Skip to main content
Topic: cleaning vinyl audio? (Read 34299 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

cleaning vinyl audio?

Reply #25
Reissues of LP on Cd are often quite different.

I will copy here a post I made some years ago discussing a technique I worked out with CoolEdit/Audition. It is quite labor intensive but can be faster than straightforward manual declicking. I don't use it often, it usually isn't necessary to get very good results, but it is a method that lets you know exactly what you are doing to the file.

EXTREME DECLICKING
I held off writing this up for quite a long time because I suspect it will have very limited appeal. The recent post about the destructed drum attacks goaded me into action. Most people will decide that the automatic declicking they now use is good enough. Perhaps they won't even want to know how much good material goes ‘out with the bath water,' but if your results are not satisfactory, there is a better way. I myself only do this for extreme cases.

Although all declicking aalso removes good material, there are two ways around most of that problem. Almost flawless results are possible if the source LP is in reasonably good condition. It is still as Graeme said, "that you can't get 'something for nothing' ", but for more time and effort you can get better results.

The straightforward method is to do only manual declicking. Manual declicking often does a better job on any individual click than is possible with automatic declicking. This difference is frequently fairly subtle but it is real and often enough the difference isn't subtle at all.  More importantly, however, one can stay away from almost all legitimate musical transients, resulting in no dulling or loss of impact.

The second method is a bit more complex but can demand less intensive effort than strict manual declicking. It is adaptable from the extreme of just preventing the worst of automatic over processing to the extreme of striving for click removal perfection. It is a variation of a technique introduced by Alofoz, but adapted to allow virtually absolute control. If you are blessed with a recording in good enough condition, simple manual declicking will take less time, but for a bad to terrible disk, this second approach is easier and offers better results.

To use this approach, one needs to check ‘limit playback to: mono' in the device settings.

Proceedings are much easier if one involves only declicking in this adventure. This means that between the baseline beginning file and the declicked working file one does no noise reduction, no equalization, no normalization, no whatever. Any and all other processing can be done either before or after the declicking, just not between the baseline file and the declicked working file.

If using the Sonic Foundry declicking plugin, you must uncheck ‘remove subsonic rumble,' or probably better, first processing the file with ‘remove subsonic rumble' checked but with the three sliders all the way to the left. Use the result as the baseline file.

The declicked working file is what comes out of automatic declicking. This can be one step of declicking or as many steps as one wishes, especially including the Younglove decrackling process. The ideal is a perfectly declicked file, but as the starting assumption for this journey is that perfection cannot be achieved via automatic declicking, we strive for a file that is so adequately declicked it needs little or no additional declicking, while not being too badly destroyed from over processing. From here, we can eliminate the destruction and tidy up the proper declicking. Over declicked is better than under declicked, but anything more than necessary just increases the time and effort of correction.

You next need a file of inverted clicks. If your declicking of choice is a one step process (e.g. no Younglove decrackling run after major declicking), and if you do it with some tool, like ClickFix or Sonic Foundry, that has a "keep clicks only" option, you can use that option to end up with a file containing just the removed clicks, no music. Sonic Foundry automatically inverts the retained clicks. ClickFix's output must be inverted before proceeding. If using CoolEdit/Audition, or some other declicker that only gives you declicked music, or doing multi pass declicking, do a Mix Paste/Invert of the baseline file into the fully declicked audio file to obtain a file of inverted clicks.

This inverted click file contains everything removed from the baseline, the good, the bad, and the terrible trumpet distortions. Eventually it will contain only the good and we will mix it into the baseline file to achieve whatever degree of declicking nirvana we were willing to strive towards. First there is some deconstruction.

From the two files, baseline and inverted clicks, we need to created two other files, but to accomplish that we first need to create four intermediate files. There are some manipulations that require less disk space, but I will describe only the most straightforward. If disk space is an issue, everything between the baseline file and the inverted click file can now be deleted. Likewise, the next four intermediate files can be deleted immediately after use, but sometimes it can be very useful to keep most  intermediated results around for reference until everything is finished.

Open the baseline file in single waveform view (i.e. not multi-track). Select all of one channel. Save that selection as a new file. Do the same for the other channel. You now have two intermediate files, mono right channel and mono left channel. Be certain to use names that will not leave you confused as to which is what after a few more steps.

Do the same thing with the inverted clicks. Now you have the four intermediate (mono) files I mentioned, the delete able ones. From these four you will build two 2 channel working files, one containing the right channel and its inverted clicks, the other containing the left channel and its inverted clicks. These also must get names that leave nothing to chance. In these, I like to have the audio as the right channel, so it appears on the bottom of the edit screen, and the inverted clicks as the left channel so they appear on the top of the screen, for both the left and right channel files. Your arraignment is your choice.

The easiest way to build these two files is via multi-track. Insert the mono inverted clicks into track 1 and the mono baseline audio into track 2. Pan track 1 100% left (-100) and track 2 100% right (+100). Select both. Mix down into a two channel file. You can also accomplish this in waveform view with copying and pasting or the Mix Paste function, but multi-track seems easier to me.

Now when you open one of the files (in edit mode) you will have the unprocessed audio on one channel and the inverted clicks on the other. When you play the file, due to the ‘limit playback to mono' setup, you will here the declicked results. If you select only one channel to play, you will hear either the un-declicked baseline or the inverted clicks; either will be center stage. This would probably be very unfunctional when trying to create a masterpiece from a raw multi-track recording, but it works quite well to clean up an LP.

I developed this so that I could work in Spectral View. Alofoz's technique, which uses multi-track,  is undoubtedly well suited to getting good sounding results, but I need, and you need, Spectral View (and destructive editing) if, perchance, we are attempting declicking distinction. In Spectral View, with a little practice, it is generally easy to see the good material that any auto declicking removes. Since you can see it, you don't have to also spend the time listening to it, you can just fix it straight away. In Waveform view, on the other hand, many clicks have a clear enough appearance, but the good material that is inadvertently removed through automatic declicking is frequently much less distinct.

From here there are many degrees of involvement. It is possible to review every millisecond, every removed smidgen, to assure that every sample is as good as the original recording into your computer will permit. It is also possible, with enough resolution and fortitude, to ignore everything except the worse problems areas (e.g. the drum strikes) and trust to the auto declicking for all else. To get the best results, one must work at a fairly high level of horizontal (time base) zoom.

If you are new to Spectral View, it may take a little practice before everything becomes obvious, but the difference between musical transients and clicks is almost always unmistakable, especially since you have available the immediate comparison of the baseline recording and the removed material. Those drum strikes, and other transients that automatic processing incorrectly declicked, can be salvaged simply by selecting the offensive material on the inverted click channel and pressing the delete key. Now your drum strike, or trump blast, is preserved in pristine glory.

There are a few things to note here. For optimum manual declicking, the click must be selected with some care. For this un-declicking, selection is faster and easier because you can select anything and everything that isn't a valid (inverted) click at one time, in one swoop. Select any number of events on the inverted click channel and any amount of blank space between them. Often, a double mouse click to select everything currently on screen is appropriate; delete it all with one keystroke.

The only time it gets difficult or especially tedious is when good material (on the inverted click channel) is practically on top of a genuine click. In this case you may need to zoom in especially close to allow precision selecting. Sometimes it will be easier and faster to delete everything in that vicinity from the inverted click channel and do a manual declick of the offensive impulse noise on the baseline audio channel.

