Skip to main content

Notice

Please be aware that much of the software linked to or mentioned on this forum is niche and therefore infrequently downloaded. Lots of anti-virus scanners and so-called malware detectors like to flag infrequently downloaded software as bad until it is either downloaded enough times, or its developer actually bothers with getting each individual release allow listed by every single AV vendor. You can do many people a great favor when encountering such a "problem" example by submitting them to your AV vendor for examination. For almost everything on this forum, it is a false positive.
Topic: clipped sample and its declipped version (Read 16345 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

clipped sample and its declipped version

Here are two very short samples (6 seconds) proving drastically that clip restauration can improve the sound quality. (It's the first time that I ever upload anything, so please forgive me if something is wrong.)

The samples are from "It's all coming back to me now * Celine Dion" from the CD "Falling into you"

The first is the unchanged original. The maximum level is about -2dB. First they clip it and then they not even use the whole available dynamic range - insane. The clipping is the original clipping (not created by myself just for testing).
The second sample is restored by myself and, of course, much quieter. At an other portion of the audio it is normalized to 0dB. Feel free to replaygain/wavgain the files, if you want to.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #1
the second sample
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #2
Here are further two 30s-samples showing two different ways of mixing & mastering. The song is "Crazy for you * Madonna". The first sample is from the album "The immaculate collection" (1990), the second from "Something to remember" (1995). Both samples seem to have the same source single tracks, but the newer one is clipressed (as well compressed as clipped) and sounds some kind of boomy and tiring. (Similar to the samples Pio had posted some time ago, but here both are studio versions.) The dynamic version is unchanged, the clipressed is normalized down about (only!) 2dB according to wavegain.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #3
second sample
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #4
I'm a little bit desperate about this thread reaching only few attention. Is there a special format that is generally preferred for downloading ? I'd like to get at least one reply confirming that the samples can be downloaded and decoded without any errors. Anyway, I'm adding flac versions for the original samples.

edit: removed the flac attachments in order to free up disc space
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #5
second sample
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #6
The difference is most noticeable with the Madonna samples... Didn't even have to put on my Sennheisers...  Unfortunately, I don't see them changing their mixing processes any time soon, because all these new clipped CDs are made to sound "nice" on low-end systems... And that's related tol the issue of perceived high quality... Rant rant...

Anyway, your uploads are working just fine!

P.S.: Regarding your question, if you post samples, it's best to use a lossless codec...
Jurg

clipped sample and its declipped version

Reply #7
Quote
P.S.: Regarding your question, if you post samples, it's best to use a lossless codec...

My samples have to fit on a floppy disc. This offers me at maximum about 12 seconds of losslessly compressed music (extra high ape)
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #8
Thanks for posting the samples, precisionist. The clip restoration improved the sound quite nicely indeed.

What declipping tool/procedure did you use?
Over thinking, over analyzing separates the body from the mind.

clipped sample and its declipped version

Reply #9
Quote
Thanks for posting the samples, precisionist. The clip restoration improved the sound quite nicely indeed.

What declipping tool/procedure did you use?

I must admit that it is a very extreme example. The results are often much worse.
The program is CEP. The two important things are:
-In order to detect the clipping in the best way as possible, the sample mustn't be changed before declipping. But you need headroom above 0dB for the added peaks. Therefore you convert up from 16bit to a float format.
-After declipping, normalize down to 0dB and then convert back to 16bit using dither. Do [span style='font-size:14pt;line-height:100%']not[/span] apply limiters/compressors to bring the peaks back into normal range, you'd make things worse. If the peaks are unnaturally high, you should rather clip away some of the added decibels.
But, if the waveform doesn't look "natural" after clip restauration, like it could have been before clipping, there's most likely too much compression/limiting, making clip restauration impossible.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #10
very nice restoration. in some cases clipped peak restoration programs will work well, especially on vocal samples. on the r3mix boards i posted once a french vocal sample, clipped (over 0 dB FS) heavier than the sample you posted, restored with sonic foundry NR2 and the results were very good. but in some cases, instrumentals, synths, drums, these clipped restoration tools don't work as well.
Be healthy, be kind, grow rich and prosper

clipped sample and its declipped version

Reply #11
Listened to the Madonna samples.  You're right that the newer version is inferior sounding, but I don't think the loudness has anything to do with it.  I couldn't hear any distortion, and the clipping I saw when I opened it up in SoundForge was hardly what I'd call "nasty".  As for the compression, it seemed that the level had nothing to do with it as what they went for left plently of headroom for the song's dynamics; rather the compression was added simply to attain a more "trendy" sound: maxed-out bass that overwhelms the mix rather than supporting it and a generally smoothed-over sound for lo-fi systems and busy traffic.

