What is true peak?

True peak in plain English

A digital file holds a list of numbers, called samples, that measure the level of a wave at regular moments: 44,100 times a second at 44.1 kHz. When you play the file, the converter in your interface or phone turns those numbers back into a smooth, continuous wave, and that wave has to pass through every sample.

It doesn’t have to peak on one. If the real top of a wave falls between two sample moments, both samples land a little way down its sides, and the smooth curve the converter draws rises above them. Those hidden tops are called intersample peaks. The highest point of the rebuilt wave, wherever it falls, is the true peak.

An ordinary peak meter only reads the samples, so it reports the highest number in the file. That’s the sample peak, measured in dBFS. A true peak meter reports how high the wave itself goes, in dBTP. The two readings are often equal, and the true peak is never lower.

This only becomes a problem near the top of the scale. If the samples of a loud master sit at −0.1 dBFS and the wave between them reaches +0.5 dBTP, the file is fine, because it stores every sample correctly. The trouble starts when something later has to deal with a wave that goes past full scale and has nowhere to put it.

Reading the lab

The three cards at the top choose what sits at the end of the chain: no limiter, a limiter that watches the samples, or one that watches the wave between them as well. Gain turns the loop up before it reaches the limiter, and Ceiling sets the level the limiter holds the peaks under. With no limiter, anything the gain pushes past 0 dBFS is cut off when the file is exported, the way a 24-bit WAV would cut it.

The big screen is a close-up of 25 samples, about half a millisecond, around the highest true peak anywhere in the loop. It’s zoomed in on the top 6 or 7 dB of the wave. The pink dots are the samples the file stores and the pale line is the wave rebuilt from them, with a white dot on its highest point. Wherever that line rises above the highest dot, the gap is what a sample peak meter misses.

Next to it, SP is the sample peak and TP is the true peak, both for the whole loop, and I is the integrated loudness in LUFS. Under the close-up is the whole loop, with an amber box showing where the close-up is looking. The panel beside it describes what’s happening, using numbers measured from the loop you’re hearing, and counts the separate places where the wave goes over 0 dBTP or over the limiter’s ceiling.

A/B switches to the untouched loop at the same loudness, which is the fair way to check whether a limiter changed the sound. You’ll find that a sample peak limiter and a true peak limiter on the same settings sound almost identical on your speakers. The difference shows up later, in converters and encoders.

Sample peak vs true peak

On the bass loop in the lab, the two readings are half a decibel apart before you touch anything. Its highest sample is at −0.5 dBFS and its true peak is at 0.0 dBTP, so the headroom the channel meter shows doesn’t exist. Normalise the same loop so its highest sample sits at −0.1 dBFS and the true peak reaches +0.4 dBTP.

How does a true peak meter work?

A meter can’t ask the converter what it’s going to do, so it does the same maths itself. It runs the audio through oversampling: it works out new points between the existing samples, using the same kind of smooth interpolation a converter uses, and looks for the highest one. The international loudness standard, ITU-R BS.1770, describes the method: raise the sample rate at least four times, from 48 kHz to 192 kHz, and read the highest point. Two meters that follow it should agree to within a fraction of a decibel on the same file.

The figure below shows the most extreme simple case: a sine wave at a quarter of the sample rate, which gets only four samples per cycle. Slide the timing and the samples can miss the crest by as much as 3 dB, reading −3.0 dBFS while the wave itself reaches 0.

Four points per sample doesn’t catch everything. For a sound right at the top of the frequency range, where each cycle lasts barely two samples, the nearest of the four points can still miss the crest by about 0.7 dB. On music that’s rare, and it’s one of the reasons the usual advice leaves a full decibel of margin instead of aiming for exactly 0 dBTP.

The lab’s meter uses the same four-point method. The pale line in its close-up is worked out at 16 points per sample so the curve looks smooth, which is why the white dot can occasionally sit a hair above the TP reading.

Do intersample peaks matter?

Inside your DAW, they do no harm. The file holds every sample correctly and plays back cleanly. Problems start when the audio goes through a process that rebuilds or reshapes the wave and has to fit the result back under full scale.

Digital-to-analogue converters

Most modern converters oversample the signal digitally before turning it into voltage, and that digital stage calculates the intersample peaks for real. If it was designed with no room above full scale, a true peak of +1 dBTP has nowhere to go and gets clipped inside the converter. Some converters have headroom built in for this and some don’t, and you have no say over what your listeners use.

MP3, AAC and Ogg Vorbis

Lossy encoders shrink a file by throwing away detail, including the highest frequencies, and when the file is decoded the wave comes back a slightly different shape. Its peaks can land higher than the ones in your master. Most streaming services send lossy copies of what you upload, at least on their standard quality settings, so nearly every release goes through an encoder. Spotify’s guidance is to keep the true peak under −1 dBTP, and under −2 dBTP for masters louder than −14 LUFS, because those are the ones most likely to distort when they’re encoded.

Sample rate conversion

Turning a 48 kHz file into 44.1 kHz means working out a whole new set of samples, and some of them land closer to the real tops of the waves than the old ones did. A file whose samples peaked at −0.1 dBFS can come out of the conversion with samples above 0, which a fixed-point file can’t hold.

Turning quiet tracks up

Spotify raises quiet tracks towards its playback level only as far as their peaks allow. A track with stray intersample peaks gets less of a lift than its loudness alone would give it. The streaming loudness lesson covers how each service handles this.

