Skip to main content


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: declicked to death? (Read 5098 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

declicked to death?

I am hoping that a number of people experienced in ABX testing will indulge this inquiry. Attached are two samples, one treated, one not. The treatment is something I’ve worked out for from-LP recordings that need fairly extreme help. It is not something I use on the majority of my transfers. It is essentially click and crackle repair, so my sample has many transients, but I use it mainly for other benefits.

I’ve never read any mention of this technique used in this form, but that doesn’t mean it might not be a secret weapon of most professionals. It seems too simple and too effective for those “additional benefits” to have been over looked. The effectiveness, for certain kinds of problems, is unquestionable, but at what cost?

This audio was not perceptually encoded so the usual artifact detection might not be a way to differentiate the sounds. Possibly differences will be more obvious from the speakers of a high quality audio system, or studio type monitors, than by using headphones.  Since software treatment of the two channels is independent, I am only downloading one channel, in order to save space and especially download time.I would be happy to try with different kinds of music if people believe that might be more revealing.

The original sample needed nothing. The treated version had a software count of 36,102 clicks repaired for the slightly less than 30 seconds of one channel. The questions are
(1) are the before and after versions distinguishable from each other?
(2) if so, what is the your subjective evaluation of the difference(s)?

declicked to death?

Reply #1
Nice surprise to hear some Stravinsky here
Version B is clear and crisp, version A is too dull. Does your declicking application allow you to hear the difference signal (clicks only)? It might reveal that you are removing parts of the signal that you'd rather keep.
Acoustical recordings like this one are full of noises that are produced by the instruments and the musicians. If you remove too much it can result in a sterile, unnatural recording. It seems to me that your declicking settings are too sensitive.

declicked to death?

Reply #2
Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2010/05/11 13:20:34

File A: D:\audio\A.flac
File B: D:\audio\B.flac

13:20:34 : Test started.
13:20:40 : 01/01  50.0%
13:20:46 : 02/02  25.0%
13:20:50 : 03/03  12.5%
13:20:53 : 04/04  6.3%
13:20:56 : 05/05  3.1%
13:21:00 : 06/06  1.6%
13:21:13 : 07/07  0.8%
13:21:16 : 08/08  0.4%
13:21:21 : Test finished.

Total: 8/8 (0.4%)

It sounds like dynamic high frequency attenuation (lower level signal = more high frequency content removed).

However, it doesn't sound like 36,102 clicks repaired!


declicked to death?

Reply #3
Thanks for replying. Given your observations, I have another inquiry, although answering it could mean more effort than would interest anyone: can a correction be made, post declicking? First, however, a number of minor observations.

I take the point about the declicking settings being too sensitive for the test material, but since the test material needed no declicking at all, anything is too sensitive. I chose that music because it is very well done and it is full of transients similar, in certain respects, to clicks. Use of any "regular" declicking would degrade this piece, most more than my efforts as presented here.

No declicking software can distinguish all that well between clicks and many other transients. My testing indicates that all such programs, at any setting that is useful, removes far more musical information than clicks, by actual count. This is an objective observation, from looking at what is done to the audio files, not based on my opinion of how the results sound. The sound can be alright, even though the "damage" has been done, especially considering the alternative. Of course I have not tested all the software out there, but the range of applications I did use convinced me that nothing else is possible for a program.

By casual tests, with several pairs of good headphones, I find that a 12kHz sine wave is quite audible (and unpleasant). Maintaining settings for which 12kHz is tolerable (i.e. not so loud as to be really painful) I can easily hear 13kHz, but it is much less loud. I  can tell when 14kHz is off or on, but it is hard to say that I actually hear much. The point of which is that what I hear in or of the music is probably not very useful much above 12kHz, so harmful effects in the higher frequencies are likely to escape me.

I fully support the position that anything, good or bad, in the recording, should be reproduced clearly and accurately by the playback equipment, at least as "clearly" as it would have been heard by an observer during the recording. However, I've always considered things like the rustling of page turning, accidental bumps of this or that, squeak of fingers changing position on strings, and other such instrument noises to be detriments. They should be reproduced well if they are there, in the recording, but they are not part of the music and are mainly a measure of how far the performance and recording falls short of what they should be. Removing them, when possible, is a good thing.

In the recording I used here, there is very little of this kind of noise that I can detect. The little that I find removed by the declicking could therefore not be a significant factor is the evaluation that the processed file is too dull -- unless there is a lot of such stuff above the frequencies that I can actually hear. If so, perhaps these sounds are important in creating the venue ambience. I can hear the space quite well in some recordings, but I don't hear anything in particular in this recording, either the uploaded segment or the full track.