To test this I went into SoundForge and ran it through L2, then did another version that was simply cut off, both at -2db.  Both versions sounded far superior to the Something to Remember version despite have roughly the same amount of headroom, and both stacked up rather well with the original.  In fact, the hard-clipped version didn't seem to clip any more than the "professional" one (I counted 8 consecutive clipped samples at most (on a snare hit no less), certainly not something I'd cry about).

So yeah, this sounds more like the mastering engineer simply being careless or lacking in taste than being told at gunpoint to "make this louder".

clipped sample and its declipped version

Reply #12
Quote
So yeah, this sounds more like the mastering engineer simply being careless or lacking in taste than being told at gunpoint to "make this louder".


Your result sounds sensible. I know there are of course much worse examples. The Madonna samples are not to proove the loudness war, they just show a good and a bad way of mixing & mastering.

I'm very interested in your restauration processes, it seems that I don't understand them. What do you mean with "L2" ? Considering this undefinable mixture of clipping and compression, I have no idea what I could do to make it sound better. (in contrast to the Celine Dion sample)
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #13
I think this is what Kuuenbu is talking about.

Edit: Here is a guide where L2 is used (combined with CEP's clip restoration) to remaster the clipped "Vapor Trails" album (by Rush).

This part explains what L2 does:
Quote
It is a DirectX plug-in that utilizes "intelligent" look-ahead algorithms to detect upcoming peaks in the audio that go out-of-bounds.  It allows for impressive volume maximizing and proper peak-limiting without introducing distortion in the process.
Over thinking, over analyzing separates the body from the mind.

clipped sample and its declipped version

Reply #14
Thank you, PoisonDan

I've misunderstood what you said, Kuuenbu. You took the unclipressed version and clipressed it to the same peak-to-average ratio, without lowering the sound quality.

No offence, but the guy who wrote the above guide has poor knowledge about what he's talking about.
Applying DC offet adjustment before declipping runs a super low-pass filter over the waveform and therefore modifies the clipped areas, so it is more difficult to detect them. It is specifically said in CEP's help that this must be avoided.
The reason for clipression is that music is just unnaturally loud. There is plenty of footroom for normalizing down after the declipping process. Applying a limiter/compressor (even a sensible one) to reduce the added peaks destroys half of the improvement.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #15
It might be interesting to attempt clip restoration on one of the worst offenders in the loudness race which was originally pointed out here by One Crazy German - the Iggy Pop & The Stooges "Raw Power" 1997 remix.

  Unfortunately I do not have audition and could not find a standalone CEP tool. If someone would point me too one that would be marvelous, but I'll upload a section for the common good, or bad 

  I cannot help but wonder how a peak restoration tool could possibly make the album sound worse.

  Beautiful, uploads appear to work now. Those seven seconds are pretty representative

clipped sample and its declipped version

Reply #16
Well.. I gave it my best.. but CEP failed miserably.. this baby's just way too clipped.. Trackgain of -17.0 db!  More than one in five samples are clipped.   

