Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Can this be right? (Read 14173 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Can this be right?

These are the statistics from CoolEdit 2000 for "Blues By Five", from the Miles Davis album Cookin'.
Code: [Select]
Davis, Miles\Cookin'\Blues By Five.wav

                   Left       Right
Min Sample Value:  -32768     -32768
Max Sample Value:  32767      32767
Peak Amplitude:    0 dB       0 dB
Possibly Clipped:  1058       1174
DC Offset:         .024       .038
Minimum RMS Power:    -74.3 dB   -70.43 dB
Maximum RMS Power: -5.77 dB   -5.7 dB
Average RMS Power:    -17.72 dB  -17.65 dB
Total RMS Power:   -16.66 dB  -16.59 dB

Using RMS Window of 50 ms



This track is from the 1999 "50th Anniversary - Special Commemorative Edition", mastered with the "20bit K2 Super Coding System" (whatever that is).



Can these numbers be right?
Can I trust the accuracy of CoolEdit?

All the tracks on this album report abnormally high numbers of "possibly clipped samples".

What's going on here?

I find it hard to believe that a Jazz label like Prestige would allow this to happen on a "special edition" of a classic album.

Any thoughts?

~esa



:edit: exchange image for text

Can this be right?

Reply #1
As minimum sample value is very low and maximum sample value is very high, I'd be very surprised if there weren't any clipping.

What do you see when you zoom in on the waveform? Are the peaks clipped?

Can this be right?

Reply #2
Can you hear clipping?
daefeatures.co.uk

Can this be right?

Reply #3
Just out of interest what are the replaygain values like for this CD?

edit:  It would be blasphemy to master Miles Davis poorly.
gentoo ~amd64 + layman | ncmpcpp/mpd | wavpack + vorbis + lame

Can this be right?

Reply #4
It is blasphemy to master anything poorly.
Average RMS is roughly -17db, so the replaygain values will be roughly -3dB. That is often enough headroom, but obviously not for this record. You can surely trust Cool Edit, this record is clipped.

I can clearly imagine how this could have happened: The record level was adjusted too high by accident. Sloppyness.
Great to see another clipped classical record ! (Of course I don't want anyone to suffer from clipping !)
I know that I know nothing. But how can I then know that ?

Can this be right?

Reply #5
Quote
What do you see when you zoom in on the waveform? Are the peaks clipped?
Most definitely.  (I was going to post some JPGs for review, but my server is down.  Perhaps later today...)

 
It's getting so I don't even want to buy new CDs anymore; they all seem to have the "clipression" issue.  I guess the only other option is to buy MoFi stuff; I've never seen/heard one of their products clip.



:edit: image links

Can this be right?

Reply #6
Quote
It is blasphemy to master anything poorly.
[a href="index.php?act=findpost&pid=259879"][{POST_SNAPBACK}][/a]

I disagree here.  I couldn't care less how Brittany Spears or the Backstreet boys CDs are mastered.  In fact, IMO, about 90% of what is played on MTV/Much Music these days deserves to be clipressed.

edit: -6 RG values are pretty good for most new recordings but, most of the older pressings of Miles Davis' CDs tend to have positive RG values, -6dB definately means too little dynamic range for Miles Davis.
gentoo ~amd64 + layman | ncmpcpp/mpd | wavpack + vorbis + lame

Can this be right?

Reply #7
Quote
I couldn't care less how Brittany Spears or the Backstreet boys CDs are mastered.  In fact, IMO, about 90% of what is played on MTV/Much Music these days deserves to be clipressed.
The only problem with that is, we'll end up with an entire generation of "music morons" who think that's how music is supposed to sound...

Can this be right?

