True Peak Limiting is a method by which a limiter adjusts for how the digital waveform will be reconstructed by playback systems which can result in actual peak levels above 0dB even when the digital peak level is technically shown at below 0dB.

Implied Curvature

Basically the way intersample peaks occur is that the quantization points of a digital waveform can at times imply a curve between them which goes up a little bit higher than those actual sample points, resulting in a louder-than-expected actual peak level coming though.

To imagine it, just draw two dots in your mind, and then instead of drawing a straight line between them, draw a slightly curved line. The two dots are at a height of zero, and the curved line connecting them is bumping a little bit above zero. See?

Interpolated Quantization

When your speaker plays back audio, its motion is not quantized between the sample points like pixels—the way the digital audio is shown when you zoom in real close—rather, it’s a continuous motion between points. Think about it: the speaker membrane has to pass through physical space to get from one point of contraction-expansion to the next. The waveform’s up-and-down shows how the speaker cone pulses out-and-in. If the audio conversion (whether from WAV to mp3 or through a playback DAC) attempts to restructure a curve which ends up going above 0dB in this manner, you can get some distortion as the waveform curvature is lopped off (just like chopping the top of a hill to make it flat). If on the other hand you have provided enough headroom to account for the extra-bumped-up waveform curve, it will play back sounding most natural, but with a higher measured peak level than you may have expected.

Imaging filling in a dot-to-dot puzzle of a sine wave that has fairly widely-spaced dots. Your hand will naturally draw in the sine shape even where there are no dots. If a waveform is moving up, then at its top goes straight across for one or more samples, then goes back down, the actual speaker movement may go up, then go up a little bit more to formulate a nice arc before it comes back down. ISP Limiters try to predict this movement and stop it before it occurs. Aggressively (non-isp) limited mixdowns can have over 2dB of secretly hidden peaks. Crazy! Your favorite non-isp limiter might not be actually curtaining playback peaks as much as you think it is, but maybe that’s why it sounds so good. If you want to push a limiter hard, do it because you like the character doing so provides.

TP for my Waveform

Because of the increase in awareness of the occurrence of this intersample (aka true) peaking in the past decade, True Peak (aka ISP) Limiters have become more of a thing. Intersample Peaks themselves are certainly not new, as many tracks from CDs as far back as the 1980s contain them.

Some limiter plugins have ISP limiting as an additional option which acts like an added final final limiter process above and beyond the original. Some limiters have ISP detection built-in to their main algorithm, and always react to them to a degree. Some limiters don’t advertise themselves as ISP Limiters, yet nonetheless upon technical analysis are shown to do a better-than-average job at curtailing True Peaks.

Nothing Is Free!

It has become common for the final limiter in mastering to be of the “true peak” variety. These limiters will limit the audio as if it’s already been converted to a physical waveform, greatly reducing the possibility of unexpected clipping above 0dB. However, they don’t magically do this for free. The side effect is that your transients get even more squashed—even more abruptly shaven—even more heavy-handedly curtailed. Unfortunately, pretty much every true peak limiter i’ve tested handles things less cleanly than its non-ISP counterpart (as seems intuitive upon reflection, given the fine-grained detail levels they are working at).
This means a likely increase in an aurally unpleasant, rough, gritty, distorted quality the more the limiter is working. Altering your timing parameters carefully of course can help alleviate this. While this distortion is often on the subliminal level, i have found some (let’s face it, pretty much all) ISP Limiters to be surprisingly rough at relatively tiny amounts of gain reduction (less than 1dB), really crunching up transients in a not-good way.

Safety Net

True Peak Limiting could be extremely valuable for realtime processes, such as a live radio show, to absolutely guarantee no overs when required. They can also be useful during mastering to ensure that the peak level remains low enough for upload to streaming services, etc.

Not Necessarily Necessary

However, i have noticed that using an ISP Limiter during mastering as a default action is not always the best idea. Since pushing a master to be louder than everyone else no longer serves any useful purpose, instead of sacrificing the natural quality of our loudest transients by True Peak Limiting them, an alternative method is to render our song, find out where the True Peak Level happens to be, and then re-render our track, simply lowering the final output by as many dB as needed to keep our ultimate peak level exactly where we want it.

For example, say i master a song, it has been approved by the client and everyone is 100% happy with the balance and dynamics, and upon rendering and scanning it, i find out that it is reading peaks of +0.5dB on both the left and right channels. Now say we want to release it on a platform that requires a true peak level of -0.1 or lower. I have two options in front of me, depending on my desires:

DESIRE A: Loudest Possible Mix
Use an ISP Limiter to squish the track by 0.6dB. Our mixdown’s TP reading will now be -0.1 and the track will be as loud as it was. But it might be a little bit more distorted on the loudest transients than it was previously.

DESIRE B: Highest-Quality Possible Mix
Render the mix again, simply lowering the final output level (post-limiter) by 0.6dB. Our mix will have the exact balance everyone loved, there will be no added distortion, and the loudness will be decreased by 0.6dB.

Which of these two disparate goals is more honorable? Considering that LU-normalized playback is becoming more and more common nowadays, and for ultimate archive quality, i definitely say option B is the one to go for. Making your song 1dB louder is not going to be noticed by anybody. And if you really want it louder (aka “denser”), it’s usually more effective to limit parts in the mix individually anyways, so that one stray too-loud transient by one instrument doesn’t punish everything else.

Hope For the Future

For my last few mastering projects, i tried to use as little ISP Limiting as possible, instead carefully adjusting other processors and analyzing renders to adjust for the target LU and ISP levels, and i think i’ve achieved a slightly cleaner and more solid sound than previously.

Since ISP Limiting is a fairly new and esoteric art, i’m giving it the benefit of the doubt that some vastly more tasteful true peak limiting algorithms will be released eventually than those that are currently on offer. Just like the first generation of digital limiters sucked pretty bad, but now they’re getting fabulously excellent. Or maybe not. Maybe it’s just something that can’t really be improved due to the laws of nature. I doubt that, since human implementation of physics is continually pushed farther than each previous generation conceives possible, but in the meantime, i’ll stick with lowering my output level a tiny amount to match true peak targets rather than adding unbeneficial distortion, when i can manage to get away with it and still hit loudness targets.

That’s all for today. See you again soon. Peace!