True peak is the maximum level of the continuous waveform that a digital audio signal represents, measured by estimating the values between the stored samples rather than reading the samples alone. In mastering, it is expressed in dBTP (decibels true peak) and is the reference used to prevent clipping in digital-to-analog conversion and lossy encoding.
The concept exists because a sample peak meter can report 0 dBFS or less while the reconstructed signal actually rises above full scale. These hidden excursions are called inter-sample peaks, and true peak measurement is the standardized way to detect them.
Why true peak differs from sample peak
A digital file stores discrete samples, but the audio that reaches a speaker is a continuous waveform reconstructed by the converter’s interpolation filter. When the waveform peaks between two samples, its maximum can be higher than any stored sample, so a sample peak meter underestimates the real level.
The difference is usually small, but heavily limited masters with dense high-frequency content can show inter-sample overs of 1 to 3 dB in extreme cases. Signals near a quarter of the sample rate, phase-shifted transients and square-like waveforms produced by hard clipping are the most likely to create these overs.
| Property | Sample peak | True peak |
|---|---|---|
| What is measured | Largest stored sample value | Estimated maximum of reconstructed waveform |
| Unit | dBFS | dBTP |
| Detects inter-sample overs | No | Yes, within the meter’s accuracy |
| Typical reading on loud masters | Exactly 0 dBFS or slightly below | Can exceed 0 dBTP |
| Used in delivery specifications | Rarely today | Standard in broadcast and streaming |
How a true peak meter works
A true peak meter oversamples the signal, typically by a factor of 4 at 44.1 or 48 kHz, using a low-pass interpolation filter, and then reads the absolute peak of the oversampled signal. The method is defined in ITU-R BS.1770, the same recommendation that specifies loudness measurement in LUFS.
Oversampling is an estimate, not an exact value. With fourfold oversampling the maximum under-reading is on the order of a few tenths of a decibel, which is why delivery specifications leave a margin below 0 dBTP. Higher oversampling factors reduce this error, and some meters use 8x or more for greater accuracy, at a higher processing cost.
| Oversampling factor | Typical context | Accuracy implication |
|---|---|---|
| 1x (none) | Sample peak meters | Misses inter-sample peaks entirely |
| 4x | Minimum in ITU-R BS.1770 at 48 kHz | Small under-reading remains possible |
| 8x to 16x | Mastering limiters and precision meters | Closer to the true analog maximum |
| 2x | Accepted for 96 kHz material | Higher base rate reduces the gap |
True peak limits in delivery standards
Most loudness specifications combine an integrated loudness target with a maximum true peak level. Broadcast standards are stricter because the signal may pass through several codec and processing stages before it reaches the listener.
| Delivery context | Loudness target | Maximum true peak |
|---|---|---|
| EBU R128 broadcast | -23 LUFS | -1 dBTP |
| ATSC A/85 broadcast | -24 LKFS | -2 dBTP |
| Music streaming (common guidance) | About -14 LUFS normalization | -1 dBTP, or -2 dBTP for very loud masters |
| Podcast distribution (common practice) | -16 to -19 LUFS | -1 dBTP |
Streaming platforms publish their own recommendations, and these change over time, so the delivery specification of each platform should always be checked. The consistent pattern is a ceiling of about -1 dBTP for ordinary material and more headroom when the master is very loud.
Lossy encoding and the need for headroom
Lossy codecs such as MP3, AAC and Opus remove inaudible components and reconstruct the signal from a reduced representation. This process changes the waveform and commonly raises peak levels, so a master that peaks at exactly 0 dBTP before encoding may clip after decoding.
The amount of overshoot depends on the bitrate, codec and density of the material. Low bitrates and heavily limited masters produce the largest overshoots. Leaving 1 to 2 dB of true peak headroom is the practical defense, and engineers often preview the encoded result to confirm that no clipping occurs.
Controlling true peak in mastering
Mastering engineers control true peak with a true peak limiter, which oversamples the signal in its detector and sometimes in its gain stage, then sets the ceiling in dBTP rather than dBFS. This is the last process in the chain, after EQ, compression and any loudness adjustment.
| Symptom | Likely cause | Fix |
|---|---|---|
| Meter shows above 0 dBTP | Sample peak limiter at 0 dBFS | Enable true peak mode, set ceiling to -1 dBTP |
| Distortion after streaming encode | No headroom for codec overshoot | Lower ceiling to -1 or -2 dBTP |
| Harsh, brittle top end | Heavy clipping before the limiter | Reduce drive, use gentler clipping |
| Delivery rejected for QC | True peak exceeds specification | Re-limit to the specified ceiling |
A common misconception is that true peak control alone makes a master safe. It addresses peak level only; loudness, dynamic range and tonal balance remain separate concerns, and loudness normalization will still turn an overly loud master down.
The simple explanation
Digital audio is stored as a series of dots, and the playback system draws a smooth line through those dots. Sometimes the line bends higher than any of the dots, like a rope swinging between two fence posts.
Ordinary meters only look at the dots, so they can miss these bulges. A true peak meter estimates the full line, which shows the real highest point. Keeping that point a little below the maximum prevents crackles and distortion when the music is played or converted for streaming.
How it is used in practice
A mastering engineer prepares an electronic dance track for streaming release. The sample peak meter shows the limiter output at exactly 0.0 dBFS, but the true peak meter reads +1.4 dBTP, because dense synthesizers and hard clipping have created inter-sample overs.
The engineer switches the final limiter to true peak mode with a -1 dBTP ceiling and previews an AAC encode at a typical streaming bitrate. The encoded version, which previously showed audible crackle on the snare, now plays cleanly.
The master loses almost no perceived loudness, because the change affects only the briefest peaks, and it passes the distributor’s quality check on the first submission.