Hi all,

I work in TV sound, and we're seeing more and more networks change over to the BS.1770 recommendations regarding average level and peak level. The issue of average level (-24 LKFS) isn't a big deal; we're using the Dolby Media Meter and keeping our mixes in the pocket.

The real problem is the peak limiting specs. Networks are now QC'ing levels in the dBTP domain (decibel True Peak), and we don't have any reliable limiters that function in dBTP. All the limiters we have at our disposal work in sample peak, and mixes are routinely being measured as "too hot" in dBTP terms.

Does anyone know of a truly reliable limiter plugin for Pro Tools that properly limits intersample peaks such that a meter like the Techtronix 7000-series shows you that you're within spec?

We've tried the iZotope Ozone, setting it to prevent intersample peaks, and we still had a show rejected for going over by 0.9dB repeatedly. So that one doesn't appear to really do the job.

What are others doing about this issue? Thanks for any thoughts/comments!!

Take a look at TC Electronics new LM2 loudness meter. It includes a feedforward "true peak" limiter to handle overshoot problems. You went 9/10 of a dB over .... crime of the century (LOL)!

Yeah, the TC Electronic LM2 looks great; but that is an expensive piece of gear! Do you know if they have a plugin version, or if anyone else has a limiter plugin that accomplishes the same thing?

Hello,

I feel for you, I had a hard time with a tv series with this issue. We tried all sorts of limiters and compressors and nothing worked. We eventually figured out that there is no way to prevent this with software. No matter what plug-in you use it will always redraw the waveform going over. Going out analogue will solve the problem but no one wanted to do that.

To make a long story longer (sorry, my brain is filling up with all that I went through on this) the only compressor fast enough to stop a quick peak is the C1 compressor. So every time I see a "Clip" indicator I get rid of the peak. Either with Audiosuite or insert on a track with the C1. I found foley and sharp fx was causing most of the overages.

Hope this helps.
Steph Carrier

Alexey,
If your consistent at 0.9 lower your threshold by 1 db and away you go.
Experiment and see. Do a test, see what works and let us all know what worked for you.

Thanks everyone for your responses! We're already operating along the lines that Marti suggested:

Up to this point our work-around has been to simply limit more aggressively. For instance, if a show's spec is -6dB True Peak, we might set our Waves L2 limiter to -7.5dB. Unfortunately there are still two main problems with this for us:

1) dBTP is a moving target, and sometimes an extra 1.5dB of sample-limiting isn't even enough. I've personally measured dBTP peaks that go as much as 2.3dB over the sample-peak limit (as measured by the Dolby Media Meter's True Peak measurement function). But we don't want to just super-squash all our mixes! Which leads to issue 2...

2) we're getting push-back from our mixers for over-limiting their mixes in layback, and I can't blame them. I wouldn't want my mix to be squashed any more than necessary either.

I've also devised a rather ugly brute-force method for tackling this issue involving upsampling, audio-suite limiting, and downsampling. It works, but it's very slow and cumbersome. Maybe I'll mention that TC Electronic LM2 rack unit to a few folks around here and see what they think. So far it appears to be the only fool-proof option.

The Massey limited doesn't let you over shoot. 79 bucks.

I have not tried it yet, but Fabfilter Pro-L features 4x oversampling, so it should work for that purpose.
FabFilter Pro-L Manual
Cheers

A mistake is to set your peak limit threshold at 0 dBFs in the first place. Unless you're using a delay line limiter, you're going to get overshoots above zero. If you set a conventional (good) limiter a couple dB lower and use a moderately fast attack time (like 300 usec), then any resulting overshoot probably won't cause the intersample problem. If you're managing levels properly, you should barely be touching a limiter threshold at -2 dBFS.

If you need more aggressive peak control, you can always run two limiters in series. The first would have a threshold around -9 to -7 dBFS with a moderate attack time (like 3 msec), and the second faster one would be setup as described above. I don't deal with plugins, so I can't help you there.