As I said, much of this can be done visually; you can learn to identify good material on sight. However, as an aid, and sometimes as a necessity, when you play both channels together, you hear the mixed results, so you can identify the dulled impact and the over processed distortion audibly. You can also learn the more subtle degradation than any automatic declicking produces. If you select only the audio channel, you get the unprocessed recording and thus hear what you will be missing if the auto clicking result is allowed to stand.

For some of the other possibilities of this technique, you do need to listen -- to both channels at once. Larger clicks too often don't get properly removed with any automatic declicking that I know about. After auto declicking, good material immediately adjacent to the click is ‘disturbed' and something from the click is left over, often a low frequency thump that is as unpleasant as the original click but more difficult to precisely locate (but here you can see exactly what and where because the original click is still right there, on the audio channel).

Often it is easier to start from scratch to manually declick these than it is to repair the auto declicking damage. Simply swipe across the click channel to select everything related to the bad fix, delete it, and take your best shot at manual declicking. These large clicks often require multi-step manual declicking for best results.

Removal of somewhat smaller clicks usually don't cause new damage, but the clicks themselves are partially left behind. This is usually only obvious when you listen to the results, not by looking at the Spectral image. Maybe this residue isn't a problem, depending upon the context, but is still isn't perfection. These partial corrections, when you care about them, like the gross ones of the preceding paragraph, are best fixed by simply deleting from the inverted click channel and manually declicking on the audio channel. If you simply manually declick, but don't delete the inverted click, your final results will still have the click -- inverted, but just as unpleasant.

Because you are selecting only the inverted clicks channel, removing via the delete key is the proper thing to do. This has a danger. If you inadvertently select both channels, disaster will result. You are working on the file of either the right or left channel of your recording at any one time. Eventually you need to get the right and left channels back together. If you delete with both inverted click channel and music channel selected, the file becomes shorter by the amount deleted. The right and left music channels, temporarily in separate files, will no longer be in sync. You will be very unhappy.

In the lower right hand corner of the screen is a box with the total file time. The last three digits, the number of milliseconds, can be used as a quick hash count check. If that number ever changes, you did the wrong thing. Pray there are enough undoes available to recover.

I don't know about other version of the software (I use CE2K) or other hardware, but after I've removed a jillion or so clicks manually, or deleted enough incorrect declicks in this scenario, my screen gets hinky. This is a sign of coming end times. I delay that eventuality by occasionally purging undoes and clicking on save -- check the total file time before saving.  If the impending crash stigmata still eventually appears, I note where I am current working (by the time display), save,  close the program. Everything is fine when I re-open, but the pk file will generally be rebuilt in the process of opening.

I have never deleted portions of a recording (right and left music together) to correct a problem, but I understand from various comments that this is not a particularly uncommon technique. To use it here you will have to select the music section to be deleted and install Que marks. Then when the right and left channels are rejoined, use the que marks to delete from both channels at once.

Once each file, right and left -- inverted click channel plus music channel together -- are satisfactory, you start consolidating. Once again there are several possible approaches: mix paste; copy/paste; channel mixer; pull apart, insert into multi-track, and mix down; convert sample type. Convert sample type is easy. Select Mono in the channels box and make sure resolution and sample rate are not being changed. Once converted into mono, Save As a new mono file with a good name. That mono file is the mix of inverted clicks and audio -- the perfectly, or at least adequately to your standards, declicked audio.

Now you have two more files, the perfectly declicked right channel and the perfectly declicked left channel. Put them back together. The multi-track solution seems easiest to me, used exactly as it was to created the inverted click-music files. Take care to insert left into track one and right into track two. Pan track 1 hard left and track 2 hard right, select both tracks, mix down, save the result.

cleaning vinyl audio?

Reply #26
wow that was incredibly deep...
i am still trying get used to editing in spectral view instead of waveform view, but i think that the realtime mixing of the inverted clicks and the original audio by forcing the output to mono is a good idea, like by listening for overprocessing in the declicked audio, instead of trying to pick out the original audio amongst the clicks like what i tried to do... but i would also like to add, that if you are using various sensitivities on the declicker, you should keep hold of that inverted clicks file in order to cancel out the detected clicks that you already know are material you want, as you continue to increase the sensitivity of the declicker

right now i am beginning to settle into a cycle in Wave Repair , press N (for next detected click) and P (to play the ~1 second or so around the marker) and keep going until i see that the click detector has landed the marker on something valid... then i press ctrl-alt-L or R to spectrally replace the click, and shift-up or down to 'nudge'(the term used in the help files) the resulting waveform until it is level with its surroundings...
this is going to take a couple of passes with different settings , instead of one single pass of mousing around the waveform like i did in Goldwave, but at least keeping one's hands on the keyboard like that is certainly good for your posture in the long run... but i wish that the declicker in Wave Repair would show more than just the number of potential clicks, like if it could somehow mark where the clicks are so i could quickly see what it has done, instead of only pressing N and V to jump between different clicks

Acoustica's 'interpolate' function is one of the best things i have ever used, its dragging your cursor acros and clicking a button instead of drawing out a new waveform, and the results are better too

i soooo wish that instead of seeing dials , sliders and numbers, i can see some graphic representation of the denoiser or decrackler on the waveform itself.... suppose that first i manually locate a click on the waveform, and then i bring up the denoiser or decrackler dialog box, things like amplitude and threshold would be represented by horizontal bands running across my view of the waveform, and similarly stuff like width would be represented by a vertical band... maybe with more abstract concepts like sensitivity or different types of noises, there would be an 'example' waveform overlaid over the true waveform to help us 'target' the correct parameters

cleaning vinyl audio?

Reply #27
right now i am beginning to settle into a cycle in Wave Repair , press N (for next detected click) and P (to play the ~1 second or so around the marker) and keep going until i see that the click detector has landed the marker on something valid... then i press ctrl-alt-L or R to spectrally replace the click, and shift-up or down to 'nudge'(the term used in the help files) the resulting waveform until it is level with its surroundings...
this is going to take a couple of passes with different settings , instead of one single pass of mousing around the waveform like i did in Goldwave, but at least keeping one's hands on the keyboard like that is certainly good for your posture in the long run... but i wish that the declicker in Wave Repair would show more than just the number of potential clicks, like if it could somehow mark where the clicks are so i could quickly see what it has done, instead of only pressing N and V to jump between different clicks

Let me give you some insider knowledge. Wave Repair's automatic click detection was added a long time ago, and since then better auto declickers have appeared on the market. In the meantime I've concentrated on making Wave Repair's manual functions as useful as I can, and left automatic processing to the growing number of other programs around. My basic attitude is that some glitches just can't be fixed with an automatic tool, and although you can fix them manually with a general purpose editor, the kind of editing operations required for typical vinyl glitches can be quite involved and so I've aimed to "package" them in Wave Repair so they are easier to perform. A couple of examples:

1. Selecting a nearby undamaged section of waveform to copy over a glitch. As well as the "copy preceding block", there's also the "block overlay" feature that allows you to slide an image of the damaged section over other parts of the file to try and find the best match.

2. The Younglove decrackling process is a multi-stage operation using standard audio editors, but is simplified in Wave Repair.

