PEACE clipping meter question!

[Robbo99999]

Maybe it's calculating reconstructed peak or something like that. basically "rounding" clipped waveforms to show were the wave would actually peak if it weren't for limiters. Though it's pretty useless, so I'm not sure why would it do that.

Are you referring to intersample overs? (what limiters are you talking about, the limiters applied during mixing & music creation in the studios?)

 

[Fluffy]

I have a meter program called Orban loudness meter. It has an option to show reconstructed peak, and in compressed music it shows peaks above 0dbfs, though the file itself never goes beyond 0dbfs according to audacity:

 

[Robbo99999]

I have a meter program called Orban loudness meter. It has an option to show reconstructed peak, and in compressed music it shows peaks above 0dbfs, though the file itself never goes beyond 0dbfs according to audacity:
View attachment 76353

Yep, read the documentation just now (ftp://ftp.orban.com/Orban_Loudness_Meter/Documentation/FreeOrbanLoudnessMeter_2.9.6_Readme.pdf), that actually measures the intersample overs (in addition to the total EQ curve), which is basically just showing when you convert from digital to analog how much over 0dB there can be....due to the curiosities of how digital samples can be converted to analog it's possible that "portion between two digital points" can end up being higher than 0dB when converted which is more likely to occur when the music is recorded close to the 0dB limit.

Maybe it's possible that PEACE is doing the same thing with it's clipping meter, although given I'm already running -4.3dB in Equaliser APO combined with -2.6dB on the Windows Volume then I'm already -6.9dB attenuated over most of the frequency range so that should account for intersample overs. It could be that the clipping meter in PEACE assumes that you're running at 100% windows volume and provides it's recommendations accordingly...assuming that PEACE's clipping meter is actually accounting for intersample overs. You can combat intersample overs by just running Windows Volume below 100% - some people say just by a couple of percent, and others say anything from -6dB to -12dB, I think -1dB is fine for accounting for most intersample overs in my general intuition after reading around.

 

[Jimbob54]

Has clipping been heard along with the clipping indicator in Peace pinging?

 

[Robbo99999]

Has clipping been heard along with the clipping indicator in Peace pinging?

No sirree, this is purely a theoretical exercise to get everything optimal......like most things! Don't you bloody ask me to blind test!

 

[Jimbob54]

No sirree, this is purely a theoretical exercise to get everything optimal......like most things! Don't you bloody ask me to blind test!

Wouldnt dream of it. But how much more pre amp reduction do you have to apply to get it to stop pinging in all circumstances? 0.5dB or 5 dB? If more the former, just do it.

I had a similar issue in Roon- clipping indicator flashing when it shouldnt based on EQ and pre amp set up. Reason: I'd selected one sample rate to upsample. No upsampling in your chain?

 

[RayDunzl]

@RayDunzl your examples are mistaken. I don't have Peace but I use EqAPO in its basic form. It doesn't matter what combined filters you have, it just matters which one is the largest boost, and you should set you pre-amp according to it. Here is a signal of 105hz+6khz normalized to -0.1 dbfs:
View attachment 76345


Now I'll add +5 db at 105hz and +3 db at 6khz:
View attachment 76344


It gives me exactly 5 db of gain above odbfs:
View attachment 76343


A loopback recording in audacity will give me a clipped signal:
View attachment 76342


Setting a -5 db pre-amp will eliminate the clipping:
View attachment 76341
View attachment 76340
View attachment 76339

I see your point. Not sure about "it just matters which one is the largest boost, and you should set you pre-amp according to it" without further investigation, for additional frequencies. Might still be true. Maybe a good "rule of thumb".

The math is tricky.

Decibels are not the same linear size at different points along a linear scale (voltage).

The individual waves in my example are at an amplitude of 0.5, or -6dBfs.