I know the digital domain, and the analog domain , but what domain exactly is the dBTP domain?

True peak in digital is whatever the digital value is assuming it has not clipped.

True peak in analog depends on the D/A converter design.
Nyquist guarantees that a lot of the recovered Analog values will be higher than the digital samples taken from the original; as were the original samples because only sampling DC would result in them being equal.

Have they fooled themselves into believing teh allged interpeak sample problem floating around the net that confuses the analog and digital domains.

Or is this something else?

What is the dBTP domain ????

I doubt the AES and EBU have fooled themselves... Intersample (true) peaks are a reality that can be measured and proven to be problematic e.g. during lossy compression (tv broadcast). Therefore TP limiter is a necessity.

Alexey,

have a look at nugenaudio.com. Not only do they have a nice and simple loudness metering, they also introduced a "Loudness Management Batch processor" with a dedicated TruePeak Limiter recently.

cheers

Why you should go so hot that relying on limiter is the only way to go?
Why should you use a limiter at all in any drama or docs?
OK if you use it for safety, but then again any good limiter like Massey L2007 or ML4000 or Sonnox will do.
If you are in commercials and obliged to give a highest volume for money...well then it's a money and you can allow to buy a tc hardware...

On a similar note, the Voxengo Elephant (Audio mastering limiter (maximizer) plugin (AU, VST) - Elephant - Voxengo) has 8x oversampling. Has anyone tested True Peak with this one?

Alistair

He also has an average level of -24 LKFS, so he probably isn't hitting the limiter very hard.

to the original thread poster I 2nd the suggestion of trying out the Nugen Audio meter, we're demo-ing it at the moment and it is really nice. (From september there will also be loudness spec's in effect in the Netherlands, though the spec's differ slightly from the US ones)

FYI. The peak oveshoot phenomena was discovered decades ago. Tektronix recognized the problem back in the early 90's, and incorporated a filtering algorithm in their digital metering equipment to emulate the overshoot that was occurring in various equipments, such as D/A converters. Although they've been preaching the gospel of true peak measurement and control ever since, most practitioners have been just been ignoring the problem....until recently.

If everyone manged levels correctly, and peaked say at -3 to -2 dBFS, with maybe an OCCASIONAL (i.e. stray) peak reaching 0 dBFS, then you probably wouldn't be having the problem. Driving or normalizing program levels to hit 0 dBFS IS the problem. It's BAD operating practice.

An easy way of dealing with this is to mix at a higher LKFS level, and then reduce the overall level of the final product to meet the -24 LKFS delivery spec. For instance: if you know that your end product typically is peaking +1.9 dBFS "true peak" with whatever limiting you're using, then mix to -22 LKFS instead of -24, and when you finish, reduce the overall level by 2 dB. That will give you the desired -24 LKFS average level and the peaks should just be reaching 0 dBFS.

Do you have a sample where this problem occurs?
dBTP levels do depend on the method of their measurement. So, it may be that Ozone and you are using different measurement standards. How did you tell that dBTP level of your program is 0.9 dB over?

Thanks again everyone for the replies.

I've tested most of these options where possible; so far I haven't yet come across anything that has the simplicity and elegance of the Waves L1-L2-L3 series limiters that simply adds inter-sample peak limiting, and functions as a plugin in Pro Tools.

In response to the question about how I'm determining the exact extent of the true peak overs, we're using the Dolby Media Meter. It conforms to the BS.1770 specifications, and therefore provides the same readings as the Techtronix 7000-series meters that are used in broadcast QC.

The Nugen Audio stand-alone application looks like it would probably do the trick, but it's a stand-alone app and it's windows-only. We're a Mac-only and Pro Tools-only shop, and our workflows and time constraints don't really accommodate working with a separate system and a standalone application. When clients finish the mix, they want to roll straight into layback.

