Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Digitizing mono vinyl: using phase-inversion to reduce noise (Read 26557 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Digitizing mono vinyl: using phase-inversion to reduce noise

Hi.

I have been using a long-winded process to reduce some extraneous noise when digitizing mono vinyl sources. With Sound Forge, I have been subtracting the difference between the two channels of my records, on the assumption that most of it is noise, and the resultant files are usually audibly better.

1. Invert the phase of one of the two channels on the original .wav file (1.wav).
2. Convert to mono by mixing the channels.
3. Save the difference (noise file) as a mono .wav file (2.wav).
4. Re-open the original "stereo" file (1.wav).
5. Convert to mono by mixing the two channels.
6. Save the sum as a mono .wav file (3.wav).
7. Open original "stereo" file (1.wav).
8. Open difference (noise) file (2.wav), copy it and overwrite onto left channel of 1.wav.
9. *edit* Invert noise channel.
10. Open sum file (3.wav), copy it and overwrite onto right channel of 1.wav.
11. Convert to mono by mixing the two channels.
12. Save changes to 1.wav.

Is there an audio editor that will allow me to subtract the difference of a stereo file in a single step, as it would save me a lot of time, and allow me to hear instantly whether the audio is significantly improved, without having to get confused by extra files? If I could work in flac, that would be even better.

If found the link below that references and old version of Cool Edit, but I gather that's been discontinued.

http://www.hydrogenaudio.org/forums/index....mp;#entry344060

Thanks.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #1
When you mix the left and right channels to produce mono, that does essentially "subtract the difference" from the original file. After all, it's the "difference" that makes it stereo!

The procedure you describe will, I think, actually result in a mono file consisting of only one of the original channels. Your last step is to mix the sum and the difference which causes one channel to be cancelled out:

( (L + R) + (L - R) ) / 2 = L

The best thing would probably be to figure out which is the quietest of L, R, or (L + R) / 2 (mono mix) and use that. I don't think your long-winded scheme is achieving anything more than that.

edit: fix spelling

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #2
There are two reasonable approaches. Which one to use depends on the disk.

Most often I find the two channels near enough to each other that I convert to mono. Do this after declicking. Automated declicking works better if done first.

Done properly, this cancels all the out of phase noise. You are done. Since there are several ways to do this in CoolEdit, it is probably also easy to do in your editor.

I prefer the Channel Mixer's LR to Mid-Side preset. This puts the mono result in the left channel and the subtracted difference in the right channel, allowing me insight into what is happening. Only the left channel is used. The Channel Mixer's Average preset produces the proper result in dual mono. Convert Sample Type to Mono (50/50) gives the correct result as a mono file.

The second case is that one channel is in much better condition than the other. This usually results from frequent play on a poorly set-up arm/cartridge. In this case use the better channel. mixing them will usually give a worse result.

 

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #3
Your last step is to mix the sum and the difference which causes one channel to be cancelled out:

( (L + R) + (L - R) ) / 2 = L


Err... you're right. Maths was never my strong point. I guess the "improvement" I was hearing was either my imagination, or the result of the remaining channel sounding better than the discarded one.

Nonetheless, is the principle of what I'm trying to do sound? That is, separating difference between the two mono tracks, and somehow using that to remove noise from the eventual true-mono audio. If so, is there software (preferably free/share-ware) that will do this for me in one or two simple steps?

Thanks.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #4
Done properly, this cancels all the out of phase noise. You are done. Since there are several ways to do this in CoolEdit, it is probably also easy to do in your editor.


What do you mean by done properly ?

The noise in one channel has no relationship to that in the other channel and therefore the two cannot cancel out. Noise is random by definition and  it is nonsense to talk about  "out of phase" noise.

Summing the two channels certainly makes the noise less noticeable because it is masked by the actual signal.