Anyhow, I digress. By using Wave Repair's auto click detection, you are starting off with a sub-optimal set of detected clicks. If you genuinely want to use an automatic click detector, there are better options. One I particularly like is called Wave Corrector (no relation). It operates in a two-pass manner. After it has detected the clicks, you can review each one, marking phantoms as such, and adjusting the repair width of others. What you can't do is manually decide on different repair strategies for individual clicks. You might find that using Wave Corrector to find the clicks and then using some other editor (eg. Wave Repair, Sound Forge, etc) to actually fix the clicks works well.

For what it's worth, here is my general approach (using Wave Repair):

1. My basic philosophy is that the only things that need to be fixed are audible glitches. Therefore the starting point is to actually listen to the file. I use headphones because they are much better at revealing glitches than speakers.

2. I listen to and repair the file in sections, typically 1 minute at a time. While listening, hit the space bar whenever a glitch is heard. This places a marker that you can return to later.

3. After marking the glitches in the section, zoom in on each glitch to see what's going on. Sometimes the waveform damage is obvious, other times switching to spectral view helps.

4. Depending on the nature of the waveform damage, I use my experience to select what I think is the most appropriate repair method. (The more you deal with vinyl records, the better you'll get at relating the visual waveform to what sort of repair will work).

5. After fixing up the glitches in a section, I listen again and once satisfied, save the edits and move on to the next section.

6. I've set up a couple of macros to ease the task of moving on to the next section and returning to viewing the entire current section.

7. I usually only use automatic declickers if the LP is in bad shape and there are just too many clicks to contemplate fixing manually. When an automatic declicker is required, I usually use Wave Corrector or CoolEdit's Audio Cleanup plugin. I've got a stack of other auto declickers that have been collected over the years, but these are the two I prefer. Two tools I've evaluated and found to work well (but have not bought) are Click Repair and Sound Forge Noise Reduction 2.0.

Hope this helps. Remember: you can't have too many tools. There is no one single audio restoration package that does it all.

cleaning vinyl audio?

Reply #28
Quote
2. I listen to and repair the file in sections, typically 1 minute at a time. While listening, hit the space bar whenever a glitch is heard. This places a marker that you can return to later.
hey nobody told me about that before! but then again i was getting used to the space = play setup from Acoustica... and im still too used to zooming in with the mouse scroll wheel , and the left and right click selection method from Goldwave, i could drag the selection marker off the view and it would autoscroll... rolling the wheel or keeping a mouse button pressed down is alot easier than mashing on the arrow keys for me...

i totally agree with manually listening for the clicks and only resorting to an auto declicker , or more likely the auto decrackler, when things get really rough....
so far i've been playing around with Acoustica Pro and i am pretty impressed by its processing capabilities, like the interpolate tool and the broadband noise reduction, but i feel really handicapped by its user interface because the view is tied to the playback, whereas in Goldwave i could freely and impulsively zoom in on where i heard a click... only being able to zoom in on the playback marker (so everything zooms past you during playback), and not being able to apply a highpass filter so i could see the clicks from far away, really is a shame, because that means i cant bring Acoustica's really nice processing capabilities to bear on the problem spots....
it kind of makes up for it with the 'listen to removed signal' feature in the declicker and broadband noise filter, on the former all you hear are just clicks and more clicks wether or not its what you meant to remove, but on the latter it is a godsend, all you do is keep turning the bands up until you start to hear music, its not trial and error like in the other automated tools, but its actually interactive because of the feedback

cleaning vinyl audio?

Reply #29
theres a saying that any technology, sufficently advanced enough, will appear to be magic... or something like that... my point is, Adobe Audition is like magic... the denoiser is pretty standard, the 'adaptive noise reduction' is still pretty strange to me, my Atom-powered netbook isnt fast enough to preview it realtime so i dont have a good idea of what the settings do(i still prefer the 'broadband noise reduction from Acoustica)... the declicker is so complex that it even figures out its own configuration for you, all you have to do is put down how 'sensitive' you want it and how little it 'discriminates' , i think its how aggressive the effect is... and it generates values for the other 9 fields in the window for you....
there are still a few pops that it misses but they are bite sized chunks compared to what it was before, i can easily deal with them with mouse editing in goldwave... what it does get rid of is so artifact-free that its almost like they werent even there in the first place

cleaning vinyl audio?

Reply #30
Automatic click repair: try VinylStudio, it does a more-than-decent job:

http://www.alpinesoft.co.uk

Also, something called Wave Corrector (see Google)

Regards - the author (of VinylStudio)

PS: Low pass filtering will *not* do it, not at all.  VinylStudio (like all other declickers you will come across) uses a statistical method to find the clicks and then an auto-regression based algorithm to reconstruct the missinmg samples.  It also has a 'manual repair' option for any clicks that the scanner misses.
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #31
Some months ago I started to digitize my vinyl and tape records.
I have been using various tools for audio recording and processing, among which:
Audacity, CD Wave Editor and Total Recorder (TR).
The latter is excellent, in my opinion, for data acquisition from various fonts, and some basic processing like normalization, the second for quick waveform examination and audio splitting, the first for the variety of tools offered.

TR also offers a simple to use audio restoration package which most of the time gives good results, but not always.
It’s so that I decided to try other packages for continuous and impulsive noise removal.
It’s so that I wasted many, many hours the days after!

Expensive professional tools were too much for my scope, so I looked for the low budget packages, starting from the most promising ones, according to what they claimed and to your postings on this forum. Also this page was of great help: http://www.delback.co.uk/lp-cdr.htm (thanks Clive).

I ended up trying almost all packages for which an unrestricted evaluation version was available... collecting a series of mainly disappointing experiences.

I may say that IMHO there is no best application in this field. Each one has it’s pro and cons, and none is really satisfying in all aspects. So you can do a better job with the joint use of several of them, not just one, but in any case the results you can get are not excellent, unless you can deserve several hours to clean up each of your records, which is something I cannot afford.

Most of these programs allow for automatic processing for clicks and pops removal , but even doing so you may end up wasting hours in trying to achieve the best compromise tweaking the parameters. But, unless the task is simple, e.g. quite classical music contaminated by sporadic large clicks, the blanket is always too small. Cleaning up almost completely your records will also rip off some part of the music, otherwise you should accept some disturbing noise here and there.

Moreover only a few applications implemented some easy way to keep the process under control.
There are basically three ways for doing this:

1) listen in real time to the processed output;
2) listen in real time to the extracted noise (the residual);
3) examine and listen to the residual after having processed the file.

The first two methods may involve adjusting the parameters while listening to the result, perhaps playing in a loop short segments while parameters are adjusted to evaluate the effect.
The first method is the most common, a few programs let you also listen to the extracted noise.

Surprisingly only Total Recorder lets you save the residual in a file! Employing the 3rd method with other programs is thus quite cumbersome! You have to subtract the processed file to the original one, which in Audacity translates in: inverting the processed file, copy and paste it in a new (stereo) track, mixing it with the original and save the result! Quite frustrating. However you can then analyze it much quicker than listening to the entire record over and over again until you are satisfied of the result.
You can in fact skip the portions where minor or no noise at all was detected and examine more carefully the critical parts, to hear if any hint of the original music is in there, which means that a certain distortion was performed, so that you may try with another set of parameters.
If instead some click and pops are left in the original you may try another pass of cleaning.