The Massey limiter doesn't actually limit intersample peaks. The Voxengo Elephant plugin doesn't come in TDM/RTAS format (I wish it did!). The FabFilter Pro doesn't seem to have a nice and simple threshold slider and output ceiling slider, so I haven't yet fully tested that - it's really geared towards music anyway. So far it seems my best hope is still that TC Electronic creates a plugin version of their LM2.

Also, just for clarification, this problem doesn't involve hitting 0dBFS and exceeding it. No broadcast network allows your peaks to get to 0dBFS; the most common spec is -6dBFS TruePeak (once again, as recommended in the BS.1770 guidelines). We even have some international deliverables requiring that we not exceed -10dBFS TruePeak. The problem is that when you set a sample-based limiter to -6dB, it's fairly easy for your inter-sample peaks to exceed that limit by a fair amount.

Anyway, thanks again for all the replies! Let's hope that an option comes along soon that's tailored to the post production sound community!

Have you tested this?

i read one early AES article abou this that was so full of errors that I am skeptical until someone shows facts not claims.

The TRUE peak for digital is the highest digital value <0
if you hit zero or clipped then all bets are off
SO DONT DO THAT !!
You do not need to have the digital values anywhere near zero to get loud sound out in the analog domain

the TRUE peak for analog depends on how far you twist the knob to the right. you can make 0dbfs have no sound at all or you can make -60 dB crank out a kilowatt of sound after the d/a.

what/where is this alleged true peak domain ???????????????
how does one convert from digital or analog domain to and from this so called true peak domain

the only problem that i can find is that people have confused digital with analog and dont understand nyquist or how d/a really works in real life not theory.

...No broadcast network allows your peaks to get to 0dBFS; the most common spec is -6dBFS TruePeak (once again, as recommended in the BS.1770 guidelines). We even have some international deliverables requiring that we not exceed -10dBFS TruePeak....

I wonder if someone screwed up with this. The EBU standard prior to the LKFS philosophy was 0 PPM = -9 dBFS and reference tone = -18 dBFS. I can see why they would want you to continue to keep the quasi-peak (DIN PPM) level to around -10 dBFS, but it makes no sense to limit true peak levels to -10 dBFS! The EBU is recommending that everyone forget the old PPM regime and to stick with the new -24 LKFS (or is it -23?) average level and to limit true peak levels to around -2 dBFS, to allow for a little headroom. The major problem with these specifications is that too many electro-political hacks get involved and just wind-up confusing the whole issue.

Hi all,

Unfortunately this problem is all too real. Please watch this YouTube video that does a better job explaining it than I could:

&#x202a;Final Cut Demo - true-peak vs peak sample metering for professional audio&#x202c;&rlm; - YouTube

In fact to illustrate the point I can create peaks that exceed their initially intended limit all in Pro Tools, in mere moments. Here's a screen grab of a 10-second clip of 10kHz tone at -20dBFS, with the gain plugin showing the analyzed level:



Now here's the exact same region after I've converted the session from 48kHz to 192kHz (a 4x up-sample).



The tone is now 0.4dB hotter! This happened simply as a result of upsampling. I didn't apply any gain at all. This is essentially exactly what's happening inside the loudness meters being used by quality control at broadcast.

Now this may or may not be particularly representative of exactly what's happening in broadcast during the D-to-A conversion on a consumer's television, but that point is unfortunately entirely moot to me; the networks to whom I must deliver content are measuring it in this way, and I need to comply with their specifications and requirements. So my hands are tied.

Anyway, I hope that further clarifies the nature of the issue I'm facing. Thanks again all for the comments and discussion!

This is not how digital audio works.

Here is a picture of a waveform I created to to illustrate the problem. (Created in Adobe Audition. I think it is the only wave editor that shows the reconstructed waveform). The sample values are shown by the square blocks. The waveform you see is what will come out of the converters after reconstruction. Note the peak at above +6 dB FS! (Even though the sample values are at 0 dB FS).



If you have your converters calibrated so that 0 dB FS is equivalent to say +4 dBu, this signal in this picture will cause a peak of over +10 dBu in the analogue domain!

