0dBFS

[Gary Gray]

Do you folks cut masters that go all the way up to 0dBFS or do you leave a little margin and if so, why?

Thanks!

gary

[cichy]

please find the answer and nice free VST (PC, Mac) here
Solid State Logic X-ISM Peak Meter at Kaos Audio

I found some different "standards" some ME master to -0.1dbfs , -0,3dbfd or -0.5dbfs

[starfighter]

Originally Posted by Gary Gray

Do you folks cut masters that go all the way up to 0dBFS or do you leave a little margin and if so, why?

Thanks!

gary

The simple answer is; FOR GOD'S SAKE NO!!!!!!!!!!!!!!

With -0.1db you make sure there's no clipping.. And no one on this forum's gonna bash you...

I always do an offline analysis on the final master before I ship just to make sure.

morten

[Torg]

please find the answer and nice free VST (PC, Mac) here
Solid State Logic X-ISM Peak Meter at Kaos Audio

I found some different "standards" some ME master to -0.1dbfs , -0,3dbfd or -0.5dbfs

+1 !

i was just looking for a free oversampled meter plugin.

Is there any other you know that displays the level, instead of just the presence of inter samp peaks?

ok, i believe i can just lower the input to the point where the indicator stops flashing and the amount i just lowered will be the level of ISP, but if the meter shows me on the fly would be much more convenient and accurate....

With -0.1db you make sure there's no clipping.. And no one on this forum's gonna bash you...

Could you elaborate more on that??

Cause i just ran a few tracks mastered to 0dBFs(not made by me) through the SSL meter and in some, i had to lower the input down to -0.5 so the "inter-sample clip indicators" (the white light in the plug) would stop flashing...

[starfighter]

Originally Posted by Torg

Could you elaborate more on that??

Cause i just ran a few tracks mastered to 0dBFs(not made by me) through the SSL meter and in some, i had to lower the input down to -0.5 so the "inter-sample clip indicators" (the white light in the plug) would stop flashing...

Doesn't that pretty much explain it?

Not that I have ever used the SSL meter, but...

Anything from -0.1 to -0.5 depending on the accuracy of the meter you use in you plugs or whatever...

morten

[Torg]

Doesn't that pretty much explain it?

Not really.

As far as i know in theory you can have up to 3dB of inter sample peak... (therefore you would need, in this case, to go down to -3.0dBFs)

So i quite didnt get you saying "With -0.1db you make sure there's no clipping"... like it is a sweet spot.

i just gave the example in my first post to simplify.

i was not looking for an explanation of why it was happening.

ed

[cichy]

some additional reading : Tech library

[thermos]

For the sake of argument, etc.

It seems like a LOT of MEs could care less about inter sample peaks. I've imported quite a few masters from the heavies that are full of inter sample peaks (most presumably from A/D clipping). Most of the time, if the master isn't cut scorching hot, you probably won't hear it. But if it is super hot, then the become pretty audible imo. I've also noticed a lot of masters burned at 0dbfs, which is always surprising. But of course, lowering it to -.1 isn't going to save you from inter sample peaks.

Also, most limiters don't catch inter sample peaks as far as I understand. Izotope Ozone, with the inter sample peaks option checked will catch them, but at the expense of a little fidelity. Anyway, just learning about this stuff myself, interested in why people care/don't care about it.

[Axon]

IIRC, the original reasons to avoid intersample peaks had more to do with keeping DACs from catching on fire (!). There was a fear that the amplifiers could latch up if the signal got too close to/over the rails, which would short out the chip.

Given that kind of prohibition in the early days it should not be surprising that standards have grown more lax since then.

[thermos]

Originally Posted by Axon

IIRC, the original reasons to avoid intersample peaks had more to do with keeping DACs from catching on fire (!). There was a fear that the amplifiers could latch up if the signal got too close to/over the rails, which would short out the chip.

Given that kind of prohibition in the early days it should not be surprising that standards have grown more lax since then.

Nice! But won't certain consumer grade D/A exhibit distortion when there are a lot of inter sample peaks present?

[Bob Yordan]

Hiya

I usually put the UAD PL to output at -0.01, then the peaks shows up as
-0.0 in WL6.

