Clipping distortion (multiple 0 dBFS samples) on commercial releases, why?

[L5730]

Hi all,

This may or may not have been covered elsewhere but it didn't come up when I searched.

Pre-waffle:
After getting some IEMs a lot of audio nasties have become extremely apparent. The unforgiving nature of such listening equipment means there is no where for 'garbage' to hide and no soft smoothing over. Sure, I notice many different problems on other setups in the home, but the closer and closer the drivers can get to the ear things show up more and more clearly - in my experience. Once the issues are noticed, they also tend to show up in much lesser playback equipment, even the crappy single speaker FM/AM radio in the kitchen!

Waffle:
I've noticed and number of CDs and lossless downloads having multiple adjacent 0 dBFS samples. There used to be 'standards' in place where CD mastering houses (or should I say glass master fabricators?) would reject if 3 or more 0 dBFS samples were found, of course the client could instruct to continue regardless. DACs cannot accurately produce a waveform from this flat-topped theoretical representation, and so are doing something on playback. I wonder how different the results are with different DACs - some DACs have headroom for clipping (or so I read) will they behave differently?

For a quick and really obvious example, take Nickelback's "Silver Side Up" from 2001. There is a nasty quality about the sound that is a product of the clipping distortion (flat-topped waveform). If a DeClip processor is used to 'reconstruct' the waveform, that nasty thin gritty sound disappears and it sounds, in my opinion, how it should do. Now that band and their music is a matter for taste, but the actual audio aberrations are quite apparent.
Turning the file down in the DAW doesn't solve the problem either, as it still has that thin crackle sound. Only reconstructing a waveform with 'real' transients seems to clean it up.

This is a fairly extreme and obvious example, but in the last 2 decades there are numerous examples that show this to a lesser extent. Modern pop is littered with clipped samples and subsequently distorted parts that I feel I need to 'fix' myself to enjoy.

My Question:
Why do many commercial releases (CDs, downloads, anything lossless digital*) contain multiple samples of 0 dBFS?


* vinyl is analogue so will have a reconstructed waveform that differs to the digital counterpart - a DAC stage was involved somewhere, as well as the who vinyl physical mastering and manufacturing process. Lossy digital encoding doesn't really have a bit depth and so the algorithms tend to reconstruct the waveform in a different fashion. Encoding lossy will product something different to the original that the original cannot ever be recovered from. Unfortunately there are also different decoders that do a better or worse job, not necessarily software but hardware implementations do differ in quality for reasons unknown.

Thanks for reading and your thoughts on this.

 

[staticV3]

Why do many commercial releases (CDs, downloads, anything lossless digital*) contain multiple samples of 0 dBFS?

Because loudness is still king and a few clipped samples among the millions that make up a song are largely inconsequential.

 

[DVDdoug]

The Loudness War.

I wonder how different the results are with different DACs - some DACs have headroom for clipping (or so I read) will they behave differently?

The data coming into the DAC is clipped. It's integer so even if the original file/data is in a format that can go over 0dB, the DAC cannot. (Some floating-point ADCs are now available so it may not be long before we have floating-point DACs.)

Some DACs are better at handling inter-sample peaks (AKA "true peaks") where the reconstructed analog goes over 0dB.

For a quick and really obvious example, take Nickelback's "Silver Side Up" from 2001. There is a nasty quality about the sound that is a product of the clipping distortion (flat-topped waveform). If a DeClip processor is used to 'reconstruct' the waveform, that nasty thin gritty sound disappears and it sounds, in my opinion, how it should do.

You had better luck than me. When I tried a clip repair tool once or twice the waveform "looked better" but it still didn't sound good. I may have made an improvement but I didn't bother with A/B comparing since it still sounded bad. (Lucky for m, it wasn't music that I was cared about.) I think there was just too much damage done by the extreme limiting & compression. It probably works better with "clean" clipping but it's impossible to know the original wave height or shape.

Turning the file down in the DAW doesn't solve the problem either,

Right... That doesn't change the wave shape.

