True Peak

Definition

The true peak is the estimate of the maximum value of the analog signal reconstructed from the samples. The method is defined by ITU-R BS.1770-4, Annex 2:

1. the signal is oversampled ×4 (for a 48 kHz file the working rate becomes 192 kHz; at 96 kHz ×2 is enough, at 192 kHz ×1);
2. the oversampling uses a linear-phase polyphase FIR interpolation filter (in the recommendation's example, 48 coefficients, 12 per phase);
3. the absolute value of the interpolated samples is taken and its maximum is computed;
4. the result is expressed in dBTP: TP = 20 · log10(max |x_os[m]|), with 0 dBTP equal to full scale.

The interpolation reconstructs the points between the original samples, where the waveform can exceed the value of the samples themselves. The residual underestimate at ×4 is at most 0.69 dB, in the worst case of a sine wave near the Nyquist frequency (the highest point can lie at π/8 of phase from the nearest interpolated sample: 20 · log10(cos(π/8)) = −0.69 dB). In musical material the underestimate is a few hundredths of a dB. The true peak is always greater than or equal to the sample peak.

Parameters and units

• Unit: dBTP (decibels true peak, referenced to full scale).
• Oversampling factor: ×4 for 44.1/48 kHz in compliant meters; many limiters also offer ×8, ×16 or ×32 to reduce the underestimate further.
• Delivery limits: EBU R128 sets a maximum of −1 dBTP; the AES TD1004 document for streaming indicates −1 dBTP; for very loud masters (louder than −14 LUFS) destined for lossy encoding, −2 dBTP is common practice.
• Limiter ceiling: −1.0 dBTP (streaming), −0.3/−0.5 dBTP only for media without lossy encoding (CD, lossless files intended for download).
• Indicators: maximum value of the song (TP max) and, in advanced meters, the number and position of overs.

Musical function

The true peak protects the master from distortion that occurs after mastering: in the D/A converter, in sample-rate conversion and especially in lossy encoding (AAC, MP3, Ogg Vorbis, Opus). Codecs split the signal into bands and rebuild it with phase and amplitude errors; in dense, bright masters the decoded peaks exceed those of the original by 0.5–1.5 dB. A true peak below −1 dBTP keeps the margin needed so that the decoded file does not clip in the player.

Practical example

A 3:20 pop master, analyzed before delivery:

Before the correction, AAC encoding at 256 kbit/s produced decoded peaks of +1.8 dBTP, that is, clipping in the player. With the limiter set to a ceiling of −1.0 dBTP, ×4 or higher, the file loses only 0.2 LU of loudness and the decoded file stays below 0 dBTP.

Common mistakes

• Setting the limiter ceiling to −0.1 dBFS in sample-peak mode and considering the master compliant.
• Measuring the true peak on an excerpt: the maximum peak is often in the final chorus or on the hit that follows a break.
• Applying further processing after the limiter (EQ, dither, sample-rate conversion, fade) without measuring again.
• Using a meter with oversampling below ×4 and comparing its value with that of a compliant meter.
• Considering a −1 dBTP true peak sufficient for any bitrate: at 128 kbit/s and below, the overs after encoding can exceed 1 dB.

How to check it

• App analysis: the True peak (×4, BS.1770) row is calculated with ×4 oversampling and a linear-phase FIR filter; the Inter-sample peaks above 0 dBFS row counts the interpolated points that exceed full scale. A sample peak above −1 dBFS always implies a true peak above −1 dBTP.
• Dedicated meters: Youlean Loudness Meter, FabFilter Pro-L 2 (true peak and loudness meter), iZotope Insight; all BS.1770-compliant.
• Encoding test: export an AAC or MP3 from the master, reload the decoded file and measure the true peak again; the value must stay below 0 dBTP.
• PLR: the app also calculates PLR (true peak − integrated LUFS), useful for seeing how much the limiter has reduced the dynamic margin.