Mostly because I dont want WL6 to do the final clipping.

I have used a inter sample peaks detector during development of my plugs and managed to minimize and remove 'some' detections from mixes that contained it. I now have to tweak the detector plug to almost max sensitivity to make it register any ISP on processed mixes.

Dont know if that is good or bad though?

The thing/point of my post is that mixes can include a lot of them and mastering is no secure way of getting rid of them as I see it.

[Torg]

It seems like a LOT of MEs could care less about inter sample peaks.

If any pros out there care to share their approach into how they handle the "inter sample peaks" problem we would difinitely like to hear...

This is not about theory(which is well explained in the articles posted in this thread), but what do YOU actually do about it ...

[Gary Gray]

Since I made my original post I've done a bit of checking. First of all here's what I've done:
1.) I've assembled my cd as a montage in WaveLab 6.0 and inserted a tc6000 on the output bus.
2.) Only one engine in use with Brickwall Limit 0dBFS.
3.) Threshold set to 0dBFS
I rendered the cd and noted that the limiter was active perhaps ten times throughout the whole record. I then performed a global analysis of the rendered wave file. The maximum peak for the whole thing was -0.02 db. I looks like the limiter leaves a tiny cushion when it gets to the wall. I haven't observed any artifacts at the limited spots, it's quite transparent in fact. i should mention that the material is orchestral so there is minimal dynamic processing overall and the level of the cd is fairly conservative. No loudness wars here! I just thought it was interesting that the tc seems to leave a bit of a buffer zone.

gary

[Darius van H]

Another very good reason to keep it below full scale is that many clients load the mastered tracks into their DAW, and if they see the red light go on they might get jumpy.

[24-96 Mastering]

Originally Posted by Darius van H

Another very good reason to keep it below full scale is that many clients load the mastered tracks into their DAW, and if they see the red light go on they might get jumpy.

Some go jumpy if they don't see red lights...

[starfighter]

Originally Posted by 24-96 Mastering

Some go jumpy if they don't see red lights...

Nailed to the floor....

morten

[technomonster]

what are intersample peaks by the way you guys.

[Lagerfeldt]

Originally Posted by Darius van H

Another very good reason to keep it below full scale is that many clients load the mastered tracks into their DAW, and if they see the red light go on they might get jumpy.

And yet just about every master from Sterling goes all the way to 0 dBFS. :-)

[jdg]

i put my ceiling at .3, but a label i work with insisted on everything being at zero, so he didn't have to normalize the tracks before ripping to mp3...

u pay by the dB, and if it cost you $1000 for the mastering, .3 is $30!!!

[Ravian]

correct me if aim wrong.

the problem is when the output of a digital limiter is set to 0dbfs,
the audio source can have overshoots,and can make a cd player skip,or not play at all.

[Lagerfeldt]

I haven't personally encountered a cd player that skipped or refused playing because of a full scale signal.

They could clip though, but that's something different.

[Nordenstam]

Hi folks!

Originally Posted by technomonster

what are intersample peaks by the way you guys.

It's actually intersample everything, but the reason most people care are the peaks. The 44100 (or whatever) sample dots in the digital system are conceptual points of infinitely small time, and therefore nonexistent. The actual/reproduced signal is everywhere else than the sample points. An intersample peak meter helps keep a check on the real peak level.



The white dots are sample points and the blue line is a reconstructed waveform. (thanx to izotope RX for the neat waveform view)
Note that the right hand side peak is the highest, even though the sample points doesn't even reach the ceiling of the others. This is only a tenth or two of a dB above zero though.

Outright squashing will give way different result. A quick and dirty handful of 3dB digital clipping applied to an already brick wall limited track produced 1.4dB of intersample peaks:


This is the lure of clipping/hard limiting. It doesn't shave off the whole peaks, only parts of it. Nearly half the peak is still there, "hidden" above 0dBFS. 1.4dB of overshot will make for quite a lot of trouble in normal playback equipment.

As noted by Thermos, ceilings of .3 or .1 etc dBFS are arbitrary. What matters are the real/intersample/reconstructed peaks. Clean processing only produces fractions of dB's overshot, aggressive loudness treatments can make for several dB's, and worst case signals can in some situations give two digit numbers of dB's overshot! Sample points meters are like analogue peak level meters, they're too slow to catch the smallest and highest peaks.