The best route is to try just left channel, just right channel and left/right summed and see which gives the best result.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #5
The noise in the two channels can be highly correlated. The left and right channels of the signal are encoded in opposite sides of the groove and result in needle motion at 90 degrees to each other, but anything in the groove that causes the needle to move horizontally, such as dust or a scratch, creates equal and opposite noise in the two channels. If the motion is not exactly horizontal then there is only partial cancelation, and if the motion is vertical then the noise adds instead of canceling.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #6
Certainly just summing the two channels will cancel the "out of phase" noise. (In the case of vinyl, this is up and down movement of the stylus.)

However it is possible that there may be slight phase differences* between to two channels, which could result in some odd cancellations of high frequencies. - So it's worth closely checking that the two channels are "lined up" before summing.

* Possible causes of slight phase differences between channels:

Cartridge alignment (although anything other than a linear tracking tone arm will always cause misalignment at some point across the record). You may be able to compensate for this a little by time shifting one of the channels on individual tracks before summing.

Spherical styli at extremes of modulation. You can't compensate for this but elliptical or better styli will give better results.

Misalignment of the the playback or recording heads on the master tape (assuming analogue master). This can be compensated by time shifting one of the channels before summing.

All this this is from (my rather poor) memory, so I might be talking crap!

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #7
I guess then that the "noise" that I have separated using phase inversion of the two "mono" channels cannot be utilised in any way - that is, removed from the sum of the two channels. Is that correct?

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #8
I guess then that the "noise" that I have separated using phase inversion of the two "mono" channels cannot be utilised in any way - that is, removed from the sum of the two channels. Is that correct?

Well, no. When you do a simple conversion to mono by mixing the two channels, you are utilizing the technique you describe. The "out-of-phase" noise in the signal is eliminated. I think the suggestion here is to compare the left  channel alone to the right channel alone to the mono mix and choose the one that sounds the best. And all three of these are easily generated by almost any basic editor (i.e. no complicated process like what you described above is needed).

OT: The same thing works with some noise in stereo recordings from FM. The noise is "out-of-phase" and is greatly reduced by mixing to mono.

edit: my --> by in last sentence

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #9
That cartridge alignment is (reasonably) correct has to be taken as a given. Of course not everyone has managed to achieve that, but lack of optimum out of phase cancellation won't be the only problem with recordings made on such a rig. If the LP has been played enough times on a mis-aligned setup, the groove isn't going to be is very good condition. One channel of the transfer is likely to be the better choice in this case.

When the TT & cartridge are working correctly, the noise is reduced by3dB, which is usually quite noticeable. This also works on cassette source, but is less certain because there are often so many azimuth misalignments. With mono LPs, summing a stereo recording to mono often makes dramatic improvements. Many click and pops will be removed or markedly reduced in level.

"Proper" is what has already been described and listed by several people. As stated, it is a standard function in any reasonable audio editor.

Mono LPs were cut with a larger tip than stereo. Groove damage changes these considerations, but the claim is that a "proper" mono stylus, which is larger and spherical, gives better results. I say claim, not evidence, because I've never had such a stylus to experiment with, but this is a pretty standard understanding. When the groove is significantly damaged, different sizes of styli may allow one to get to a less damaged depth (higher or lower than where it was played to destruction).

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #10
OT: The same thing works with some noise in stereo recordings from FM. The noise is "out-of-phase" and is greatly reduced by mixing to mono.


Another trick that works quite well with noisy FM stereo signals is to mix just the high frequencies down to mono. This retains separation at low and mid frequencies while greatly reducing the high frequency noise.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #11

OT: The same thing works with some noise in stereo recordings from FM. The noise is "out-of-phase" and is greatly reduced by mixing to mono.


Another trick that works quite well with noisy FM stereo signals is to mix just the high frequencies down to mono. This retains separation at low and mid frequencies while greatly reducing the high frequency noise.


What's the easiest way to do this?

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #12
1. Sum and difference the L and R channels to get L+R and L-R channels
2. Lowpass the L-R channel
3. Sum them back up to get L and R

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #13
1. Sum and difference the L and R channels to get L+R and L-R channels
2. Lowpass the L-R channel
3. Sum them back up to get L and R

Lowpass: 18kHz or 10Hz?
Why do you want to get back to L and R?

