You are here

LUFS and True Peak Max

For everything after the recording stage: hardware/software and how you use it.

LUFS and True Peak Max

Postby stefan.pynappels » Tue Jul 30, 2019 8:30 pm

Hi All,

First post here, hoping the learned collective can help me.

I have recorded the voice track for a tutorial video at work using a lav mic into a Zoom H4n Pro, and did the post processing in Audacity. I am by no means experienced in audio processing, but I was able to get the audio sounding good, and loudness normalised to -14 LUFS using the Youlean Loudness Meter plugin.
There was not a great loudness range, as expected for a voiceover track, but there were a number of points where a hard "ch" or "t" sound caused a peak. Even with a hard limit applied using the default Audacity limiter, set at -1dB, and the LUFS at -14, there were still a small number of occurrences where the True Peak Max went to around +1.0dB.

Now, as I understood it, there should never be any occurrence of the peak going above 0dB as this causes clipping (even if it is inaudible). So how do you go about avoiding this happening? Take the peaks out before normalisation using targeted attenuation? Some other way? Avoid the problematic sounds being picked up at the recording stage somehow?

Don't get me wrong, the audio in the video was fine, and there were only 4 momentary peaks over the course of a 6 minute video, but I was just wondering if there was some better technique I could apply to avoid any positive True Peak Max instances at all.
stefan.pynappels
New here
Posts: 6
Joined: Tue Jul 30, 2019 5:18 pm

Re: LUFS and True Peak Max

Postby blinddrew » Tue Jul 30, 2019 9:32 pm

If you're only looking at a couple of instances I'd just automate them out - drop the volume for those syllables only. It's probably quicker than trying to apply any global processing.
We've got a couple of VO specialists on the forum here though, so if they contradict me, listen to them not me! :)
User avatar
blinddrew
Jedi Poster
Posts: 7659
Joined: Sun Jul 05, 2015 12:00 am
Location: York
Ignore the post count, I have no idea what I'm doing...

Re: LUFS and True Peak Max

Postby Hugh Robjohns » Tue Jul 30, 2019 9:34 pm

stefan.pynappels wrote:First post here...

Welcome Stefan!

Even with a hard limit applied using the default Audacity limiter, set at -1dB, and the LUFS at -14, there were still a small number of occurrences where the True Peak Max went to around +1.0dB.

Most limiters can only react shortly after a signal transient occurs, so it's common for there to be a short 'overshoot'. In the analogue world, even if these overshoots cause brief (less than 1ms) clipping (and the associated harmonic distortion) the ear/brain tends not to notice. Unfortunately, if they cause a digital overload the resulting distortion is anharmonic (the distortion components are not musically related to the source signal), and that sounds very unnatural, so even the most brief of overloads tends to get noticed.

So how do you go about avoiding this happening?

Setting the peak limiter's threshold lower would be the easiest and most obvious way. Try -4 or -5dBFS instead of -1dBFS, for example.

Or you could try a faster attack time if that option is available -- although doing that could actually result in a different kind of audible (transient) distortion, even if the signal peak remains well below clipping, so it may not be a practical option.

An alternative approach -- and this is what I tend to do -- would be a quick bit of manual editing of the track to attenuate the offending short sections by a few decibels. (Or use fader automation to do the same thing as Drew suggests... But I find editing quicker, easier and more precise).

Or you could find a 'look-ahead limiter' -- a plugin that effectively feeds its side-chain from slightly in advance of the signal itself, so that it can attenuate the level just in advance of the peak arriving, and thus avoid overshoots completely.

H
User avatar
Hugh Robjohns
Moderator
Posts: 24088
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: LUFS and True Peak Max

Postby stefan.pynappels » Tue Jul 30, 2019 10:27 pm

Thanks for the responses, that makes a whole lot of sense.

It looks like I have a few avenues to pursue, with a very small number of peak, attenuating then manually may be the best option, although I also found a different limiter plugin that might be able to do the look-ahead limiting thing, it's called Limiter No.6 (https://vladgsound.wordpress.com/plugins/limiter6/).