Besides failures in click detection and false detection, applications may do a different job in reconstructing the damaged parts. Techniques based on spectral analysis of nearby segments should give better results than those based on simple interpolation of the waveform in the time domain. 

Resuming in brief my experience with each package I tried.

With the Total Recorder add-on you have to adjust 3 parameters. The configuration space is thus enormous and to find one which cleans up the record without taking out too much is often impossible. It also misses some important clicks even in the most aggressive settings.
ClickRepair has a good reputation, however I didn’t find it that superior in unattended declicking. Semiautomatic or manual editing is not that flexible and friendly as in Wave Repair.
This latter offers the most comprehensive set of tools for manual editing, but the automatic declicker is quite difficult to set up and doesn’t give very good results.
Wave Repair is very good instead for continuos noise removal. Probably the best of the lot in this. Listening to the residual I verified how removed noise adapts to the music level, and I could hear no hint of the original music. Something that happened with other packages instead.
Groove Mechanic has only one parameter to adjust (plus a few buttons) and does a relatively good job, but you have very little control and feedback on program operations. However I was surprised by the fact that if you keep reprocessing the file it continues to find new clicks forever!
Finally I tried GoldWave. It also has only one parameter. It was the only one which repaired a record badly scratched, but, even if at moderate settings, analyzing the residual, I discovered that a lot of the music was also cut off.

As a sort of conclusion I may say that while noise reduction is a mature technique with a sound theoretical base, clicks/pops detection and removal is still a (time consuming) art with unpredictable results.

Being an engineer and a software developer myself I try to advance some suggestions to those of you involved in developing these tools.
I know it’s easy to say and could be very hard to do, but this algorithms should be self adaptive! It’s stone age to think of letting the user adjust the parameters. It remembers me of the beginning of the radio era... when tuning a station was a enterprise :-)
Algorithms should perhaps examine the noise residual in the time and frequency domain, and dynamically optimize the parameters in order to maximize the probability of having just noise in it.

cleaning vinyl audio?

Reply #32
Not convinced yet to give it up, I tried VinylStudio too.
The program is well designed and easy to use. Very good the idea of keeping the list of corrections that one may examine and correct if necessary. By manually reviewing and adjusting them, one should achieve a high quality restoration.
The task is simpler and the result probably better than what can be done with other applications.
However I counted a relatively high number of false positives and false negatives, as well as some incomplete corrections. This happened with various settings.

Thus I conclude that this program too it’s not a panacea, and should not be used, like the other ones, for automatic unattended processing of your files.
The trial version doesn’t allow you to save the corrections as a wav file, thus I could only apply method 1) of my previous post.
I wanted to try also Wave Corrector, but the trial version is too limited. Anyway, the interface and principle of operation looks like quite similar in many aspects to VinylStudio. Someone can tell if it works really better to justify a trial and the more expensive license?

Finally, let me say that I expected some comments on my previous posting, which seems to have gone overlooked instead, from you specialist of the field.

Isn’t it any worth what I said?

I suspect that most applications if not all of them, only implements noise recognition in the time domain (am I wrong?) whilst in spectrograms clicks are usually manifest. So I wonder why recognition in the frequency domain or perhaps a joint recognition in the time and frequency domains is not attempted. Wouldn’t it drastically improve a correct discrimination of music and noise?

Let me say it frankly, I see here a vast space of improvement, at least for the low budget packages targeted for home usage. I don’t know how professional tools work, however there is a huge gap in the cost of the differently targeted applications. If I had to pay some hundreds dollars or euros, to digitize a few dozens of LP’s, I would rather buy equivalent CD’s, but I would certainly pay 50-70$ for an application capable of a reliable, quality restoration, with minimal manual intervention and no waste of time.

cleaning vinyl audio?

Reply #33
There are dozen of applications. Most of us who do very much of this, perhaps all of us, use a variety of tools, depending upon the material. Speaking for myself only, while hearing about other programs can spark a little temporary interest, I am disinclined to spend any more money or time on something new. I worked out my techniques over time, they have done what I need for hundreds of albums now, and I don’t feel any great need for something better.

Those other programs I have aroused myself to try seemed a little different, but not better overall that what I’ve been using for some time. Quite possibly there are applications that would achieve the same end results by another route, but they would have to consistently save significant time and effort to be worth a change in my habits.

Declicking takes time. A few, very few, LPs that started out very scratchy were fully corrected satisfactorily by my series of “automated” processing steps, but I still had to listen through each one carefully before I knew that. Mostly the manual declicking part is straight forward, but it can’t be hurried.

cleaning vinyl audio?

Reply #34
@GlaMas:
There are, of course, audio restoration tools around that operate in the frequency domain.
One that comes into my mind is iZotope's RX. But the price tag is "professional" ($349).
Sadly, I can't give you any recommendations - I just watched a demo video of it's operation.
But after using/trying some of the tools mentioned above I can only second that there's no automatic process that's really satisfying...

.sundance.

cleaning vinyl audio?

Reply #35
I agree with AndyH-ha on this. I've been restoring vinyl, cassette and 78's for almost 10 years now. The tools I use now are mainly the ones I started out with. Partly that's because I'm familiar with them and so can get reasonable results quicker than with software I'm not familiar with. But it's also because I've not found anything better. I do check out the new offereings from time to time but, as I said have not found significantly anything better than my old tools. The only exception is Vinylstudio. I find it gives reasonable results and it runs much quicker than my old tools - it will go through an entire LP in under 1 minute. However, it conatins lots of features that I don't need or use and also doesn't have some features that I routinely need (and that you can only really expect to find in a full-blown audio editor) so it's not superceded them.

In summary I think most automated tools that I've tried over the years do a fairly good job in that the end result is usually a lot better than the original audio - assuming you take the time to learn how to use them. However, none of them give perfect results. To achieve that you need to put in a lot of time with manual corrections

cleaning vinyl audio?

Reply #36
I put the majority of my effort into cleaning the physical vinyl itself, which typically leaves not much to clean up digitally: light crackle on very quiet sections, and the occasional obvious click/pop.

Typically I get rid of the clicks and pops with the auto single click filler in CoolEdit and for light crackle on quiet sections just run over it with the medium amplitude audio preset in CoolEdit auto decrackling. That way I get an album done in a few minutes, and fix only noise that is not sufficiently masked.

That said I don't mind a little vinyl noise in my audio so I will rarely redo any album: I put it down to period ambience and authenticy.

cleaning vinyl audio?

Reply #37
Allright, I almost figured out my way too. As a conclusion (for now) I think that, of all the tools that I have tried, I will keep Wave Repair (WR) and VinilStudio (VS). They are relatively cheap and good value for the money.

As I said Wave Repair gives the best results in noise reduction (NR). You have an adequate control of the process, in that you can set accuracy of the FFT, you can select precisely a noiseprint, also by examining its frequency characteristics, and you can enable amplitude dependent filtering, something not found elsewhere. With NR done this way (at 1.5 sensitivity) I didn’t experience any undesirable effect.
This is also more than adequate for hum and rumble reduction. In fact, although I read recommendations (from the Authors of WR and VS as well) to use specific hum and rumble filters for this purpose, I found that the response of the notch and high pass filters available is not step enough (at least in WR), so these filters take out more of the bass notes than general NR does.