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #14
Back to removing noise from mono LPs: I've always wondered if something like the "intelligent" vocal cut algorithms could be used in reverse to do a better job than simple summation?

With vocal cut, if something is on the L channel and the R channel in roughly equal amounts, then you want to remove it, and keep the rest.

With mono LPs, if something is on the L channel and the R channel in roughly equal amounts, then you want to keep it, and remove the rest.


The intelligent vocal cut algorithms FFT the audio (or similar) and analyse it to remove the common information, while keeping both stereo channels reasonably intact. It works nearly OK - you can easily introduce the kind of warbly mess that arises by chopping things around in the FFT frequency domain, but you can avoid it with some care.


I wonder if you could apply this principle to mono records? The simple summing only reduces the unwanted stuff by 6dB relative to the wanted signal. The FFT method would work better than that. I suspect the problem is that most unwanted stuff is broadband (and often impulsive in nature), and so not very responsive to frequency analysis.


If you have foobar2k, you can try this...

http://www.hydrogenaudio.org/forums/index....showtopic=46611

Cheers,
David.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #15
The noise in the two channels can be highly correlated. The left and right channels of the signal are encoded in opposite sides of the groove and result in needle motion at 90 degrees to each other, but anything in the groove that causes the needle to move horizontally, such as dust or a scratch, creates equal and opposite noise in the two channels. If the motion is not exactly horizontal then there is only partial cancelation, and if the motion is vertical then the noise adds instead of canceling.



I agree, except that it is vertical movement that represents the difference signal (i.e "out of phase") otherwise summing the two channels of a mono signal would also cancel the actual signal as well as the noise.

You say that "anything in the groove that causes the needle to move horizontally, such as dust or a scratch, creates equal and opposite noise in the two channels": surely the mono signal also causes the needle to move hoizontally and would, by your reasoning, also be cancelled.

I still have a problem with the term "Out of phase" when applied to noise but I think "inversion" is what is really meant here.

The best way to play a mono recording must be with a mono cartridge that cannot respond to vertical movement.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #16

The noise in the two channels can be highly correlated. The left and right channels of the signal are encoded in opposite sides of the groove and result in needle motion at 90 degrees to each other, but anything in the groove that causes the needle to move horizontally, such as dust or a scratch, creates equal and opposite noise in the two channels. If the motion is not exactly horizontal then there is only partial cancelation, and if the motion is vertical then the noise adds instead of canceling.



I agree, except that it is vertical movement that represents the difference signal (i.e "out of phase") otherwise summing the two channels of a mono signal would also cancel the actual signal as well as the noise.

You say that "anything in the groove that causes the needle to move horizontally, such as dust or a scratch, creates equal and opposite noise in the two channels": surely the mono signal also causes the needle to move hoizontally and would, by your reasoning, also be cancelled.

I still have a problem with the term "Out of phase" when applied to noise but I think "inversion" is what is really meant here.

The best way to play a mono recording must be with a mono cartridge that cannot respond to vertical movement.

Yep, got that one backwards. Thanks for the correction.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #17
The best way to play a mono recording must be with a mono cartridge that cannot respond to vertical movement.

If you are only playing it to listen, that may be true. You can also sum the two channels of a stereo cartridge to cancel the difference; that is what most of the 'mono' cartridges are these days. However, if you are doing transfers, which you intend to clean up for the best possible music, recording with a normal stereo cartridge (perhaps with a mono stylus) is significantly better.

Declicking works better overall on the stereo cartridge output. I'm not sure why, but I've played with it enough times to know it is so. Therefore automated declicking and decrackling come first, then mixing to mono. Simply mixing to mono will remove and/or reduce some clicks, but the end result will be better declicking first. As I wrote before, the final results are often a dramatic improvement.

Except when one channel output is clearly superior to the other because of damage to the LP. Summing in that case gives a result much poorer than the better channel alone. Without the stereo cartridge transfer, you don't have a choice. I don't find many mono LP that haven't suffered over the last half century.