True Peak is about the reconstructed waveform as shown in the picture above. It is a prediction of what will happen when the digital signal is reconstructed in the analogue domain. The sample values in the digital domain are just that, sample values. They are not the waveform as it will be after reconstruction.

I suggest you do a bit more research about how digital sampling really works. You might find that it isn't as intuitive as it first seems.

This white paper by Dan Lavry should get you started: http://www.lavryengineering.com/docu...ing_Theory.pdf

Alistair

Perhaps we should petition TC <g>! I'd love to have an LM2 plug-in also!

yes
conversions can cause problems
DONT DO THAT !!

Record at highest multiple you can of the final sample rate
2x conversions do not cause problems
fractional ones can cause problems

The conversion is not causing the problem. The problem is already present in the signal before any SRC. The SRC just makes it visible in the non-oversampled meter of the AudioSuite gain plugin.

In a modern SRC, there is no quality difference between integer ratio sample rate SRCs and non-integer ratios.

I suggest you do a bit more reading about how digital audio works before commenting. Here is a good book: Principles of Digital Audio, Sixth Edition Digital Video/Audio: Amazon.co.uk: Ken C. Pohlmann: Books

Alistair

FS is meaningless in the analog domain
you can turn the knob to the left to control how large the analog signal will be. you can turn a hot signal in the digital domain to -60db in the analog if you want to.

digital Samples do not equate to analog signals that way!!
thta is the way you drew them. you cannot mix domains!!!

nyquist guarantees that the analog will be higher than the digital samples in many places cause the digital was lower than the analog when it was sampled originally and you COULD get back the same signal.

but when you get it back you can adjust the absolute max out of the D/A with the gain knob. you can get back bigger or smaller signal with the same shape.

dB FS levels are not entirely meaningless if the levels have been calibrated, hence my mention of calibrating, but the whole point of this thread is that the relationship is not as direct as one might presume. At least with most basic meters included in most DAWs.

Anyway, imagine taking a sine wave signal in the digital domain that peaks at -6 dB FS. Calibrate your system so that you measure -12 dBu peak at the outputs in the analogue domain. Now feed the signal in my picture to the same DAC without touching the volume/gain control! This signal will peak at above 0 dBu and not at -6 dBu as one might suspect. Get it?

What are you basing this on? Did you read Dan Lavry's paper?

I didn't draw anything. The software shows a reconstructed waveform. That is what it would look like in the analogue domain if you attached an oscilloscope to the output of your DAC when fed with that signal.

This is a meaningless comment as it stands.

This is not what we are talking about.

On second thought, this book might be more accessible than Pohlmann's: Digital Audio Explained: For the Audio Engineer: Amazon.co.uk: Nika Aldrich: Books

Alistair

You are correct that waht you claim is not hwo it works
Thanks.
I did this in graduate school.
I know how sampling works.
Most fo the folks on the internet only think they know how it works. Most cant state nyquist correctly.
just read three papers in various forums about this.
THEY ALL MADE BASIC MISTAKES.
ERRONEOUS ASSUMPTIONS (Implicit ones THAT ARE WRONG)
Confusign the two domains.
Dont know how REAL LIFE D/A works.
Dont know the limitations on the theory due to approximations that are made in real life.
The list goes on.

I've only seen this feature in Waves L1+. Is this option really present in L2 and L3? I thought that they had pulled it out.

Anyway, if you use Ozone's Inter-sample option with Character set to higher than 4, it should safe-guard your dBTP levels just as L1+ did.

BTW mind you that no plugin can guarantee absolute and strict safety of dBTP levels! This is because the ITU standard does not specify the exact way of their calculation. It only mentions that a 4x oversampling has to be applied, leaving it up to the manufacturer to implement this oversampling in a particular way. So, if you want to be safe with your production levels, use both Inter-sample feature and -0.1 dB ceiling for some extra headroom.