Increasing the 105Hz by 5dB adds gives a ratio of 1.778279 x 0.5 = 0.8891395 linear value - Adds 0.3891395 linearly

Increasing the 6000Hz by 3dB gives a ratio of 1.412538 x 0.5 = 0.706269 linear value - Adds 0.206269 linearly

Summing those gives 1.5954085 linear value

Reducing that summed voltage by 5dB (the largest boost): Ratio of 0.562341 x 1.5954085 = 0.897163 (not clipping), a linear drop of 0.6982455 which is more than the combined increases of 0.5954085

 

[Robbo99999]

Wouldnt dream of it. But how much more pre amp reduction do you have to apply to get it to stop pinging in all circumstances? 0.5dB or 5 dB? If more the former, just do it.

I had a similar issue in Roon- clipping indicator flashing when it shouldnt based on EQ and pre amp set up. Reason: I'd selected one sample rate to upsample. No upsampling in your chain?

-2.2dB extra preamp above & beyond, only on one track that to me is bass heavy around my highest point in EQ curve whilst also being a track recorded at high loudness, so worst case of worst case if you will. I'm running 44.1kHz on sound card and also on players & content. Windows set at 24bit, ITunes set at 24bit, and (of course) 16bit content being played back.

 

[Robbo99999]

I see your point. Not sure about "it just matters which one is the largest boost, and you should set you pre-amp according to it" without further investigation, for additional frequencies. Might still be true. Maybe a good "rule of thumb".

The math is tricky.

Decibels are not the same linear size at different points along a linear scale (voltage).

The individual waves in my example are at an amplitude of 0.5, or -6dBfs.

Increasing the 105Hz by 5dB adds gives a ratio of 1.778279 x 0.5 = 0.8891395 linear value - Adds 0.3891395 linearly

Increasing the 6000Hz by 3dB gives a ratio of 1.412538 x 0.5 = 0.706269 linear value - Adds 0.206269 linearly

Summing those gives 1.5954085 linear value

Reducing that summed voltage by 5dB (the largest boost): Ratio of 0.562341 x 1.5954085 = 0.897163 (not clipping), a linear drop of 0.6982455 which is more than the combined increases of 0.5954085

He did the same experiment as you though and it showed no clipping, so we don't need to be blinded by the maths right? All he did was set the negative preamp to what is generally expected based on his total EQ curve. (So what's the distinction between your two conflicting results?)

 

[RayDunzl]

we don't need to be blinded by the maths right?

I wanted to understand my error in the given example.

Now I do.

I didn't expect "set the attenuation to the largest boost" to be sufficient, but it may be so.

 

[Robbo99999]

I wanted to understand my error in the given example.

Now I do.

I didn't expect "set the attenuation to the largest boost" to be sufficient, but it may be so.

Ah, ok, you both did two different experiments then. OK......so your initial replies to me were based on a false understanding of what "we" were doing......it's all there in the first post of this thread though in the screenshots as well as my words. Just applying a negative preamp equal to the sum of all the EQ filters (ie Total EQ curve as seen in graphs in PEACE and also in Equaliser APO). So, that means that you're not sure why PEACE is showing clipping in my (our) situations when we shouldn't expect it based on the total EQ curve? (You just had the wrong end of the stick before, is that correct? Nothing wrong with that, just I had an inkling from the get go (and in subsequent posts) you might have missed some important things I was trying to show in my first post that sent you down the wrong track.)

 

[Sgt. Ear Ache]

I've recently started playing around with Mathaudio Headphone EQ in foobar and noticed the same thing. Seems to need an extra db of attenuation to keep the meter from triggering...

 

[Robbo99999]

I have a meter program called Orban loudness meter. It has an option to show reconstructed peak, and in compressed music it shows peaks above 0dbfs, though the file itself never goes beyond 0dbfs according to audacity:
View attachment 76353

Wow, that's a pretty cool program, I downloaded and ran an analysis of some of my loudest tracks, I analysed around 40 tracks in total and the Highest Reconstructed Peak Level was 2.6dBFS on a Greenday track, whereas all others were under 2.0dBFS and most under 1.5dBFS......therefore I think we should run at -2dBFS on the main Windows Volume Control to counter these effects, which equates to 87% Windows Volume.