WR also has an impressive set of tools for manual repair to experiment with, if this is the case.
It also offers the spectral view of your data, which is not available in other similar applications, something essential to precisely locate otherwise hidden noise.
Spectral replacement is also available, although it doesn’t always blends smoothly with the context.

However, if you have a severely damaged recording with thousand of clicks, something that can deliver a reasonable automatic correction, with not too much efforts, is the only choice. The output of VS in these cases is in fact much more enjoyable that the original, although it can have missed some clicks, or made more corrections than necessary here and there. Anyway, audible artifacts seems to be rare. I succeeded in producing the residual (extracted noise) of a Chopin Nocturnes in real bad shape, where the program made more than 50000 corrections (!). At a random examination of it no evident subtraction of musical content resulted. (But with rock or jazz it can be another story... In other cases, as I said, I experienced less positive results.)

I wish however that VS had an option to directly save the residual! As well as a more handy way to operate with wav files, without specifying collections, etc..

Is the Author still reading these notes?

I experimented with the filters of VS too. Unfortunately there is a (minor) bug, I suppose, as the processed file is a few ms shorter in this case. So I couldn’t examine the result in more detail.

cleaning vinyl audio?

Reply #38
Hi GlaMas,

Just got your PM.  I didn't receive any email notification of your original posts, hence no reply.  Sorry about that.  I do indeed find your comments valuable, thank you for taking the time to post them.

I can see it would be useful to be able to listen to the residual.  I will put it on the todo list, it would not be hard to do.  I have not looked into what can be achieved, manually or automatically, with a spectrogram but one problem I forsee is that VinylStudio needs to know *exactly* where a click is in order to repair it, which is why we scan for them in the time domain.  But it's not an area I know much about so I will look into this in more detail when I have time.

With regard to false positives and false negatives when scanning for clicks, *all* programs suffer from this to a greater or lesser extent.  With VinylStudio, you can see them in the Corrections List which perhaps makes them more obvious.  To compare like-for-like, you might like to import the results of a scan in WaveRepair into VinylStudio's list of Recorded / Imported files in the Cleanup Audio window for the album in question. You can then flip backwards and forwards between the two files and if you are luccky the waveforms will line up exactly.  If so, you can then compare (visually) the output from VinylStudio with the output from WaveRepair.  I think you might be surprised at what you find.

I'd also be interested in your results for Jazz and Rock albums.  The problem is always percussion, which is why classical music cleans up so well.  I have experimented with various algorithms for differentiating a percussion burst from a click, and it always comes down to a compromise between false positives and false negatives.  My goal has been to to make VinylStudio's declicker's settings easy enough to use that non-technical users can get decent results.  Did you experiment with the PP (persussion protection) setting?  Turning this up eliminates a lot of false positives, and one can subsequently deal with any false negatives by rescanning the affected section(s) with more aggressive settings.  I am ashamed to admit that we default to a fairly low PP setting because it's what our target market seems to expect.

With regard to hiss, hum and rumble filtering, the hiss filter (which is a broadband noise filter) is not a good choice for removing hum as you will lose bass notes if you fail to turn the hum filter on.  This is because the hum filter is a narrowband 50 (or 60) Hz notch filter whereas the hiss filter works over a broader frequency range.  By enabling the hum filter, you remove the hum from the 'noise print' used by the hiss filter, which in turn reduces the impact on bass notes in general.  Again, we have focussed on ease of use in terms of the filter settings we offer.

Why does the fact that VinylStudio knocks a few ms off the end of the file make it impossible for you to evaluate the efficacy of the noise filters?
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #39
Quote
I have not looked into what can be achieved, manually or automatically, with a spectrogram but one problem I forsee is that VinylStudio needs to know *exactly* where a click is in order to repair it, which is why we scan for them in the time domain. But it's not an area I know much about so I will look into this in more detail when I have time.


Using Wave Repair clicks usually appear in spectrograms as neat narrow strips of different color (thus I bet that also an automatic recognition should be feasible). The location is not *exact*, but should be a good guess. It’s surprising how very short spikes, that go otherwise overlooked in the waveform, stand out in the spectrogram.

Quote
I'd also be interested in your results for Jazz and Rock albums. The problem is always percussion,
which is why classical music cleans up so well.


I got quite good results with a few classical records I tried. The LP’s were in bad conditions and the list of corrections huge (as I said). I used the default setting for classical music. I randomly checked them and found that almost all were correct.
On the other hand today I tried with an LP by Paolo Conte (http://en.wikipedia.org/wiki/Paolo_Conte). This album was in good conditions and only a dozen of audible clicks were present. I tried various sensitivity settings and also with PP to the max. Nevertheless about 2000 corrections were made and I verified that, this time, almost all of them were unnecessary!
Thus I decided to use WR for manual correction. I resorted to the spectral view most of the time to spot the noises.

Quote
I have experimented with various algorithms for differentiating a percussion burst from a click, and it always comes down to a compromise between false positives and false negatives.


In fact. That’s why recognition in the frequency domain could be more effective.

Quote
Why does the fact that VinylStudio knocks a few ms off the end of the file make it impossible for you to evaluate the efficacy of the noise filters?


Because the original and the processed file were no more aligned, so I couldn’t extract the residual to see what VS did.

Quote
With regard to hiss, hum and rumble filtering, the hiss filter (which is a broadband noise filter) is not a good choice for removing hum as you will lose bass notes if you fail to turn the hum filter on. This is because the hum filter is a narrowband 50 (or 60) Hz notch filter whereas the hiss filter works over a broader frequency range. By enabling the hum filter, you remove the hum from the 'noise print' used by the hiss filter, which in turn reduces the impact on bass notes in general. Again, we have focussed on ease of use in terms of the filter settings we offer.


Should be certainly so, but as I said, I couldn’t verify this.
I checked the results of WR instead. I listened to the residual and I realized that, in this case, on the contrary, the hum and rumble filters (HR) took out more bass notes than filtering with noiseprint (at maximum accuracy).
In the residual of the HR filters I could perceive part of the original music, with noiseprint just the noise! Maybe that WR and VS have a different implementation of filtering algorithms. In my case hum and rumble are at –50/60dB level, other broadband noise well below.
It would be interesting to know Clive’s point of view on this issue.

cleaning vinyl audio?

Reply #40
Using Wave Repair clicks usually appear in spectrograms as neat narrow strips of different color (thus I bet that also an automatic recognition should be feasible). The location is not *exact*, but should be a good guess. It’s surprising how very short spikes, that go otherwise overlooked in the waveform, stand out in the spectrogram.


OK, it sounds like a very useful tool, for humans at least.  I will look into it when I have time.  But 'close' is not close enough to perform an automatic repair unfortunately.  It needs to be exact.  It might be useful to help localise areas for scanning at higher sensitivity than the rest of the file though.

I got quite good results with a few classical records I tried. The LP’s were in bad conditions and the list of corrections huge (as I said). I used the default setting for classical music. I randomly checked them and found that almost all were correct.
On the other hand today I tried with an LP by Paolo Conte (http://en.wikipedia.org/wiki/Paolo_Conte). This album was in good conditions and only a dozen of audible clicks were present. I tried various sensitivity settings and also with PP to the max. Nevertheless about 2000 corrections were made and I verified that, this time, almost all of them were unnecessary!
Thus I decided to use WR for manual correction. I resorted to the spectral view most of the time to spot the noises.