It is also possible that my ability to discriminate is very poor even at frequencies my ears can process well enough. I suppose one possible test would be to apply a low pass filter to each version (perhaps at 12kHz) and find out how much of the differences you notice are still audible. That would be quite interesting to me.

Looking at the frequency distribution of the difference between the two files I uploaded, I see virtually a flat line from the lowest frequency to about 3500Hz. Since the level of the original audio rises significantly from 3500Hz down to about 100Hz, I would say the declicking does, proportionally, effect higher frequencies more in this range.

Going higher from 3500Hz, the level of the difference trace is on a steady downward curve to about 15kHz, then it rises rather steeply. This entire range fairly closely matches the frequency distribution curve of the original audio, so I'm not sure if increasing attenuation as frequency rises is a correct interpretation. I don't quite see how to get a grip on the data. The overall difference between the two files, independent of frequency, averages about -66dB RMS, has a maximum of about -51dB RMS, and a peak difference of about -24dB, all of which should be very difficult to hear.

Although the processing done here was not through any kind of EQ-like filter, the question is: could the missing quality be usefully restored with a graphic or parametric equalizer? More importantly for my proposes, could the settings be determined by the data rather than through a subjective evaluation, the later of which would seem to be useless to me. Is there another treatment that might be more useful?

As stated in my original post, this processing is not something I apply to most recordings. Where it is needed, the results are sometimes dramatic. I just used it on an old sound effects LP of a heartbeat. The count was 4.2million+ corrections on one channel of a little over 20 minutes. This was the fifth round of automated declicking on the file, the earlier steps being for larger clicks. All legitimate audio in the heartbeat recording seems to be below 800Hz; I believe these processing results are all for the good.

declicked to death?

Reply #4

Low pass filtering at 12kHz makes absolutely no difference to my results - the A vs B difference is still there, in the same amount.

For me, the audible difference between "A with 12kHz LPF" vs "B with 12kHz LPF" is far greater than the audible difference between "B" vs "B with 12kHz LPF".

I think it would be hard to correct for the high frequency loss, because it sounds like it changes dynamically, probably based on the level of the input signal. There are moments where the high frequency loss seems insignificant, and other moments where it sounds quite severe.

Hope this helps.

Now, if you tell us what you're doing...  there might be a way of knowing what's lost at each moment and restoring it. Doesn't sound like there's anything to worry about with the heart beat record, but for more general use it would be necessary IMO.


declicked to death?

Reply #5
Thanks again for the interesting, and hopefully useful, information. No feedback is a hazard of working in isolation. I would appreciate one more indulgence, which is to point out a few of the particularly severe problem points and a few where damage is much less. While I apparently can’t hear what you hear, perhaps it is a matter of perception rather than not physically receiving. I have learned to hear some things for which I started out having little or no awareness.

Still I don’t think this is a large problem, as I normally only apply the process where it is needed, usually to small sections, often enough less than a second in length.  Differences, at least the differences I’m looking for, are readily apparent to me. But since it can do damage I’m not hearing, better perception could only be helpful.

As to what I am doing, you are most likely aware of it in the general form. The  Younglove Decrackling Script is well enough known . There have been many discussions about it on the Audiomasters forums
(mostly in the Syntrillium forum archive on the same site). While it is mostly considered wholly benevolent, I discovered that results vary markedly depending partly on the noise profile and the noise extraction parameters, but mostly on the declicking method and declicking parameters.

These declicking parameters take very small bites, but a great many of them. One of the interesting things about the method is that doing the same declicking directly on the audio, rather than on the extracted noise, produces markedly lesser changes. In regard to the current problem, I think the process effects the audio to the same degree, regardless of signal level. This makes for a greater net change as the original signal level lowers.

It has two significant uses for me: reducing the shusssh of scuffing, and ameliorating the harmonic distortion of in-groove damage. It  can’t return things to normal, but the result sounds much better than with the problem ignored. I haven’t found another way to achieve anything so useful on these kinds of damage.

declicked to death?

Reply #6
30+ downloads. Is anyone else able to notice the differences discussed between the files -- and willing to point out the time of two or three points with the greatest, and perhaps the least, noticeable difference?

declicked to death?

Reply #7
Since this has been bothering me, I started listening to the selections again, concentrating first on shorter sections at the lowest amplitude. Some differences in the sounds of certain instruments soon became apparent and easy enough to recognize that I can differentiate the before and after 100%. On median level parts (relative to the entire 29 second section) I find differentiating by sound very difficult and inconsistent, even though I can see the differences on screen readily enough. I have so far been unable to hear any difference on the highest level segments.

