True peak vs sample peak: why 0 dBFS can still clip.

What happens between the samples

A digital file is a list of numbers, one per sample. A sample peak meter reports the largest one: if the loudest sample is 0.891 of full scale, it reads −1.0 dBFS. That's accurate as far as it goes, but nobody listens to a list of numbers.

A digital-to-analog converter draws a smooth line through the samples, and so does every sample rate converter and every lossy decoder. That line doesn't have to peak on a sample. When the real peak falls between two samples, it can be higher than both. That's an inter-sample peak, and a sample peak meter can't see it.

The textbook case is a sine wave at a quarter of the sample rate, with its samples landing 45° away from the crests. Every sample reads 0.707 of the real height, so a sample peak meter shows −3.0 dBFS while the actual waveform peaks at 0.0. Music rarely goes that far, but heavily limited masters often show true peaks 0.5 to 1.5 dB above their sample peak.

How true peak is measured

ITU-R BS.1770, the standard that defines LUFS, defines true peak too. The meter oversamples the signal, 4× for 48 kHz audio, with a filter that fills in the values between the original samples, then takes the largest absolute value. The result is written in dBTP, decibels true peak.

Oversampling is an estimate, not a perfect reconstruction. Two conforming meters can differ by a few hundredths of a dB on true peak, and at very high frequencies a 4× meter can read slightly low. That's one reason the usual ceiling is −1 dBTP rather than −0.1. The Loudmeter analyzer uses 4× oversampled true peak and marks every moment that reads +0.1 dBTP or higher in red on the waveform.

Why true peak matters more than ever

• Lossy encoding. MP3, AAC, Ogg Vorbis and Opus throw data away and rebuild the waveform on playback. The rebuilt peaks usually land higher than the originals, and the louder and denser the master, the more they grow. Every streaming service encodes your file.
• Sample rate conversion. Going from 96 kHz to 44.1 kHz is a reconstruction too. Inter-sample peaks become real sample values, and those can pass full scale.
• Consumer converters. The converters in phones, cars and TVs often have little or no headroom above 0 dBFS, so an inter-sample over can clip in the analog stage.

None of this shows on a sample peak meter in your DAW. The master looks clean and the stream crackles. Spotify asks for −1 dBTP, or −2 dBTP for masters louder than −14 LUFS, for exactly this reason.

How to keep true peak under control

1. 01Use your limiter's true peak (or ISP) mode if it has one, and set the ceiling to −1 dBTP.
2. 02If the limiter only watches sample peaks, set the ceiling lower, around −1.5 to −2 dBFS, and check the result on a true peak meter.
3. 03Measure the bounced file, not the session. The export is what the platforms receive.
4. 04If one moment sets the true peak, fix that moment. A snare hit or a sharp consonant is often easier to clip or de-ess than limiting the whole track harder.

dBFS, dBTP and the numbers you'll see

• dBFS is decibels relative to full scale. 0 dBFS is the largest value a fixed-point file can store, and sample peak is measured in dBFS.
• dBTP is the same scale measured after oversampling. It can read above 0 in a file whose samples all sit below 0 dBFS.
• 32-bit float files can hold samples above 0 dBFS. They're fine inside your DAW, but they clip the moment you export to a 16- or 24-bit file.
• Clip counters in DAWs count samples at or above full scale. A master can show zero clips and still have true peak overs.

FAQ

Keep reading

Measure your file now.

Free, in your browser. Your file never leaves your device.