Don't let the numbers put you off.  Most of these 'false positives' do not actually matter.  If you zoom in on a few of these corrections and take a look at the change that has actually been made to the waveform you will see that, in the majority of cases, it is just a tiny adjustment and is not audible.  But, occasionally, such a'repair' can dull the sound of a percussion burst. VS's 'Percussion Protection' algorithm catches most of these but not all.  It's pretty effective though, especially at higher PP settings.  If you compare the results with Wave Corrector for example, I think you will see what  mean.

I can see though that if you have the patience to perform repairs manually, a spectrogram is a useful tool to help you find the clicks.  I think that's what it comes down to, for you at least, and that currently gives WR an edge if you want to work that way.  As I say, I will look into providing one when I have time (busy with something else, currently).

Quote
Quote
I have experimented with various algorithms for differentiating a percussion burst from a click, and it always comes down to a compromise between false positives and false negatives.


In fact. That’s why recognition in the frequency domain could be more effective.


I can't comment on that without looking into it properly, but have you compared the results of an automatic scan in WR with a similar scan in VS (using the technique described in my previous post)?  I'd be curious to know how many 'significant' false positives WR generates.  Of course, it's hard to track them down without a corrections list (why did we ever put that feature in? ), but you might use VinylStudio's false positives (and percussion protection markers) as a guide.  This is what I do when comparing VS with competing applications.

Quote
Quote
Why does the fact that VinylStudio knocks a few ms off the end of the file make it impossible for you to evaluate the efficacy of the noise filters?


Because the original and the processed file were no more aligned, so I couldn’t extract the residual to see what VS did.


I just checked this, and the waveforms seem to be aligned OK.  PM me your email address for a full license key so that you can save the cleaned up files and perform your own tests.

Quote
I checked the results of WR [hum and rumble filters] instead. I listened to the residual and I realized that, in this case, on the contrary, the hum and rumble filters (HR) took out more bass notes than filtering with noiseprint (at maximum accuracy).
In the residual of the HR filters I could perceive part of the original music, with noiseprint just the noise! Maybe that WR and VS have a different implementation of filtering algorithms. In my case hum and rumble are at –50/60dB level, other broadband noise well below.
It would be interesting to know Clive’s point of view on this issue.


That's probably because of phase shifts introduced by the rumble filter.  Try turning it off.  I'm not sure that listening to the residual is a viable test in this case though.  What I noticed during testing that a quietly plucked bass string almost disappeared when the hiss filter was used to filter out hum, but was restored when the hum filter was turned on.  Another curiosity which might pique your interest is that rumble can cause the hiss filter to generate faint clicks (I think because the low frequency component introduces small DC offets which the overlapped FFT's cannot entirely eliminate).  Turning on the rumble filter eliminates this and we do it by default when you enable the hiss filter.

I think I'd sum all this up by saying that the most important thing (to me) is what the cleaned up music actually sounds like (I find a the loss of the initial attack of a finger plucking a bass string quite noticeable, for example, and overly agressive declicking can do that).  I think it would be easy to get side-tracked and we do get quite a bit of positive feedback on the declicker from our (largely non-technical) user base.  Anyway, thanks again for posting and PM me for that license key.
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #41
Quote
I have not looked into what can be achieved, manually or automatically, with a spectrogram but one problem I forsee is that VinylStudio needs to know *exactly* where a click is in order to repair it, which is why we scan for them in the time domain. But it's not an area I know much about so I will look into this in more detail when I have time.


Using Wave Repair clicks usually appear in spectrograms as neat narrow strips of different color (thus I bet that also an automatic recognition should be feasible). The location is not *exact*, but should be a good guess. It’s surprising how very short spikes, that go otherwise overlooked in the waveform, stand out in the spectrogram.

Just to reinforce what Paul was trying to explain:

Yes, you can use a spectral view to easily identify the *approximate* location of a click. But the very nature of the spectral view is that every "point" is in fact the representation of an FFT of the surrounding samples. (In the case of WR, it's 128 samples wide). So having noticed the spike in the spectrogram, you know that the actual click is somewhere in the 128 sample range. (In fact, chances are that it's smack in the middle). If you zoom right in, you'll see that the "spike" is in fact an extended "smear". At those sort of zoom levels it's fairly easy to pick the middle of the spike and switch to waveform view - you should then see the glitch on the waveform. (It's surprising how subtle some of these glitches can be, and yet still cause audible ticks. I've seen what look like tiny "pimples" on the side of a much larger waveform that are still audible).

cleaning vinyl audio?

Reply #42
It's surprising how subtle some of these glitches can be, and yet still cause audible ticks. I've seen what look like tiny "pimples" on the side of a much larger waveform that are still audible.


It sure is!  The human ear is an amazing device, and scanning for clicks is more of an art than a science.  But we really must get that spectrogram view implemented.

And hello Clive, nice to meet you.
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #43
(A little later that day...)

I had a little play with spectral view in Cool Edit and I can see what you are getting at GlaMas.  However, I found that only when you zoom in to a view that is 10-20 seconds wide can you see any clicks.  I personally would not have the patience for that.  I also found - just a quick test, you understand - that VinylStudio, even at its lowest sensitivity setting, finds clicks that are hard / impossible to see by eye (although whether you can hear them is another thing).  This makes me wonder whether we should make the lowest setting a bit lower still, but I find that if you go too far, broader clicks / pops - some of which are very noticeable - start to slip through.  I don't think these would show up in spectral view either as they lack the high-frequency component that form the basis of all click-repair (or should I say click-detection) software, visual or automated.  They're often easy to see in the waveform display though, if they're big enough.

Having said which, I can see that a spectral view is a useful tool.  We'll definitely put it on the to-do list, along with listening to (and perhaps saving) the residual of the declicking process.

Input like this is incredibly useful by the way, please keep it coming.  I don't have a background in DSP and am painfully aware of my limitations in this area.  But for now, back to the housework 
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #44
Alright, looks like we all agree on spectrograms. I am sure that Paul will find a way to integrate the tool in his program.
Also what you guys said goes pretty much in the direction of finding a way to automate a joint recognition in the time and frequency domain, i.e. what I suggested in one of my previous posts. Take it for what it counts, but my flair of old sw analyst, says it’s worth investigating in this sense.

Quote
I'm not sure that listening to the residual is a viable test in this case though.


Oh, well, I think that the residual is always revealing! If your music is in the residual it can only mean that the filter has subtracted too much! Isn’t it?

Quote
What I noticed during testing that a quietly plucked bass string almost disappeared when the hiss filter was used to filter out hum, but was restored when the hum filter was turned on.


Hmm... I didn’t notice anything like that. Can This depend on the accuracy used in the “hiss” filter?
WR offers various degree of accuracy, as well as amplitude dependent filtering, which should minimise undesirable side-effects like this. Isn’t it so?

Quote
rumble can cause the hiss filter to generate faint clicks (I think because the low frequency component introduces small DC offets which the overlapped FFT's cannot entirely eliminate). Turning on the rumble filter eliminates this and we do it by default when you enable the hiss filter.


Same question of above. In light of all what we said, would you post on this Clive?


cleaning vinyl audio?

Reply #45
If your music is in the residual it can only mean that the filter has subtracted too much! Isn’t it?


