You bounced a mix that never once kissed 0 dBFS, uploaded it, and somewhere between your hard drive and someone's earbuds it started spitting crackles. Your meter said the file was clean. It was — just by the wrong measurement.

The short answer: True peak measures the real analog level your samples will reconstruct into, which can climb higher than any single sample. Master with your limiter's ceiling at −1 dBTP so those inter-sample peaks and the gain that lossy codecs add still have headroom to land without clipping.

What a digital sample actually is

Your audio file isn't a continuous waveform. It's a string of measurements — 44,100 of them per second at CD rate — snapshots of where the wave was at each instant. The smooth analog curve only exists again after a D/A converter (or a codec's decoder) reconstructs it by drawing a curve through those dots.

Here's the catch: the reconstructed curve doesn't have to stay at the level of the dots. It passes through them, but between two samples it can bulge upward. If two adjacent samples both sit at −0.1 dBFS on a steep waveform, the actual peak of the reconstructed curve between them can reach +0.6 dBFS — over full scale — even though no individual sample ever read above −0.1.

That bulge is an inter-sample peak, and a standard sample-peak meter is blind to it. It only reads the dots. A true-peak meter does what the converter does: it upsamples the signal (typically 4×) to estimate where the reconstructed curve actually goes, then reports the highest point. That's the dBTP number.

Samples can all sit under 0 dBFS while the real waveform between them overshoots
Samples can all sit under 0 dBFS while the real waveform between them overshoots — leave a dB of room.

Why the overs cause real damage

When the reconstructed peak exceeds 0 dBFS, two things can clip it:

  • The D/A converter in a phone, laptop, or cheap interface. Many have little or no headroom above full scale, so the over gets squared off into audible distortion on playback.
  • Lossy encoders. MP3, AAC, Ogg, and the codecs Spotify, Apple Music, and YouTube use all reshape the waveform during encoding. That reshaping routinely adds level — an encode can come back 1 to 1.5 dB hotter at the peaks than the file you fed in.

A master sitting right at 0 dBFS sample-peak can clip twice: once from the codec pushing it over, again from the listener's converter choking on the result. Set the ceiling at −1 dBTP and you've reserved a full dB of room for both.

The number, and how to dial it in

The practical recipe is short:

  1. Put a true-peak-capable limiter last in your master chain (FabFilter Pro-L 2, Ozone Maximizer, Waves L2 in TP mode, or any limiter with a "true peak" / "ISP" switch — turn it on; it's often off by default).
  2. Set the output ceiling to −1.0 dBTP.
  3. Drive your gain or threshold to hit your loudness target. For streaming, −14 LUFS integrated is the standard reference; most platforms normalize to somewhere between −11 and −16 LUFS, so landing around −14 keeps you from being turned down and avoids wasted headroom.
  4. Bounce, then verify the rendered file in a separate true-peak meter (Youlean Loudness Meter is free and reads both LUFS and dBTP). Trust the render, not the limiter's live display — they can diverge.

That −1 dB isn't superstition. It's roughly the worst-case gain a transparent lossy encode adds, plus a small cushion for converter headroom. And one dB of loudness is genuinely inaudible as "quietness" — nobody has ever A/B'd two masters and picked the louder one because it was 1 dB hotter at the true peak.

Studio control room
A treated control room buys you honesty. Most of us are chasing that honesty in software instead. (Photo: Unsplash)

The myth that costs people clean masters

This is the one that gets producers burned, because their DAW actively reassures them they're fine.

Myth: "My DAW's meter never touched 0 dBFS, so my master is safe for streaming." Wrong. Your DAW's main meter is a sample-peak meter — it reads the dots, not the reconstructed curve between them. Sample-clean files produce inter-sample overs all the time, and the codec stage adds even more on top. You need a true-peak meter to see what the sample meter physically cannot.

If the only number you've ever checked is the one on your master fader, you've been flying blind on the metric that actually matters at delivery.

The other camps

−1 dBTP is the safe default, not a law of physics. Reasonable engineers land elsewhere depending on where the music is going.

  • Lossless / high-res delivery (−0.3 dBTP). If you're delivering WAV or FLAC that won't be re-encoded — vinyl pre-masters, lossless tiers, archival — there's no codec adding gain, so you can sit much closer to the ceiling. −0.3 to −0.5 dBTP reclaims that headroom for loudness with no real risk.
  • Cautious mastering houses (−1.5 to −2 dBTP). Engineers mastering loud material, or anyone hedging against the worst encoders, go further. Spotify's own guidance says if you master louder than −14 LUFS, target below −2 dBTP — hotter masters leave codecs less room and provoke larger inter-sample overs. The louder you push, the more margin you owe.
  • The sample-peak-margin minority. A few engineers skip true-peak limiting entirely and leave generous sample-peak headroom — ceiling at −3 dBFS sample, no ISP processing. It works in practice because the margin absorbs the overs, but you're guessing at the cushion instead of measuring it. I'd rather know the number.

The thread connecting all of these: the riskier the destination (lossy, loud, unknown playback gear), the more headroom you reserve. −1 dBTP sits in the middle of that range for a reason — it covers streaming without throwing away audible loudness.

One honest caveat about true-peak limiting

Clamping inter-sample peaks isn't free. To catch a peak that only exists between samples, the limiter has to oversample and react to a level no individual sample contains — and aggressive true-peak limiting can soften transients or add faint pumping that sample-peak limiting wouldn't. On a punchy drum master, listen for it.

The fix is usually to limit a touch less hard and let the −1 dB ceiling do its job, rather than slamming into a −0.1 dBTP ceiling and asking the limiter to perform miracles. Headroom is cheaper than processing artifacts.

Set the ceiling, enable true peak, and verify the rendered file — not the live meter — before you call it done. The −1 dBTP master that measures clean after encoding will always beat the louder one that crackles on someone's phone, and the only way to know which you've made is to check the meter your DAW doesn't show you by default.