What causes intersample peaks?

Sharp corners in the waveform push true peaks up. A smooth, slow wave doesn’t have much room to rise between two samples, while a wave with steep edges and lots of high-frequency content can overshoot them by a decibel or more. Clipping is the worst offender, because it turns rounded peaks into flat tops with sharp corners at each end, and the rebuilt wave rings above and below those corners.

The lab shows how big the effect can be. Push the bass 6 dB into the export with no limiter and 0.3% of its samples get cut off at 0 dBFS. The sample peak meter can’t read higher than 0.0 after that, so it looks as if nothing worse happened, but the true peak is at +2.0 dBTP and the wave goes over 0 in 131 separate places.

Hard limiting does the same thing more gently. A limiter that clamps down within a fraction of a millisecond reshapes the waves around each peak, and a clipper used to shave transients before the limiter adds its own flat tops. Both are standard on loud masters, which is why loud masters are where true peak problems show up.

The sound itself matters too. With the same sample peak limiter on the same settings, the full mix in the lab ends up with its true peak 0.3 dB above the highest sample, while the bass loop’s gap is 1.4 dB. So measure each master on its own, even if a setting left the last one clean.

Sample peak and true peak limiters

A limiter holds the level under a ceiling, and the question is which level it reads. A sample peak limiter checks the samples. Set it to −1 dB and no sample will ever pass −1.0 dBFS, while the wave between the samples is free to go higher.

In the lab, the bass pushed 6 dB into a sample peak limiter at −1 dB comes out with every sample at or under −1.0 dBFS. Between the samples the wave crosses the ceiling in 82 places, and the true peak lands at +0.4 dBTP. A channel meter would tell you this master is a full decibel below 0.

A true peak limiter oversamples the signal it’s listening to, the same way the meter does, and turns the sound down early enough to hold the rebuilt wave under the ceiling. Most also use lookahead, a short delay that lets them see a peak coming. Switch the lab to the true peak limiter and the true peak sits at −1.0 dBTP with nothing crossing the line.

It costs a little loudness, because the limiter now has to turn down the hidden peaks as well. On the bass, the loudest hit comes down 7.1 dB instead of 6.5 and the loop ends up at −11.7 LUFS against −11.5. That’s 0.2 LU for a master with a full decibel of room left for the encoder.

Most mastering limiters have a true peak switch. FabFilter Pro-L 2 has one, and so does the Maximizer in iZotope Ozone. If yours has no such option, set the ceiling lower, around −1.5 dB, and check the result on a meter that reads dBTP. The free Youlean Loudness Meter shows true peak alongside LUFS. The limiting lesson covers the rest of what a limiter does.

What true peak ceiling should I use?

For a master you’re releasing, −1 dBTP with the limiter in true peak mode. It leaves room for converters and for most encoding, and it’s what Spotify asks for and the maximum in the EBU R 128 broadcast standard. If your master is louder than −14 LUFS integrated, go to −2 dBTP, as Spotify recommends, since streaming services turn loud tracks down anyway and the extra decibel costs very little.

Dropping from −1 to −2 dBTP has a cost. On the full mix in the lab, the same settings read −10.9 LUFS at −1 dBTP and −11.6 LUFS at −2, so the extra decibel of safety takes 0.7 LU off. With its default normalisation, Spotify turns both versions down to −14 LUFS, so they end up playing at the same level there.

How to check and control true peak, step by step

1. Put a true peak meter last on the masterAny BS.1770 loudness meter will do. Make sure the reading you’re watching says dBTP, since some meters show sample peak in the same spot.
2. Switch the final limiter to true peak modeLook for a button labelled True Peak or TP. Leave limiters earlier in the chain as they are.
3. Set the ceiling to −1 dBUse −2 dB if the master reads louder than −14 LUFS integrated.
4. Set the loudness with the limiter’s gainPush in until it sounds right, comparing with the limiter bypassed at matched loudness.
5. Watch the meter through the loudest sectionThe true peak should stay at or just under the ceiling. If it goes over, something after the limiter is adding level, or its true peak mode is off.
6. Measure the exported fileLoad the bounce into the meter on its own. If you change sample rate or bit depth on export, this is the reading that counts.

Common true peak mistakes

Trusting the channel meter

A DAW’s channel meter usually reads samples. A master that never lights its red clip light can still have true peaks above 0 dBTP, so check the true peak figure on a loudness meter before you export.

Normalising to 0 dBFS

Normalising sets the highest sample to the level you choose, so a file normalised to 0 dBFS or −0.1 dBFS will often have true peaks above 0. The bass in the lab reaches +0.4 dBTP after normalising to −0.1. If you normalise at all, normalise to −1 dBFS or lower and check the true peak afterwards.

A −0.1 dB ceiling in sample peak mode

That setting holds the samples just under full scale and does nothing about the wave between them, which is exactly how intersample overs get through. Use true peak mode, or if your limiter doesn’t have it, a lower ceiling and a true peak meter to check it.

Only measuring the session

Dither, sample rate conversion and anything on the master after the meter can change the peaks a little on the way out. The exported file is what listeners get, so that’s the one to measure.

True peak limiting on every bus

Only the final ceiling needs it. A limiter on a drum bus is there to shape the drums, and inside a floating-point DAW, levels between plug-ins can go above 0 without harm. The gain staging lesson explains why.

Questions people ask

Keep going