Reply #8
Quote
Quote
You can surely trust Cool Edit, this record is clipped.
Thanks for the confirmation.  I do have a follow-up question, though...
When I check the file in WaveRepair (v4.8.7 - with the Clip Detection options set to 1 sample @ 0.0db), it reports 649 clipped samples.
Any reason for the discrepancy?
Is it just that the two programs use different algorithms?
Which program is likely to be more accurate?
[{POST_SNAPBACK}][/a]

Rather than write it out again (and possibly bore people who have seen it before), I'll point out that I summarised how Wave Repair decides if there is clipping in [a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=29900]this thread[/url]. The weird thing is that in your case, CoolEdit found more clipped samples than Wave Repair, whereas in the other thread it found less. So I am a little perplexed as to how CoolEdit decides what constitutes clipping.

I am the first to acknowledge that CoolEdit is a great program, and would not expect it to get this wrong. But then as the author of Wave Repair I'm fairly confident that it will have done what it's supposed to. So I can only assume that CoolEdit's algorithm for clip detection is fundamentally different to Wave Repair's.

Can this be right?

Reply #9
Quote
In fact, IMO, about 90% of what is played on MTV/Much Music these days deserves to be clipressed.

The "I don't care unless it affects me" attitude is wrong.
daefeatures.co.uk

Can this be right?

Reply #10
Quote
Quote
Can you hear clipping?
Occasionally - I certainly don't hear thousands, though.
]


If they are consecutive you won't hear them as seperate clips.

Can this be right?

Reply #11
Quote
The weird thing is that in your case, CoolEdit found more clipped samples than Wave Repair, whereas in the other thread it found less. So I am a little perplexed as to how CoolEdit decides what constitutes clipping.
Exactly.  I even set the Clip Detection options in WaveRepair to 1 sample @ 0.0db.
That's why I restated the question here.  (I had not forgotten our earlier discussion.  )
Sometimes WaveRepair finds more clipping; sometimes CoolEdit finds more.  It's very odd.


BTW, thanks for the excellent program, cliveb..!  I'm using it quite a lot, now.

Can this be right?

Reply #12
Quote
If they are consecutive you won't hear them as seperate clips.
That makes sense...  thanks, Garf.

Can this be right?

Reply #13
[off-topic]
Quote
Quote
In fact, IMO, about 90% of what is played on MTV/Much Music these days deserves to be clipressed.

The "I don't care unless it affects me" attitude is wrong.
[a href="index.php?act=findpost&pid=259959"][{POST_SNAPBACK}][/a]

Thats not the attitude I have.  I was just expressing an opinion.  My opinion is that ~90% of popular music has no artistic merit, however some of it does and I do like a considerable amount of music that ends up being clipressed.  Thus the problem does affect me too.  Since I do not believe that a large amount of music in the mainstream these days is worth listening to, I do not consider it "a shame" that this music is mastered poorly.  Certainly the prevalence of poor mastering is a problem and I would ultimately like to see it eliminated, even in music I do not appreciate but I don't now, and never will, consider it a "shame" that Brittany Spears' music is mastered poorly.  I can appreciate where you are coming from with this point but I don't feel that it applies to me and I do feel that it was unnecessarily vindictive.[/off-topic]
gentoo ~amd64 + layman | ncmpcpp/mpd | wavpack + vorbis + lame

Can this be right?

Reply #14
Quote
Quote
I couldn't care less how Brittany Spears or the Backstreet boys CDs are mastered.  In fact, IMO, about 90% of what is played on MTV/Much Music these days deserves to be clipressed.
The only problem with that is, we'll end up with an entire generation of "music morons" who think that's how music is supposed to sound...

[a href="index.php?act=findpost&pid=259955"][{POST_SNAPBACK}][/a]



Sad to say ('cos the music is great), the album your avatar image is taken from is one of the more poorly mastered CDs I've bought in the last year or three...what was Prince thinking? I'm a bit afraid to see what some of those Rainbow Children tracks would look like in Cool Edit.

Can this be right?