I'm not sure, it's not something I commonly do, but for filters which might change the phase or otherwise shape the signal, I would think it could be misleading.  It's a fair test of the declicker though.

Quote
Quote
What I noticed during testing was that a quietly plucked bass string almost disappeared when the hiss filter was used to filter out hum, but was restored when the hum filter was turned on first.


Hmm... I didn’t notice anything like that. Can This depend on the accuracy used in the “hiss” filter?
WR offers various degree of accuracy, as well as amplitude dependent filtering, which should minimise undesirable side-effects like this. Isn’t it so?


Possibly, but the point is really moot as you have a better option.  I don't know what WR's accuracy parameter means, but it probably means a longer FFT length.  What I do know is that hiss filtering is inherently subtractive and you don't want anything in your noiseprint (which is what the subtraction is based on) that you can get rid of in a more selective way.  Our hum filter (and no doubt WR's) is a steep-sided notch filter which takes advantage of the fact that we know exactly what the mains frequency is and can therefore target it precisely.  VinylStudio makes this a bit easier by measuring the noiseprint on the fly, after declicking, hum filtering and the rumble filter have all been applied (assuming you turned them on).  In most conventional audio editors, you would need to (and should) scan first with these filters enabled before taking a noiseprint and applying the hiss filter.

We decided on a fixed FFT window length (8192) based on listening tests and with one eye on performance and ease of use.  Progressively shorter windows sounded noticeably worse, including a noticeable pumping sound when the level of background noise is high which renders the hiss filter useless in my opinion.  Longer ones sounded much the same but carry a performance penalty we were keen to avoid as we wanted to provide real-time hiss filtering on a 500MHz pentium CPU.  One of the nice things (well, I think it's nice) about VinylStudio is that filtering is done on the fly, so you don't need to scan the entire file and then overwrite (or make a copy of) your original file (unless you want to).

Having said all of which, 8192 might not be the best compromise on a fast PC.  I will revisit this at some point, perhaps provide a bit more flexibility (i.e. 8192 or more), which is simple enough to do.  It may also be the case that overly long FFT sizes are less effective, I'm not sure. I lack the theoretical background and haven't performed enough listening tests.  Presumably there has to be a point in the averaging process where white noise becomes indistinguishable from the overall spectral content of the music and you just end up with a fancy graphic equaliser.  Someone out there can no doubt enlighten me.

Our hiss filter is inherently amplitude (or, rather, SNR) dependent.  Again, on the basis of listening tests, this stood out as the right thing to do, although what we do may do it in the same way that WR does.  Loud passages of music are certainly much less affected by the subtraction process than quiet passages.  Soft cymbals are a particular challenge, for which we have a 'quality' slider which lets you determine, in effect, how quickly the hiss filter reacts to changes in SNR.  The bass note I mentioned was quite soft, and given that there were bass components in the noiseprint, the hiss filter thought it was fair game.  It's that kind of subtlety that we seek to preserve, especially for non-technical users.

One thing that really upsets hiss filters is clicks, so take these out first (VinylStudio always does things in that order) and make sure there are none in the noiseprint.  They seem very effective on noisy 78's though (ours does, anyway), provided you don't overdo it.  I would *not* apply the hiss filter to a recording that doesn't need it because the FFT .. inverse FFT process (subtraction or no) is inherently imperfect with a finite length FFT.

Quote
In light of all what we said, would you post on this Clive?


I believe he is already monitoring this thread.  I can't give you a timescale for that spectral view by the way.  There are other things (such as support for FLAC as a recording format) which we want to do first, but I will post developments back to HA in due course.  You might like to watch this thread:

http://www.hydrogenaudio.org/forums/index....showtopic=66260

Or you can sign up to the RSS feed on the AlpineSoft website.  Also, if you are still using it, VinylStudio will offer to update itself when the time comes (and gives you a summary of what's new).

Ciao!  Got your PM by the way.  Kind of you to offer to pay but a deal's a deal, I'll send you the key.
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #46
Ok Paul, thanks a lot.

You almost convinced me not to use the hiss filter and just apply the rumble and hum filters.
I will make some more comparisons.

However I think that VS would benefit of some flexibility here. I.e. what kind of filters it applies exactly?
One size fits all is not always true. For instance, in my case I don’t have a 50Hz hum at all, I have instead a noise peak at 100Hz (at nearly –50db) and a smaller one at 200 Hz. This means that the hum I have is not introduced by the more common earthing problems, but by poor filtering in the power supply of some device in the signal path. Does the VS hum filter cope with this case too? It would also be nice if one could specify the characteristic of the notch filter. 20 dB attenuation would be enough in my case, and nearby frequencies would be less affected (by phase shifts as well) than by a –60 dB filter or so. The same applies to the rumble filter as well.

As for the FFT size, a larger one, although slower, could perhaps be more adequate for batch processing of files, e.g. in my case I don’t care of real time processing.
BTW, remember to place in your to-do list a direct access to wav files (like all other programs do).

cleaning vinyl audio?

Reply #47
YW

The hum filter filters out the common harmonics too.  The rumble filter is a 3-stage IIR high pass filter rolling off at 32Hz.  I think other programs that dedicated filters like these do much the same thing.

We have a difference in philosophy, you (and perhaps the rest of HA) and I.  Our goal is to put competent tools in the hands of (careful now!) non-technical users.  There's a lot of settings we could offer, but we don't and probably won't because it's not what those people want or need (perhaps I am in the wrong forum here; oh well...)

Instead, we put our time into figuring out what the majority of the parameters should be and code them in, providing adjustments for the ones that we find make the biggest difference to the results.  Put another way, VinylStudio is not, and does not seek to be, an alternative to CoolEdit (which I think is a truly miraculous program).  I think some control over the FFT size in the hiss filter might be desirable though, trade off time against quality.  I will look into it.

As things stand, we cannot reasonably provide support for loading a WAV file directly as your collection is where VinylStudio keeps track of your filter settings, trackbreaks and suchlike.  I acknowedge that this is not what everybody wants and I do plan to review it at some point, with the possible aim of moving the information currently stored in the MCF file into smaller, one-per-album files.  It would be safer, for one thing.  It's not a high priority though and we'd still encourage users to create a collection to keep track of it all.  Again, for someone of a certain age with a pile of albums or tapes to digitize, it's appropriate.

Perhaps you'd like to do a few informal listening tests, comparing results from VinylStudio with whatever other cleanup tools you currently favour, tell us what you make of it.  I'd certainly be interested to know.

Oh yes.  Because it applies filters on the fly, VinylStudio cannot show you the effects in the waveform display.  To get around this, save a cleaned-up copy in the Cleanup Audio window.  The waveform display of the copy will show you the filtered output, and you can switch back and forth between the 'before' and 'after' views.  You can delete the copy when you are done with it.

Cheers - Paul.
I am an independent software developer (VinylStudio) based in UK

cleaning vinyl audio?

Reply #48
Wow, lots of new posts in this thread over the past couple of days (which I've spent mainly in bed with "man flu", ie. a cold). I feel that perhaps I should make some contribution, so I'll gather them together in one posting...

And hello Clive, nice to meet you.

Hi Paul, nice to meet you too. I see you're located in "London-ish". So am I (Watford, actually). So if you feel the urge to meet up sometime it could probably be arranged. I once met Derek Higgins (author of Wave Corrector) and we had a good old chin-wag.