I still don’t know if what I find is responsible for the treated sample being call “dull” vs “clear and crisp.” While the differences I identified are easy to hear since I became aware of them, I wouldn’t say “high frequency loss sounds quite severe” here, or anywhere else, in the treated version nor do the differences seem significant vis a vis the music itself.

Since they are not part of any vinyl defect that would need attention (if this had come from vinyl), these differences are not something I care about when working on an album. They are also not something I pay any attention to when listening to music. Whether or not they play some important part, below conscious level awareness, I can’t say right now. I don’t know whether I will ever become more aware of such small details without actively looking for them. When in music listening mode I don’t tend to be very analytic.

I am interested in any other comments as to whether or not differences between the two versions seem important. This is a different consideration from “can you tell any difference.”

declicked to death?

Reply #8
Andy, the differing audibility on louder / medium / quieter sections you report matches my own perception.

I do think it's important and objectionable. I think I'd have noticed it without being asked to listen out for it. (The same is true of the side-effects of most declickers and many denoisers).

Whether it's more objectionable than the artefact the process helps to remove, or more objectionable that the side-effect of other methods of removing that artefact, is probably a more relevant question. If you had a CD quality master to start with, you wouldn't be doing this processing.



declicked to death?

Reply #9
Thanks. I am hoping to get more comments, to see if there are a variety of reactions, to find out if your experience is mostly the norm. I say it matters from the viewpoint of the ideal transfer but not from the standpoint of being able to fully enjoy the music. It isn’t anything I would notice if I were not looking for it. Whether that is because I have been so focused on solving vinyl defects for the past eight years that I’ve trained myself to ignore some finer points or that my hearing isn’t the same as yours, I have no idea.

If more of the people who bothered to even listen fall on one side or the other, that might shed a little light on what is important (or even noticeable) to most people, but so far people are not too forthcoming.

And no, as I wrote earlier, I don’t even do this particular thing on many of the LP’s I transfer. However, as I also wrote, I don’t get much feedback on anything I do. Possibly other processing I use more widely also ruins the recording for most people.

declicked to death?

Reply #10
Possibly other processing I use more widely also ruins the recording for most people.
As long as you don't have clients to satisfy that's fine
IMO you have to make up your mind about the purpose of the declicking:
-only remove clicks and other artefacts related to the storage medium, leaving the original recording as much untouched as possible
-or adjust the recording to your liking by making it more pleasant to your ears.
Both are legitimate, although the first is usually preferred for public releases.
Your example shows that your declicking method is affecting the music. The difficult part of declicking is that in most cases we don't have the click-free original as a reference. I don't know your declick tools, but it might be an idea to find less coarse settings that only remove what you want.
As a test you could take a clean recording (like your Stravinsky sample) and mix vinyl clicks to it. Then you could try to declick the mix. The advantage is that now you do have a reference.
Have you ever tried doing several declick passes with different settings, removing large clicks first and increasing sensitivity gradually in the following passes ?

declicked to death?

Reply #11
IMO you have to make up your mind about the purpose of the declicking:
-only remove clicks and other artefacts related to the storage medium, leaving the original recording as much untouched as possible
-or adjust the recording to your liking by making it more pleasant to your ears.
Both are legitimate, although the first is usually preferred for public releases.
If only that were true!

The "first do no harm" principle doesn't get very close to most commercial releases.

At best, engineers use the best tools available, and skilfully balance improvements vs artefacts such that the artefacts introduced are more than compensated for by the improvements. At very best, most people would be hard pressed to notice any artefacts on 99% of the recording.

At worst, engineers use very cheap tools, introduce horrible artefacts by over-using them, and then try to hide the faults they've created by applying reverb, eq, exciters etc.

Many of the transfers which are praised for "not overdoing it" are from engineers who only have cheap tools, but have the good sense not to overuse them. In truth, such releases often contain clicks which the better tools could have removed perfectly.

The market is split between people who want to buy CDs very very cheaply, people who quite like the original noisy sound, people who want as "clean" a sound as possible (however much this wrecks the music itself), and people who just want the specific music and will forgive any faults (clicks or processing artefacts) because "old recordings just sound like that".

As the CEDAR patents expire, and/or as tools become cheaper and better, there's probably a market* for release flat transfers with no restoration at all - for picky people to restore at home themselves. Then if they don't like the result, they only have themselves to blame!

* - well maybe me and three other people


SimplePortal 1.0.0 RC1 © 2008-2021