* vinyl is analogue so will have a reconstructed waveform that differs to the digital counterpart

Right. The analog cutting & playback process alters the waveform making some peaks higher and some lower. That can "hide" clipping that exists in the master and create better measured dynamics (higher crest factor) without changing the sound of the distortion, or the sound of the dynamics. Of course, vinyl can also add distortion and frequency response variations and it always adds noise.

Encoding lossy will product something different to the original that the original cannot ever be recovered from. Unfortunately there are also different decoders that do a better or worse job, not necessarily software but hardware implementations do differ in quality for reasons unknown.

MP3 encoding does "similar" things as vinyl, making some higher peaks that often go over 0dB, and again "improving" the crest factor. If you play it at "full digital volume" you'll clip your DAC, but as far as I know that slight clipping isn't audible and if you hear a compression artifact it's probably something else.

 

[kemmler3D]

I wonder how different the results are with different DACs - some DACs have headroom for clipping (or so I read) will they behave differently?

There was a thread on this and intersample-overs a while back. I am not sure where it landed. IIRC it was somewhat contentious in that the expected behavior isn't super-well defined, and there was debate on exactly how audible it would or could be in practice, and whether the behavior should be part of the standard test suite.

Why do many commercial releases (CDs, downloads, anything lossless digital*) contain multiple samples of 0 dBFS?

Like @staticV3 said, people all along the production chain tend to favor loudness over dynamic range. Why? My guess - the prevailing opinion is that most people don't have especially good listening equipment at home. This would be a good assumption - the most popular headphones on Amazon by FAR are the cheapest ones.

 

[Sokel]

There was a thread on this and intersample-overs a while back. I am not sure where it landed. IIRC it was somewhat contentious in that the expected behavior isn't super-well defined, and there was debate on exactly how audible it would or could be in practice, and whether the behavior should be part of the standard test suite.


Like @staticV3 said, people all along the production chain tend to favor loudness over dynamic range. Why? My guess - the prevailing opinion is that most people don't have especially good listening equipment at home. This would be a good assumption - the most popular headphones on Amazon by FAR are the cheapest ones.

There are several talking about intersample overs,one of them described by experts is here (down the page):

DSP Measurements and Rising Noise Floor

And an updated summary table based on tests run so far: Michael

www.audiosciencereview.com

and a more closer,here:

Practical example of intersample peak greater than +6 dBFS

I've mentioned it before here on this forum, but it's entirely possible to have music that has intersample peaks beyond +6 dBFS, or one whole bit. It was mentioned again later on this thread by @LTig. Here is a 30-second excerpt of "Get Away" from the Electronic Soundtrack album Code Lyoko by...

www.audiosciencereview.com

 

[L5730]

I did just stumble upon a video mentioning something interesting. It was a company selling their own plugins to do what hardware would have been used for.
Take a project and send it to a DAC, then back in to the DAW via an ADC hard clipping the samples. This approach just seems like the wrong way to create a sound in the digital world - one doesn't know how it will ultimately play back on end user equipment and is it not the job of the mastering engineer to make the best compromise of the material for the potential places it will be played?
The discussion in the video did hit on seeking maximum loudness in the material, so this largely seems a product of loudness wars and not really an honestly real creative choice, I may be wrong or just misunderstanding.

I am not talking about a few ISPs, that is neither here nor there as, yes I would still say it's technically broken digital audio if they exist, but if the transients are fast and few, it's very likely inaudible in the real world.

I am in agreement that 'reconstructing' a waveform by using DeClipping tools is always an approximation and sometimes more harm can be done than good. Without the original non-clipped files, it's all we have. We can also sit there and re-draw the sample points ourselves if we've the time and patients. I have done this to fix glitches, but they were more isolated.
Oversampling does do some things visibly. I will have to listen to see if they make enough of a change. This would be a quick option to get closer to the analogue waveform before hitting the DAC.

Agreed, lossy can and will still clip on playback/decoding, but may do so in a less undesirable way. Contending with lossy artefacts in another issue however. 0 dBFS is still 0 dBFS once decoded, and there's no ignoring that.

I will look into the posted links later.

Thanks to all for your contributions to this thread.