I had a little play with spectral view in Cool Edit and I can see what you are getting at GlaMas.  However, I found that only when you zoom in to a view that is 10-20 seconds wide can you see any clicks.  I personally would not have the patience for that.

I agree about the need to zoom in to see the clicks. It's the same with Wave Repair. My personal view is that spectral view is useful when you're doing a manual repair by listening. When you hear a click, zoom in around the general area, switch to spectral view, and the click stands out quite clearly.

Other glitches that lack high frequencies (plops and thuds) are much harder to spot. Sometimes you'll see an isolated "splodge" in the spectrogram. Quite often it's easier to spot plops and thuds in waveform view: there can sometimes be an obvious DC-like shift in the general waveform. But even once you've found such a glitch, fixing it can be maddenly difficult. I admit to sometimes giving up and leaving the original thud present, because all attempted repairs make things no better or worse.

Quote
I'm not sure that listening to the residual is a viable test in this case though.


Oh, well, I think that the residual is always revealing! If your music is in the residual it can only mean that the filter has subtracted too much! Isn’t it?

My view is that the ultimate goal is to improve the musical signal. Music in the residual is "collateral damage" and it could be that the most effective repair does result in some collateral damage. If you obsess about avoiding any music in the residual, then you might prevent yourself from achieving the best result.

Quote
rumble can cause the hiss filter to generate faint clicks (I think because the low frequency component introduces small DC offets which the overlapped FFT's cannot entirely eliminate). Turning on the rumble filter eliminates this and we do it by default when you enable the hiss filter.


Same question of above. In light of all what we said, would you post on this Clive?

Regarding clicks introduced by the hiss filter. This of course is pure speculation on my part, but assuming VinylStudio uses FFTs to implement its hiss filter, my experience has been that unless you apply some sort of windowing so that the overlapped FFTs are "faded in" to each other, discontinuities at the boundaries occur, leading to "ticking". A filter might remove those discontinuites. But if this is what's happening in VinylStudio, then the rumble filter is simply masking something that probably ought to be dealt with by windowing the overlapped FFTs. But I reiterate that I have no idea how VinylStudio's code actually works, so if you already window your FFTs, please don't take offense!

I don't know what WR's accuracy parameter means, but it probably means a longer FFT length.

That's right. Wave Repair's accuracy settings mean an FFT length of: Low=2048, Medium=4096, High=8192, V.High=16384.

Our hum filter (and no doubt WR's) is a steep-sided notch filter which takes advantage of the fact that we know exactly what the mains frequency is and can therefore target it precisely.

Wave Repair doesn't have a "hum filter" per se. It has a generalised filter that can be configured as low-pass, high-pass, or notch, with adjustable Q and turnover frequencies, and you can run up to three of them simultaeously. There happen to be some preconfigured "presets" to deal with hum at 50, 50+100, 60 and 60+120, but the user is at liberty to adjust the frequencies and filter steepness however they like.

We have a difference in philosophy, you (and perhaps the rest of HA) and I.  Our goal is to put competent tools in the hands of (careful now!) non-technical users.  There's a lot of settings we could offer, but we don't and probably won't because it's not what those people want or need (perhaps I am in the wrong forum here; oh well...)

Seems like we're aiming at different user types. Wave Repair is primarily aimed at manual repair - fiddling with the waveform at ultra-fine levels of detail. My view is that automatic declickers sometimes do a good job on "medium level" clicks, but I've yet to find one that deals with big pops and splats, and most of them miss the tiny ticks. I'm kind of obsessive about my restorations, so I have to get in there and fix up things that auto declickers get wrong.

I've never seen VinylStudio, but from what I've read here it sounds as if it's targeted at a more automated restoration process. So the two tools could be complementary. (Seems I should download an evaluation copy and try it out). On the other hand, people who are only interested in automatic cleanup can just ignore Wave Repair - it's definitely a tool for "geeks".

VinylStudio is not, and does not seek to be, an alternative to CoolEdit (which I think is a truly miraculous program).

And neither is Wave Repair. CE2000 is indeed a wonderful program - IMHO the finest affordable general purpose audio editor there ever was. What a shame that when Adobe bought out Syntrillium, they chose to dump CE2000. You & I aren't trying to replace it, but somebody needs to. For sure, the install files are readily available (from my own website, for example), but activation keys are not. New users are either forced to use illegal keys or miss out on a wonderful program.

All the best,
Clive

cleaning vinyl audio?

Reply #49
Quote
Our goal is to put competent tools in the hands of (careful now!) non-technical users. There's a lot of settings we could offer, but we don't and probably won't because it's not what those people want or need


Alright, these notes are for those who want to experiment, or spend some time to get, perhaps, better results. However you may also think of providing VS with an interface for “normal” users and another one for perfectionists 

Quote
> If your music is in the residual it can only mean that the filter has subtracted too much! Isn’t it?

... for filters which might change the phase or otherwise shape the signal, I would think it could be misleading


You are right. High pass (HP) or notch filters introduces substantial phase shifts, this means that alignment of samples is lost, thus it’s not possible to extract a residual the usual way, and it doesn’t make sense.

Given this, how one could compare the results attainable with hum and rumble (HR) filters vs. noise removal (NR) with a fingerprint? (The otherwise said “hiss removal”).

I thought of applying different filtering processes to files of pure white noise (generated with Audacity). As NR is level dependent I tried also with different intensity levels: 0 dB, -20dB and –24 dB. I made these experiments with WR as it offers configurable filters. The results are interesting I think, and are as follows.

1) Rumble is better attenuated by appropriate notch filters, rather than one or two cascaded HP filters!
In my case I have rumble distributed around two peaks at 15 Hz and 25 Hz, while hum has a peak at 100 Hz and a negligible one at 200Hz.
For hum a notch filter at 100 Hz is mandatory, for rumble I experimented with one and two cascaded HP filter with corner ranging from 30 to 45 Hz.
Alternatively, I used two “broad” notch filters, tuned at 15 and 25 Hz
I found that this way the resulting response is steeper, thus frequencies between 40 and 100 Hz are less affected.

2) Applying HR filters may introduce some clipping here and there.

3) Filtering with NR is always level dependent, but if the “level dependent” option in WR is selected this characteristic is enhanced, so that NR filtering has a really negligible effect at a 0dB level,  it is nevertheless modest otherwise (without checking the option).

4) At –20dB level things are more interesting. The results, without checking the “level dependent” option, are very similar to those produced by the HR filters, but the response is steeper! Morever all information in the noiseprint are used, so that some other frequencies are lightly attenuated.

5) At the –24dB level attenuation of frequencies were noise peaks are located is greater.

6) Would it be better to apply HR filtering and then NR with a noiseprint of the residual noise?
I tried also this. The result is a slightly stronger attenuation of rumble, but at the expense of a less steep response, resulting in more degradation of (ultra)bass notes.

The only thing I could not simulate is the dynamic adaptation of the NR filter. I suppose this could be only judged by actual aural test.

However you should agree with me that I didn’t find any evidence that applying HR filters rather than direct NR is better!

Based on this tests I think I will stick with applying NR without applying HR filters first.
Isn’t it?




 
SimplePortal 1.0.0 RC1 © 2008-2020