Hidden clipping made visible.

[Netlist]

I accidentally found this website a few weeks ago and, curious to find out how my cd collection would stand the test, I can only confirm what the author wrote.
I ripped a few cd’s with EAC and imported them in CoolEdit.
Many tracks, all from different labels, artists, have clipped samples but don’t show up in the statistics window of Cooledit.
Zooming in, however, clearly shows many flatten tops.
CoolEdit’s clip restoration feature recommends to convert the data to 32bit and run the tool with certain parameters.
Doing so results in a frightening amount of clipped samples that can be brought back into range with the build in hard limiter.
The main question is why the clipping doesn’t show up in the stats window although the clipping is clearly present. A bug or do I overlook something?
Picture below of a 16 bit sample showing the clipped samples.

/Hugo

[Netlist]

Statistics window:

[fmak]

[QUOTE]Originally posted by Netlist
[B]I accidentally found this website a few weeks ago and, curious to find out how my cd collection would stand the test, I can only confirm what the author wrote.
I ripped a few cd’s with EAC and imported them in CoolEdit.
Many tracks, all from different labels, artists, have clipped samples but don’t show up in the statistics window of Cooledit.
Zooming in, however, clearly shows many flatten tops.
CoolEdit’s clip restoration feature recommends to convert the data to 32bit and run the tool with certain parameters.
Doing so results in a frightening amount of clipped samples that can be brought back into range with the build in hard limiter.
The main question is why the clipping doesn’t show up in the stats window although the clipping is clearly present. A bug or do I overlook something?
Picture below of a 16 bit sample showing the clipped samples.
------------------------------------------------------------------------------------

I have found this even on test CDs. This is why I always attenuate level by 1 to 3 dB

[Netlist]

Thanks fmak.
I’m getting a grip on how things behave.
Somewhere along the line, that is the mastering or at my side the ripping process has hard limited the waveform to about -0.01db.
That is just below clipping and probably the reason why the bad samples are not detected.
If things went wrong while ripping the track it can probably be fixed.
If an engineer did a bad job, shame on him.
Attenuating the signal doesn’t really help as the flat lines are not corrected in that process. Only converting the waveform to 32bit and restoring the samples seems to bring some relief. This practice leaves me with peaks of +6 to +8db sometimes. Attenuating the entire wave with –8db or a bit more looks like the best solution I can come up with for now.

/Hugo

[BlackCatSound]

You can't restore it as you don't know whats gone missing.

As the stats show the peak values at 32392 and -32393 these are still within the limits of 16bit PCM. The signal was clipped before it was mastered.

[dr.strangelove3]

Attached is an older elektor design, meant to detect those clippings on cd's.

I sure want to build it some day to test my cd collection.

Sorry, it's only in Dutch.

hmm.... file is too big to post.

Anyone interested, I can send it by mail.

[Pjotr]

Hi Stranglove,

Know that circuit but that looks only at the digital limits. Not much of help because many CD’s do clip below maximum level. Nowadays there are many wave viewers/editors out there to do it more accurately like i.e. Goldwave ( http://www.goldwave.com/ )

Cheers

[darkfenriz]

Look here
The curvature of signal (=1/radius) has a huge peak followed by several zeros when clipping occurs.
This tool can detect clipping, statistics only count the number of maximum and minimum samples.

[Netlist]

Originally posted by BlackCatSound
You can't restore it as you don't know whats gone missing.

Indeed, like every restoration, the result is only an estimation of what once was there.

As the stats show the peak values at 32392 and -32393 these are still within the limits of 16bit PCM. The signal was clipped before it was mastered.

Correct, that's what I found out yesterday. I don’t have to blame EAC.

/Hugo

[Netlist]

This math looks nice (and complicated to me) but how to integrate it in a working application?

/Hugo