Reply #15
Quote
I even set the Clip Detection options in WaveRepair to 1 sample @ 0.0db.
[a href="index.php?act=findpost&pid=259964"][{POST_SNAPBACK}][/a]

Can I just point out that setting clipping detection to only 1 sample probably isn't very informative. A properly mastered CD could very well hit 0dB on isolated samples without clipping. Similarly, if you take an unclipped WAV file and normalise it to 0dB, you'll get at least one 0dB sample. It probably makes more sense to set the number of samples for clipping to 2.

Can this be right?

Reply #16
Quote
Sad to say ('cos the music is great), the album your avatar image is taken from is one of the more poorly mastered CDs I've bought in the last year or three...what was Prince thinking? I'm a bit afraid to see what some of those Rainbow Children tracks would look like in Cool Edit.
I know; it's really bad.
CoolEdit and WaveRepair don't report any clipping on The Rainbow Children, but, for the most part, the wave forms look like solid green bars in CE - especially The Everlasting Now.
Very sad because, as you've said, the music is great.
 


Quote
Can I just point out that setting clipping detection to only 1 sample probably isn't very informative. A properly mastered CD could very well hit 0dB on isolated samples without clipping. Similarly, if you take an unclipped WAV file and normalise it to 0dB, you'll get at least one 0dB sample. It probably makes more sense to set the number of samples for clipping to 2.
Understood.
I just set it at 1 sample to see if it would then report the same number of clipping instances as CoolEdit.  I was experimenting a bit to see if I could find out how CE calculates clipping - with no luck.  But, I generally leave the setting at the default value (4).

(Set at 2 samples, WaveRepair reports 393 instances of clipping; at 4 samples (the default value), it reports only 15.  This, compared with CE's 1,058 (left) and 1,174 (right), is what made me want to see what WR would report using just 1 sample (694).)

Can this be right?

Reply #17
I am appalled by reading this thread.  Soem greedy record company re-re-re-releases material, completely ruins it with a lousy mastering job, and sells it to unwitting customers who are impressed with the technojargon on the cover?  This is one of my favorite albums too...I'm glad I don't have this version of it....

Can this be right?

Reply #18
Quote
I am appalled by reading this thread.  Soem greedy record company re-re-re-releases material, completely ruins it with a lousy mastering job, and sells it to unwitting customers who are impressed with the technojargon on the cover?
And it's getting more and more rampant.
They even screwed up the 30th-anniversary reissue of Pink Floyd's Dark Side Of The Moon - though it has far less clipping than Miles' Cookin'.

Looks like MoFi is our last refuge....

Can this be right?

Reply #19
Wow....I have that version of Dark Side of the Moon, but never listened to the CD layer.  Seems like I'm OK so long as I stick to the SACD layer, but I'm still a bit distressed that such recordings, which deserve to be in the hands on none but the best audio engineers, could have fallen victim to such desecration.

Can this be right?

Reply #20
As far as I know and from my tests, CEP counts the samples with absolute value 32767 or 32768 and their number are the "possibly clipped" samples. Anyway, these are very poor methods and I don't use them. I decide the clipression status of any audio by full and precise zooming, by hearing and from my rich experience.
Cliveb, as soon as there are at least 2 samples at the same (or at +/- 1 the same) value in a waveform, probability of clipping (or limiting) is 99,999...%. Peaks have to be very different.

music_man,
be careful to not make me angry. You can't afford to not care about something you don't like whether it's clipressed or not. There's too much danger that your own beloved music might also become clipressed. This thread just shows that clipressed classical tends to become more frequent.
For every guy saying "Techno is really worth being well-mastered, but I don't care about metal." you can find someone other: "Squash techno to death, but metal is great music."  Just an example. Please don't call anything which someone calls music non-artistic.
I know that I know nothing. But how can I then know that ?

Can this be right?

Reply #21
Quote
I just set it at 1 sample to see if it would then report the same number of clipping instances as CoolEdit.  I was experimenting a bit to see if I could find out how CE calculates clipping - with no luck.  But, I generally leave the setting at the default value (4).