More info in this thread.

Originally Posted by Torg

If any pros out there care to share their approach into how they handle the "inter sample peaks" problem we would difinitely like to hear... This is not about theory(which is well explained in the articles posted in this thread), but what do YOU actually do about it ...

I use the IS peak meter like the other metering/test/visualizing stuff - mostly to do technical tests, only occasionally throwing a glance at it while mastering. The interesting thing is the difference between IS peak and sample point peaks. Some processing options are more prone to creating IS peaks than others. By observing the meter while fooling around and testing, I can quickly find how the processing fares in regards to the IS peaks. Some things, like clipping, are guaranteed to create IS peaks - so I step very cautiously if that is going to be used. Other stuff, like harmonic boosting (ala HEDD) will not create much IS peaks at all. Some processing options falls in between those extremes. So to me - the most useful aspect of the IS peak meter is to avoid the "bad" processing options in the first place.

How much overload that can be tolerated is highly dependent on the playback system and the musical content. I don't use a cheap DAC to test how it can sound, as there is no "perfect cheap DAC" that will sound bad in the same way as every other cheap DAC. Instead, I let the IS peak meter tell me how much above zero things goes, and from there I try to make a guesstimate as to how that may interfere with the listening joy. If the peaks are very small and/or only occur at noisy parts (like snare drum), some distortion is tolerable IME. If the peak occurs in say a bassy tone, the overload may create an obvious splat of distortion in the playback system. It's guesswork, but the guesswork can at least be quantified.

If the final destination is CD only, IS peaks are less troublesome than if the audio is going to be transfered to other media. In these days, that means psychoacoustic coding. Just about any lossy coding will change the waveform and create a new set of sample points on the output of the decoder. If the signal is free from IS peaks and have about half a dB of headroom, coding and decoding at high rates can often be done without hitting the ceiling. As soon as the bitrate shrinks the peaks will change more, demanding more headroom. The typical case in point is the way loud CD's sitting right at the ceiling with a dB or two of IS peaks. This is guaranteed to create a lot of extra distortion when lossy coding is used. This is IMHO, a contradiction, as the masters are created to sound great on crap systems(no dynamics, little bass, etc), yet it will have distortion beyond the ME's control when being used in the typical crap system playback scenario - lossy coding.

Hope this helps.


Regards,

Andreas Nordenstam

[Lagerfeldt]

Thanks Andreas, yet another great explanation. I've added this and your other answer to my link list on my Danish audio forum Lydmaskinen - Index page under Digital Theory. I hope you keep the screenshots on your server?

[Darius van H]

Originally Posted by Lagerfeldt

And yet just about every master from Sterling goes all the way to 0 dBFS. :-)

I had it several times (people asking why it seemed to be clipping).........so i never go to 0..............normally -0.1 or sometimes -0.3 if it's something acoustic and delicate.

[Torg]

Thanks andreas!

Helped a lot.

I've seen another post from you on older thread that relates to this one so i'll just post below as it may be interesting to many others as well:

Originally Posted by Lupo

Mp3's aren't much easier. Try this: clip something, flat-topping it. Change gain by -3dB. Render this to a Mp3 and decode it back to the wave editor. Check the peak levels! They'll be far above the previous limit of -3dB.

Now think about the battery size of the ipods and other tiny little players. How much headroom there is available in a 1 volt DC portable audio system. Or a regular CD player with about 1VAC max output. Compare this to the typical +-15Volt driven studio gear. Digital can have 'hidden' overshots up to 10dB above the maximum sample value. That's a lot of voltage to ask for.

No sane manufacturer will spend time, energy and money to develop more headroom in their gear when all it takes to playback perfectly fine is a bit of (un)common sense when setting the digital level in the source. Why should gear manufacturers raise the prices of their product to cope with illegal samples? They won't. They spend the finances to get more of whatever they need more of to sell more. Tiny physical size, larger storeage capacity, better hardware protection, better design and so forth.

based on what you said, can you explain what happens to the overload if i dont lower the gain by 3dB??

ok.. it will be the amount that will distort.... but how is this distortion introduced in the mp3 file when its converted?