Has anyone had any experience using that?

Thanks for the help so far, at least it's given me some further things to read up on.
stefan.pynappels
New here
Posts: 6
Joined: Tue Jul 30, 2019 5:18 pm

Re: LUFS and True Peak Max

Postby cyrano.mac » Tue Jul 30, 2019 10:48 pm

With only 4 peaks, I'd try manual volume adjustment...

I mean, why does everything need to be (possibly) mangled by a plugin?
cyrano.mac
Poster
Posts: 32
Joined: Sun Jun 02, 2019 5:00 pm

Re: LUFS and True Peak Max

Postby hobbyist » Tue Jul 30, 2019 11:27 pm

stefan.pynappels wrote:Hi All,

First post here, hoping the learned collective can help me.

I have recorded the voice track for a tutorial video at work using a lav mic into a Zoom H4n Pro, and did the post processing in Audacity. I am by no means experienced in audio processing, but I was able to get the audio sounding good, and loudness normalised to -14 LUFS using the Youlean Loudness Meter plugin.
There was not a great loudness range, as expected for a voiceover track, but there were a number of points where a hard "ch" or "t" sound caused a peak. Even with a hard limit applied using the default Audacity limiter, set at -1dB, and the LUFS at -14, there were still a small number of occurrences where the True Peak Max went to around +1.0dB.

Now, as I understood it, there should never be any occurrence of the peak going above 0dB as this causes clipping (even if it is inaudible). So how do you go about avoiding this happening? Take the peaks out before normalisation using targeted attenuation? Some other way? Avoid the problematic sounds being picked up at the recording stage somehow?

Don't get me wrong, the audio in the video was fine, and there were only 4 momentary peaks over the course of a 6 minute video, but I was just wondering if there was some better technique I could apply to avoid any positive True Peak Max instances at all.


True peak is meaningless until you convert D/A.

If no actual peak is at 0dBFS or above there is no problem yet.

Just reduce the gain before converting. But you need to be aware of how your D/A is designed. A bad D/A design can still cause problems if you are close to 0dBFS. A good D/A design will have lowered your signal enough before doing the D/A so you would not have a problem.

And some D/A have problems near 0dBFS anyway. So pushing as close as possible can still lead to a lessened quality analog signal.

Best to keep at least below -6dBFS before sending to the D/A.

Personally I prefer to be at -18dBFS and just turn the knob on the amplifier to the right if I want to hear it louder on my stereo.

If you are in the louder is better camp then you have to decide how risky you want to be.

Also if you are in the wide DR camp you may also have a problem meeting the new standards for loudness.
hobbyist
Regular
Posts: 280
Joined: Mon Jul 01, 2019 12:52 am

Re: LUFS and True Peak Max

Postby Hugh Robjohns » Tue Jul 30, 2019 11:57 pm

hobbyist wrote:True peak is meaningless until you convert D/A.

I don't agree -- it's not meaningless... But I can see where you're coming from!

It's true that the digitised signal is perfectly intact and undamaged if its samples always remain at or below 0dBFS, and that the True Peak value is a close approximation of the reconstructed analogue signal which effectively only exists at the output of the D-A (or within some internal digital signal processing etc).

However, I don't think it's right to say the True Peak value is meaningless until conversion because (a) in the context of producing material for a loudness-normalised medium the delivery spec usually states the True Peak value must be under -1dBTP, or -3dBTP, or whatever. And (b) any signal processing involving over-sampling or complex filtering will encounter those inter-sample peaks and overload as a result -- so their presence has meaning and it's therefore very useful to know about them so they can be corrected!

So, in those increasingly common circumstances, the True Peak value most definitely does have a critically important meaning -- at least if you want your material to be accepted rather than rejected, and clean rather than distorted when processing -- regardless of whether it's being sent to a D-A converter anywhere nearby!