When there's a HUGE clipped area, CEP doesn't really know what to put in the restored area, so it creates a giant round hump of a soundwave, which winds up sounding like a pop on dirty vinyl.  Have a listen.

clipped sample and its declipped version

Reply #17
Yes     
This 'pop' problem is well-known to me. CEP is hopelessly overtaxed by this sample (it's not what I ...ahem... prefer to listen to). Using some audio technician tricks I was able to make it sounding a bit better, though, but nothing that satisfied me. CEP tends to overestimate the height of the additional peaks, especially if the clipping is so excessive.
This sample is actually the worst one I've ever seen and I was really shocked by the two rectangels. The clipping is different than on all other examples so far: It is deliberate; applied like an effect. On all other samples it was just the (hopefully unwanted) result of a volume set too high.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #18
Sorry, I'm having problems with uploading. I didn't want to make this post.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #19
Quote
The samples are from "It's all coming back to me now * Celine Dion" from the CD "Falling into you"

I just found this thread via a Google search on clip restoration, lol. I started doing some research into "the loudness wars" since I keep running into the term here and there. That wasn't the only reason though. I ripped this very CD the other day and noticed frequent clipping in the resulting MP3's. Checked the CD's, and it's there too, so it wasn't a LAME problem. Ah, the wonders of Sennheiser headphones. So anyways, I'm looking into ways to accomplish "clip restoration". Maybe someone could outline what is required (is Adobe Audition any good?), and more importantly the proper steps to take? That would be really great, thanks! Take care all!

clipped sample and its declipped version

Reply #20
Quote
So anyways, I'm looking into ways to accomplish "clip restoration". Maybe someone could outline what is required (is Adobe Audition any good?), and more importantly the proper steps to take? That would be really great, thanks! Take care all!


I think I had already given an answer in this post above... Ask, if it's not enough or not understanable.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #21
Here is my attempt of restoring the worst-ever-seen sample posted by Audible!.

Why did this last soooooooooooooo long ?
I had already done this restauration in July, but some things (especially the forum software) prevented me from uploading. Here's what I still remember:

Thanks to some tricks I was able to isolate the added cliprestored material in order to apply dynamics compression only to the added peaks, not to the added peaks + the original flattops below them. after that I mixpasted (added) the compressed peaks over the original clipped sample. This procedure eliminated most of the rumbling sound caused by the clip restauration and left the reduced distortions untouched; I think it sounds a bit better than the original.
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #22
Quote
It might be interesting to attempt clip restoration on one of the worst offenders in the loudness race which was originally pointed out here by One Crazy German - the Iggy Pop & The Stooges "Raw Power" 1997 remix.

While it could certainly be considered an offender, from what I heard it had absolutely nothing to do with the loudness race.  Supposedly Iggy made it as loud as possible because he wanted to trade one lo-fi style for another rather than giving in to industry pressure to release a "cleaned-up" version.  In other words, the obscene levels on the CD were done as a middle finger to the record labels and mundane audio trends, as well as well to fill Iggy's warped-out vision.

Also the CD was released in 1997, back when squashed-sounding CDs were the exception rather than the rule, and there was no way you could get away with something along the lines of Californication or Borrowed Heaven—at least in the mainstream.

clipped sample and its declipped version

Reply #23
Post belongs to this thread.
Screenshots we see usually show the way it shouldn't be, so I uploaded 2 positive examples showing my personal reference.
wavegain stats:
Code: [Select]
   Gain   |  Peak  | Scale | New Peak | Track
---------------------------------------------
 +0.86 dB |  32717 |  1.10 |    36122 | You'll be the one who'll lose
 +0.13 dB |  32751 |  1.02 |    33245 | Forever

Note that both tracks peak very near to 0dB so they actually use their roughly 20dB headroom. (I had to disable the clipping prevention in wavegain.)
"You'll be the one who'll lose*Kim Wilde" is from the CD "Close" and had not even a single clipped and/or compressed sample. I normalized it to 0dB. The single high peaks you see are additional dynamic drums.
"Forever*Mariah Carey" is from "Daydream" and the cliprestored version.

PS: First time I offer images, any tips appreciated...
I know that I know nothing. But how can I then know that ?

clipped sample and its declipped version

Reply #24
Quote
PS: First time I offer images, any tips appreciated...
[a href="index.php?act=findpost&pid=261058"][{POST_SNAPBACK}][/a]

BMP IST DEATH!!1 

You'd save HA a huge amount of bandwidth if you'd use PNG instead...
Over thinking, over analyzing separates the body from the mind.

 
SimplePortal 1.0.0 RC1 © 2008-2021