[24-96 Mastering]

Great post, Lupo. I shall use this in my workshops if someone asks about fs headroom. Is that OK with you?

By the way, if you don't mind a life of repetetive work, bad pay and frustrating burocracy and limited recognition, you really should get into teaching...

[Nordenstam]

Originally Posted by Lagerfeldt

I've added this and your other answer to my link list on my Danish audio forum Lydmaskinen - Index page under Digital Theory. I hope you keep the screenshots on your server?

Cool! Pics are there to stay. (btw; they're on a Danish web hotel )

Originally Posted by Torg

based on what you said, can you explain what happens to the overload if i dont lower the gain by 3dB??

ok.. it will be the amount that will distort.... but how is this distortion introduced in the mp3 file when its converted?

Short story: If the signal is above zero internally and the output is an integer word, it will clip.

The long winded one: The mp3 format itself does not have a specific bit depth. Internally, it's "bit-less" and if floating point, also clipless. From what I've gathered, most coders today use floating point internally. The interesting thing is therefore what comes out of the decoder.

There's a whole slew of stuff going on inside the coder that might change the peak level of the decoded file. First, mp3's are most often low pass filtered(either on it's own or as a part of resampling to lower sample rates like 32KHz). Resampling will place the new sample dots somewhere else on the waveform. Then the signal hits a filter bank that splits it into different bands. If information is altered in the frequency bands (as is the point with the coder), the final waveshape will again change. As noted, if the coder is floating point, none of these effects will make the signal clip internally in the coder. (if not, all hell is loose internally in the coder) But.. When the signal leaves the decoder, it's most usually presented as an integer word of PCM. If there was any sample points in the floating point presentation above zero, they will fail to be output correctly - distorting when they hit the ceiling instead of going above zero. Id est - clipping.

There are some ways to avoid this. Changing gain at the source(limiter output ceiling?) prior to coding is the easiest way but it can be uncertain as the gain change needed is unknown. Changing gain in the encoder is fairly transparent too - (e.g. --scale switch in lame), as this is done in floating point in the coder. Most up to date players supports the Replay Gain function (scaling the signal as it's being decoded) and so the level can be set lower using the APE tags (as opposed to the ID3 tags that contain the usual MP3 descriptions). The free mac/PC MP3Gain application can run a clipping check on MP3 files prior to conversion, lowering the replay gain level til the clipping stops is a sure way to avoid decoding clipping. Another way is to find a decoder that spits out PCM data in floating point format and lower the level before conversion to integer(as must be done before playback).

Originally Posted by 24-96 Mastering

Great post, Lupo. I shall use this in my workshops if someone asks about fs headroom. Is that OK with you?

Of course!

Originally Posted by 24-96 Mastering

By the way, if you don't mind a life of repetetive work, bad pay and frustrating burocracy and limited recognition, you really should get into teaching...

Sounds like a description of being self employed.. So what's the difference? Hehe.. Seriously; have been thinking about that. Maybe... But first; time to study! Am applying for a precourse in math and physics in spring and will hopefully start on electrical engineering next fall. Seems like the perfect compliment to the audio related knowlegde.


Cheers,

Andreas Nordenstam

[Lagerfeldt]

Originally Posted by Darius van H

I had it several times (people asking why it seemed to be clipping).........so i never go to 0..............normally -0.1 or sometimes -0.3 if it's something acoustic and delicate.

-0.12 dBFS here seems to do the trick.

When I try to explain the issue their eyes seem to glaze over.

[StephenMarsh]

FWIW - I stay .2 dB lower on the output of the last digital piece in my chain. Never had problems there before though I have had a client report 'static' with a 0dBFS master that I wasn't able to reproduce in the studio. I chauked it up to cheap playback converters not being as robust as the studio convertors. The little drop fixed it so no worries.

Bottom line - I'd challenge anyone to claim the master is 'too quiet' with such a small difference in level so there's little reason to push the envelope. IMO - If you're concerned, pull it back a touch and don't worry too much about it.

[macc]

I watch the oversampling meters in RME digicheck and turn it down until they stop showing white...

Sometimes it's -0.4dB, does this make me a softie?