Interestingly in relation to the Clipping Meter showing clipping in PEACE, I'm pretty sure that it is also taking into account Reconstructed Peaks above 0dBFS as the worst overall offender in terms of number of peaks above 0dBFS was indeed my Supermassive Black Hole track that was triggering clipping in PEACE, here you can see the screenshot in Orban showing that there is a massive 125417 peaks above 0dBFS which is around say a 100 times more peaks than a lot of other loud tracks, with the highest peak being 1.8dBFS - so that correlates with the fact I was seeing the most clipping in PEACE with this track:

Interestingly, you can see patterns in some albumns in terms of how they have been recorded that even some loud albumns don't have reconstructed peaks above 0dBFS, it's almost like the engineer did it this way on purpose to prevent them, for example this is a very loud track from Florence & The Machine and no peaks above, which is the pattern on pretty much all the tracks on that albumn:

Yes, so in conclusion I think PEACE clipping meter is indeed showing valid clipping....it's taking into account the usual Total EQ Curve as well as Reconstructed Peaks above 0dBFS (aka intersample overs), so those 2 factors are being considered. To combat the Reconstructed Peaks I think the best strategy is to run -2dBFS in the main Windows Volume Control (which is 87% Windows Volume), and then you run an additional negative preamp on top of this to take into account your usual Total EQ Curve (the latter being the normal understood practice). So in my case I need -4.3dB negative preamp to account for my Total EQ Curve, and then I'll be running -2dBFS (87% Windows Volume) to account for Reconstructed Peaks, so that will be -6.3dB altogether.

 

[restorer-john]

I'll have to ask you tomorrow, because it's too late for me think clearly on it

Absolutely appreciate your honesty. Nothing worse than trying to get your head around esoteric numbers late at night, with or without a few drinks and a hard day under your belt.

Sometimes, I've typed out a reply, looked at it and decided to delete the draft and go to bed.

 

[jae]

I've recently started playing around with Mathaudio Headphone EQ in foobar and noticed the same thing. Seems to need an extra db of attenuation to keep the meter from triggering...

I've been using this recently too. The -6 dB limit on pre-amp and the lack of ability to copy and paste (or directly load) a preset in commonly found formats (like whats found on autoeq/oratory github) from are my two biggest gripes with it.

 

[Robbo99999]

Absolutely appreciate your honesty. Nothing worse than trying to get your head around esoteric numbers late at night, with or without a few drinks and a hard day under your belt.

Sometimes, I've typed out a reply, looked at it and decided to delete the draft and go to bed.

True, although it turned out Ray was the "confused one" this time, I think we got to the bottom of it if you see my previous post.

 

[Fluffy]

@Robbo99999 I don't think lowering your windows volume will help you here. Once a sample has reached beyond 0dbfs, any information above is lost – in other words, it clipped. the result is that the clipped part is sent to conversion as 0dbfs, or as the highest maximum digital value. I believe the information lost has already happened by the time it's sent to the windows mixer, so lowering the volume won't help to regain that lost information.

I did the following experiment: I took a song which had many clipped samples (reconstructed peak at about +1.5 dbfs), and made a loopback recording into audacity twice. First time at 100% volume, second time at 80%. I then time aligned the recordings and normalized both to 0 db. Then I inverted one recording, and mixed them together to show the difference. the result is a perfect null – which means, the samples clipped before reaching the mixer, and lowering the volume didn't "reveal" any information that exist above 0dbfs.

 

[dasdoing]

Wow, that's a pretty cool program, I downloaded and ran an analysis of some of my loudest tracks, I analysed around 40 tracks in total and the Highest Reconstructed Peak Level was 2.6dBFS on a Greenday track, whereas all others were under 2.0dBFS and most under 1.5dBFS......therefore I think we should run at -2dBFS on the main Windows Volume Control to counter these effects, which equates to 87% Windows Volume.