(Set at 2 samples, WaveRepair reports 393 instances of clipping; at 4 samples (the default value), it reports only 15.  This, compared with CE's 1,058 (left) and 1,174 (right), is what made me want to see what WR would report using just 1 sample (694).)
[a href="index.php?act=findpost&pid=260070"][{POST_SNAPBACK}][/a]

OK, I've been doing some experiments and I believe I know what causes the difference.

Firstly, I think CoolEdit just counts the absolute number of samples at maximum possible value; I see that precisionist seems to have deduced the same thing. Wave Repair doesn't work like this. If it sees a consecutive number of samples greater than the defined number (default 4), it regards this as *one* example of clipping. So if you have, say, 7 maximum samples in a row, CoolEdit will report 7 potential clips, while Wave Repair will report just one (and place a marker at the starting point). Even if you set the number of samples for Wave Repair to use down to 1, it still only counts 1 clip for any number of consecutive samples at max: it waits for the waveform to drop back down before looking for the next clip. You can see this if you zoom right in on a clip mark.

So this explains why CoolEdit reports significantly more clips than Wave Repair does: CoolEdit reports number of individual clipped samples, whereas Wave Repair reports number of flattened wavetops. And there's something else we can deduce from this. The greater the difference in the number of clippings reported by the two programs, the wider each individual clip is likely to be, and therefore the more extreme the clipping.

Secondly, I noticed that CoolEdit failed to report a few clips that Wave Repair picked up, and this is because those samples happened to be precisely -32767. As you know, the actual range of sample values goes from -32768 to +32767. Wave Repair regards consecutive samples at -32767 as a possible clip, whereas it seems that CoolEdit requires -32768.

Can this be right?

Reply #22
Quote
Quote
Sad to say ('cos the music is great), the album your avatar image is taken from is one of the more poorly mastered CDs I've bought in the last year or three...what was Prince thinking? I'm a bit afraid to see what some of those Rainbow Children tracks would look like in Cool Edit.
I know; it's really bad.
CoolEdit and WaveRepair don't report any clipping on The Rainbow Children, but, for the most part, the wave forms look like solid green bars in CE - especially The Everlasting Now.
Very sad because, as you've said, the music is great.
 



I'm curious to know generally, does a 'green bar' that doesn't generate clipping reports, come from backing off levels on an already-clipped file, or from extensive limiting?

Can this be right?

Reply #23
Quote
...if you have, say, 7 maximum samples in a row, CoolEdit will report 7 potential clips, while Wave Repair will report just one...

So this explains why CoolEdit reports significantly more clips than Wave Repair does: CoolEdit reports number of individual clipped samples, whereas Wave Repair reports number of flattened wavetops.
OK, got it.


Quote
Wave Repair regards consecutive samples at -32767 as a possible clip, whereas it seems that CoolEdit requires -32768.
Does this explain the instances in which WaveRepair report significantly more clipped samples than CoolEdit?

Fiona Apple's "Criminal", from the album Tidal, for example:
CoolEdit reports "0 possibly clipped samples", whereas WaveRepair reports "clipping found at 368 places" (with clip detection set at the default 4 samples). 

It's obvious that this track clips more than the local barber, but since CE reports -32767 as the minimum sample value (L&R), the clipping seems to have escaped CE's notice.  It looks like you're right about CE requiring -32768 before reporting a clip.

Quote
And there's something else we can deduce from this. The greater the difference in the number of clippings reported by the two programs, the wider each individual clip is likely to be, and therefore the more extreme the clipping.
Good to know!


Thanks for all the input, guys.  You're really helping me suss this out.

Can this be right?

Reply #24
Quote
I'm curious to know generally, does a 'green bar' that doesn't generate clipping reports, come from backing off levels on an already-clipped file, or from extensive limiting?
AFAIK, this could only be the result of massive limiting/compression.