Best to keep at least below -6dBFS before sending to the D/A.

I certainly agree with you on that point! All of my digital replay systems handling commercial CDs and downloads are actually configured with 6dB of attenuation to provide some much-needed headroom in the digital domain, specifically to cope with unintended inter-sample peaks present in most commercial (peak normalised) music recordings.

Personally I prefer to be at -18dBFS ...

That seems rather excessive headroom for mastered commercial mixes... Or are you talking about tracking/mixing headroom?
User avatar
Hugh Robjohns
Moderator
Posts: 24088
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: LUFS and True Peak Max

Postby hobbyist » Wed Jul 31, 2019 12:57 am

Hugh Robjohns wrote:
hobbyist wrote:True peak is meaningless until you convert D/A.

I don't agree -- it's not meaningless... But I can see where you're coming from!

It's true that the digitised signal is perfectly intact and undamaged if its samples always remain at or below 0dBFS, and that the True Peak value is a close approximation of the reconstructed analogue signal which effectively only exists at the output of the D-A (or within some internal digital signal processing etc).

However, I don't think it's right to say the True Peak value is meaningless until conversion because (a) in the context of producing material for a loudness-normalised medium the delivery spec usually states the True Peak value must be under -1dBTP, or -3dBTP, or whatever. And (b) any signal processing involving over-sampling or complex filtering will encounter those inter-sample peaks and overload as a result -- so their presence has meaning and it's therefore very useful to know about them so they can be corrected!

So, in those increasingly common circumstances, the True Peak value most definitely does have a critically important meaning -- at least if you want your material to be accepted rather than rejected, and clean rather than distorted when processing -- regardless of whether it's being sent to a D-A converter anywhere nearby!

Best to keep at least below -6dBFS before sending to the D/A.

I certainly agree with you on that point! All of my digital replay systems handling commercial CDs and downloads are actually configured with 6dB of attenuation to provide some much-needed headroom in the digital domain, specifically to cope with unintended inter-sample peaks present in most commercial (peak normalised) music recordings.

Personally I prefer to be at -18dBFS ...

That seems rather excessive headroom for mastered commercial mixes... Or are you talking about tracking/mixing headroom?


TP is just their guess for the converted peak at the sample rate they assume.

Real D/A converters are still iffy near the max.

-6dBFS should be sufficient to avoid true clipping as well as the problems real life D/A converters may have.

-18dBFS is just my being cautious to avoid problems. There is still way more S/N and DR left than anybody would ever need. So why push it to the max unless you are trying to win the loudness war.

Or as you note, if you are submitting to some media or streaming service that has requirements to do it higher.

I don't use them but I think I read, maybe I misread, that if you submit lower they will bring it up to the level they want it at. No problems with that that I can see. Unclear to me but maybe you do need to limit the DR to whatever range they say first.

But give them something too hot and they will fix it in a way that may make your file worse than it could have been if you had not exceeded their specs.
hobbyist
Regular
Posts: 280
Joined: Mon Jul 01, 2019 12:52 am

Re: LUFS and True Peak Max

Postby stefan.pynappels » Wed Jul 31, 2019 6:38 am

Thanks hobbyist and Hugh.

I get that keeping the peaks lower will give more headroom and that the potential issues depend on the A-D too.

As I'm dealing pretty much only with video voiceovers I'm not looking for extreme DR or even extreme loudness, this was more a question for me to learn from. Thank you for your clear explanations, I now have a lot more reading to do.
stefan.pynappels
New here
Posts: 6
Joined: Tue Jul 30, 2019 5:18 pm

Re: LUFS and True Peak Max

Postby blinddrew » Wed Jul 31, 2019 8:09 am

Some streaming services turn up but not all. Hence sites like Loudness Penalty can be useful when you're reaching the mixdown stages.
User avatar
blinddrew
Jedi Poster
Posts: 7659
Joined: Sun Jul 05, 2015 12:00 am
Location: York
Ignore the post count, I have no idea what I'm doing...