Interestingly in relation to the Clipping Meter showing clipping in PEACE, I'm pretty sure that it is also taking into account Reconstructed Peaks above 0dBFS as the worst overall offender in terms of number of peaks above 0dBFS was indeed my Supermassive Black Hole track that was triggering clipping in PEACE, here you can see the screenshot in Orban showing that there is a massive 125417 peaks above 0dBFS which is around say a 100 times more peaks than a lot of other loud tracks, with the highest peak being 1.8dBFS - so that correlates with the fact I was seeing the most clipping in PEACE with this track:
View attachment 76400

Interestingly, you can see patterns in some albumns in terms of how they have been recorded that even some loud albumns don't have reconstructed peaks above 0dBFS, it's almost like the engineer did it this way on purpose to prevent them, for example this is a very loud track from Florence & The Machine and no peaks above, which is the pattern on pretty much all the tracks on that albumn:
View attachment 76401

Yes, so in conclusion I think PEACE clipping meter is indeed showing valid clipping....it's taking into account the usual Total EQ Curve as well as Reconstructed Peaks above 0dBFS (aka intersample overs), so those 2 factors are being considered. To combat the Reconstructed Peaks I think the best strategy is to run -2dBFS in the main Windows Volume Control (which is 87% Windows Volume), and then you run an additional negative preamp on top of this to take into account your usual Total EQ Curve (the latter being the normal understood practice). So in my case I need -4.3dB negative preamp to account for my Total EQ Curve, and then I'll be running -2dBFS (87% Windows Volume) to account for Reconstructed Peaks, so that will be -6.3dB altogether.

believe it or not, but recently many engeniers use clipping as a "feature, not a bug"....in other words, it's used as an effect

this may sound stupid, but don't forget we have a whole genre based on distorted guitars

 

[Fluffy]

believe it or not, but recently many engeniers use clipping as a "feature, not a bug"....in other words, it's used as an effect

this may sound stupid, but don't forget we have a whole genre based on distorted guitars

If your goal as a producer is not to recreate a real world sound but to produce a sonic experience, clipping is as valid a method as any other.

 

[Robbo99999]

@Robbo99999 I don't think lowering your windows volume will help you here. Once a sample has reached beyond 0dbfs, any information above is lost – in other words, it clipped. the result is that the clipped part is sent to conversion as 0dbfs, or as the highest maximum digital value. I believe the information lost has already happened by the time it's sent to the windows mixer, so lowering the volume won't help to regain that lost information.

I did the following experiment: I took a song which had many clipped samples (reconstructed peak at about +1.5 dbfs), and made a loopback recording into audacity twice. First time at 100% volume, second time at 80%. I then time aligned the recordings and normalized both to 0 db. Then I inverted one recording, and mixed them together to show the difference. the result is a perfect null – which means, the samples clipped before reaching the mixer, and lowering the volume didn't "reveal" any information that exist above 0dbfs.

Hmm, interesting, the clipping of intersample overs I think turns out to be more analog clipping if running at 100% windows volume if your DAC doesn't have any built in overhead, but also I think I heard that running below 100% windows volume allows for Windows to still reconstruct those peaks that go over 0dBFS, but I'm a bit hazy on the specifics here......however I do know for a fact that a JDS Labs engineer mentioned in a forum that all folks should be running below 100% Windows Volume to counter Intersample Overs (which is the same thing as Reconstructed Peaks over 0dBFS), and that was based on loads of DACS on the market that the engineer tested to see how they behaved, so there is some credibility behind the theory that lowering Windows Volume can help counter Intersample Overs (it's also an understanding & adopted method used by experienced members on this forum, incl @JohnYang1997 who I believe is a Topping Manufacturer). The fact that your experiment showed no positive effect of lowering windows volume is strange, I'm not sure how to explain that.

believe it or not, but recently many engeniers use clipping as a "feature, not a bug"....in other words, it's used as an effect

this may sound stupid, but don't forget we have a whole genre based on distorted guitars

Ha, well that's a good point, maybe I'm not being "true to source" by keeping the intersample overs at bay! However, some DACS are immune to intersample overs, it's a specific type of DAC, I can't remember the class of name...ah, I think they're called multibit or R2R DACS - therefore there are some DACS that would go against you theory of intersample over distortion being a planned part of the music. I think on balance I don't really see intersample overs as being a proper planned part of the sound.