1 dBTP ceiling: the loudness number gets you heard at a sensible level, and the true-peak margin keeps the encoded file from clipping once it reaches a listener.
You control it with a true-peak limiter: a limiter switched to its true peak or inter-sample mode, set to a ceiling a little below 0. A common default when mastering for streaming is around -1 dBTP, which leaves roughly 1 dB of margin for the encoding stage to use without pushing a peak into clipping.
The distinction is in what the limiter watches. An ordinary sample limiter guards the stored sample values and can still let inter-sample peaks slip past it. A true-peak limiter oversamples the signal, so it sees the peaks that fall between samples and holds the real, reconstructed level under the ceiling you set. Most modern limiters and dedicated loudness meters offer this mode, sometimes labelled true peak and sometimes inter-sample. The practical steps are to switch it on, set the ceiling, and confirm your final meter is reading dBTP rather than plain sample peak.
Leave the margin rather than mastering flat to 0. If you deliberately push loudness harder, some engineers drop the ceiling a touch further, to around -2 dBTP, because louder and denser material tends to generate bigger inter-sample peaks and encodes less forgivingly. These are common starting points, not fixed rules, so treat them as a safe zone and adjust by ear and by target.
It also helps to see where true peak sits among the other readings you watch at the master, because it is easy to confuse them.
| Reading | What it measures | What it is for |
|---|---|---|
| Sample peak (dBFS) | the tallest stored sample value | a quick clip check, but it can under-read the real peak between samples. |
| True peak (dBTP) | the tallest point of the reconstructed wave, inter-sample peaks included | keeping the real level safe through encoding and playback. |
| Loudness (LUFS) | the average felt loudness over time | how loud the track sits next to other tracks. |
| Headroom | the gap between your loudest peak and the ceiling | room left over for the next stage of the chain to work in. |
In short, true peak and sample peak are both about the highest instant the signal reaches, but true peak counts the peaks between samples that sample peak misses. Loudness is a different measure entirely, an average of how loud the track feels rather than how tall its peaks are. And headroom is the space you keep above your peaks so the later stages, encoding included, have somewhere to go. You watch all of them at the master, but they are answering different questions.
True peak is a mastering check, the very last stage. It only matters once you have a mix worth mastering, and getting there is the real work. A limiter can catch an overshoot at the end, but a clean, controlled source is what keeps the trouble from starting in the first place.
The Collection is our three instruments together, ARGISH, SILT and REHEAT, with three separate licence keys. This page is about holding the real peak under control at the end, and that job starts easier when the sound going in is already well behaved: our instruments output at sensible levels with clean peaks, so your project keeps room to spare instead of arriving hot and fighting the ceiling. ARGISH is a self-playing chord synth, SILT is a tape-loop instrument, and REHEAT writes acid lines, and whatever you make with them is yours to release. Each one ships as AU, VST3, AAX and a standalone app on macOS, signed and notarized, and a VST3 on Windows.
You can hear one free in your browser first, no install and no account, then decide.
We write these when there is something worth writing down. One email when a new one lands or a new Tunary instrument ships. No newsletter, no schedule.
Tunary