Re: LUFS and True Peak Max

Postby Zukan » Wed Jul 31, 2019 9:00 am

I'd echo Hugh's advice here but just as an aside and maybe it might help you, I use a technique I created a video about. It involves looking for errant spikes/transients and removing them using a clipper plugin. I use GClip and it finds errant spikes pretty well. Once found you can dial them down to the desired level. Works better than a limiter in cases where there are the odd dodgy transients riding high.

In fact, and I hate to reveal this here, it is the same technique certain mastering engineers use to maximise loudness.
User avatar
Zukan
Moderator
Posts: 7870
Joined: Fri Sep 12, 2003 12:00 am

Re: LUFS and True Peak Max

Postby Hugh Robjohns » Wed Jul 31, 2019 9:36 am

hobbyist wrote:TP is just their guess for the converted peak at the sample rate they assume.

No, the True Peak value is not 'a guess'. It's a well-defined calculation of the reconstructed waveform's true amplitude, with an equally well-defined error-margin. The higher the oversampling rate used in the calculation, the smaller the error margin.

There is absolutely no guesswork involved!

That might just have been an unfortunate turn of phrase... but this is the way misleading and damaging myths build! ;)

Most True Peak metering algorithms employ a 4x oversampling rate which results in a maximum possible error margin of 0.688dB. That's why the BS.1770 specs set the maximum acceptable True Peak value at -1dBFS -- to ensure there is absolutely no possibility of digital clipping.

A TP metering system using 8x oversampling would have a maximum error of 0.169dB, but the additional processing overhead to achieve 8x oversampling is often considered excessive for the relatively small benefit in precision that it provides. Higher oversample rates obviously result in even smaller error margins.

The whole story (and the nature of the TP algorithm calculations) are described in Annex 2 of ITU-R BS.1770-4: https://www.itu.int/rec/R-REC-BS.1770-4-201510-I/en

H
User avatar
Hugh Robjohns
Moderator
Posts: 24088
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: LUFS and True Peak Max

Postby stefan.pynappels » Wed Jul 31, 2019 9:39 am

Zukan wrote: I use GClip and it finds errant spikes pretty well.

Interesting, it looks like it is a Windows only plugin, so should be worth a look on my work computer. Thanks for the tip.
stefan.pynappels
New here
Posts: 6
Joined: Tue Jul 30, 2019 5:18 pm

Re: LUFS and True Peak Max

Postby Mike Stranks » Wed Jul 31, 2019 9:48 am

Welcome!

I'm with Hugh in his advice to set the threshold of the peak limiter slightly lower. I tend to work at about -4 to -6 for voice. (Of course, it depends on the voice.... I'm fortunate that most of the people I work with know how to 'do' V/Os and can add emphasis without increasing the volume dramatically.) But also, as advised, be careful you're not inadvertently doing damage to some transients. Our ears/brains are acutely aware of distortions/anomalies with voice sounds - especially if the voice is exposed and not blended with other sounds.

... and coming in on another tack, I wouldn't use a lav mic for this type of work. Maybe it was all you had or circumstances dictated that it had to be a lav. As always, life is easier if you can get the source sound as good as possible. I've found that lavs - however good - can often introduce their own 'interesting features'! :)
Mike Stranks
Jedi Poster
Posts: 6729
Joined: Fri Jan 03, 2003 1:00 am

Re: LUFS and True Peak Max

Postby stefan.pynappels » Wed Jul 31, 2019 9:53 am

Mike Stranks wrote:... and coming in on another tack, I wouldn't use a lav mic for this type of work. Maybe it was all you had or circumstances dictated that it had to be a lav. As always, life is easier if you can get the source sound as good as possible. I've found that lavs - however good - can often introduce their own 'interesting features'! :)

Unfortunately it was all I had, although I have dug out an old Behringer C-1 and will do some testing on that. I know Behringer have a variable (!) reputation, but it'll be good to see whether it's better than the lav.

Thanks for the response :-)
stefan.pynappels
New here
Posts: 6
Joined: Tue Jul 30, 2019 5:18 pm

Re: LUFS and True Peak Max

Postby The Elf » Wed Jul 31, 2019 10:07 am

+1 for GClip, but use it as a last resort, not a first.
User avatar
The Elf
Jedi Poster
Posts: 12460
Joined: Tue Aug 14, 2001 12:00 am
Location: Sheffield, UK
An Eagle for an Emperor, A Kestrel for a Knave.

Re: LUFS and True Peak Max

Postby hobbyist » Wed Jul 31, 2019 4:58 pm

Hugh Robjohns wrote:
hobbyist wrote:TP is just their guess for the converted peak at the sample rate they assume.

No, the True Peak value is not 'a guess'. It's a well-defined calculation of the reconstructed waveform's true amplitude, with an equally well-defined error-margin. The higher the oversampling rate used in the calculation, the smaller the error margin.

There is absolutely no guesswork involved!

That might just have been an unfortunate turn of phrase... but this is the way misleading and damaging myths build! ;)

Most True Peak metering algorithms employ a 4x oversampling rate which results in a maximum possible error margin of 0.688dB. That's why the BS.1770 specs set the maximum acceptable True Peak value at -1dBFS -- to ensure there is absolutely no possibility of digital clipping.

A TP metering system using 8x oversampling would have a maximum error of 0.169dB, but the additional processing overhead to achieve 8x oversampling is often considered excessive for the relatively small benefit in precision that it provides. Higher oversample rates obviously result in even smaller error margins.

The whole story (and the nature of the TP algorithm calculations) are described in Annex 2 of ITU-R BS.1770-4: https://www.itu.int/rec/R-REC-BS.1770-4-201510-I/en

H


If there is any uncertainty or error margin then that is effectively the same as a guess.

The 0.688 may be close enough for govt work though:)

And using the 'TP' at -1 would allow for it completely. What you can not do is compute exactly the actual TP after conversion.

The key point that most people do not realise is that if you do not go over 0dBFS in the digital domain then you do not have a problem in the analog domain *IF* you lower the signal so your A/D can handle it without conversion problems. And most A/D need a little more as they are less than perfect when right up near their max digital value that they can handle well.
hobbyist
Regular
Posts: 280
Joined: Mon Jul 01, 2019 12:52 am

Re: LUFS and True Peak Max

Postby Hugh Robjohns » Wed Jul 31, 2019 5:29 pm

hobbyist wrote:If there is any uncertainty or error margin then that is effectively the same as a guess.

Again, we'll need to agree to differ on that one. To me a guess is a conclusion arrived at with insufficient information to be sure of being correct. A guess can be completely wrong.

The TP value is a calculation, not a guess, and it is guaranteed to be correct within a known error margin -- which is a perfectly legitimate way of expressing a scientific result. Technically, the TP value is an approximation to with 0.688dB (for 4x over-sampling)... Which isn't a guess! ;-)

If it was desirable or necessary, the true peak value -- ie the absolute waveform amplitude -- could be calculated with total precision since the amplitude of the waveform at every point in time is known absolutely from the available samples -- as defined by Nyquist's theorem. If this were not the case sample rate converters wouldn't work!

However, calculating the amplitude at every point in time is clearly very wasteful in terms of processor cycles and, more to the point, not actually necessary in the metering application defined in BS.1770. Instead, a simpler calculation to provide a very close approximation with a known error margin has been deemed acceptable, and an allowance is built in to the specification to accommodate it.

The key point that most people do not realise is that if you do not go over 0dBFS in the digital domain then you do not have a problem in the analog domain *IF* you lower the signal so your A/D can handle it without conversion problems.

You mean D-A, but yes, I agree -- as discussed earlier.

H
User avatar
Hugh Robjohns
Moderator
Posts: 24088
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound


Who is online

Users browsing this forum: Aural Reject