I don't know about David's suggestion for a different approach, whether it could produce a better overall result than 'normal' summing to mono or not. How can the program be expected to determine what you want to keep and what to throw away? I suspect that just doing a second step, with one of the better NR routines, would beat it for quality.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #18
I find...

  stereo > declick > decrackle > denoise > mono

...works better than...

  stereo > mono > declick > decrackle > denoise

...but wonder whether that's true for all declick/decrackle/denoise algorithms and records? There might be some which work much better on the cleaner mono signal. (I haven't found any, but then no one has given me a CEDAR system to play with at home yet!  )

Cheers,
David.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #19
It's a trade-off. There is some advantage to mixing to mono first because that process alone can remove many clicks, especially larger ones. For awhile I believed the advantages weighed in on that side, in spite of general opinion. The disadvantage is that too many of the partially removed clicks are harder to deal with afterwards. Automated declicking routines recognize fewer of them and some are harder to find, and harder to correct, manually.

There is a sort-of middle ground but I've pretty much abandoned it too. Using CE's Channel Mixer with the L-R to Mid-Side preset puts the mono file in the left channel and the throw-away stuff in the right channel. On thing this process makes clear is that very large clicks often have consequences that extend for a few hundred milliseconds beyond the obvious peak. That is at least one of the reasons why the usual declicking too often leaves low frequency thumps that are harder to remove. In a normal stereo mix, these are often difficult to see.

Anyway, dividing the spoils this way gives one clearer access to some of the larger (and not so large clicks). One can manually remove some clicks from the right channel, then mix back to the original using the Mid-Side to L-R preset. You are back to the beginning, ready for automated declicking, minus some number of difficult clicks.

Unfortunately  it doesn't always work well because of two difficulties. One is that all of the disruption isn't always in the right channel; sometimes the bulk of it is in the left. The other is that the same applies to smaller clicks that you don't even care to go after while in this mode. Some of those will, none the less, be effected by your efforts on near-by larger clicks. Once the channels are back together, the program(s) can have the same problems properly recognizing and correcting some clicks as when simply converting to mono and working on the mono file.

Of course, that doesn't answer the questions about more expensive tools.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #20
It's a trade-off. There is some advantage to mixing to mono first because that process alone can remove many clicks, especially larger ones.

Surely unless a click is identical but out-of-phase on both channels, then it can't be completely removed by mixdown to mono? And it strikes me that the chances of a click being identical and out-of-phase are vanishingly small.

I agree that conversion to mono can produce an improvement (ie. a reduction in amplitude) where the click is present mainly on only one channel, but there are more effective repair techniques in this case: If you're the sort of person who is prepared to manually consider the best repair method for each individual click, then by recording a mono LP in stereo, you can copy small sections of one channel to the other in order to remove glitches that appear on only one channel. Straightforward mixdown to mono doesn't produce a repair as good as this.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #21
The overall advantage is in declicking first. However, "completely removed" vs "reduced to the point of inaudibility when listening with  closed back headphones" might be debated from a philosophical viewpoint (i.e. what is acceptable or desirable? how do you actually tell?). Often, many clicks are gone as far as any practical consideration goes. In the case of large clicks that are only partially removed, they are often enough made easier to deal with. First converting to mono was always "first," never the end of the process. My earlier discussion deals with why I eventually abandoned that approach.

I appreciate the aesthetic viewpoint of trying for the ‘perfect' fix but in practical terms that is often irrelevant. One might sometimes, or often, say with reasonable confidence that the removal of a small click produces exactly what would have been recorded had the disk defect not existed, but in many cases it isn't possible to know unless one has another, undamaged, source against which to make a comparison (essentially never in my case). Besides, we are talking about recording from an LP. Do it twice, or ten times, and you get something slightly different each time.

I've done complete albums manually, with WaveRepair and with CoolEdit. I get the urge upon infrequent occasions when the LP condition is especially good (there will only be some thousands, or tens of thousands, of clicks to remove, instead of hundreds of thousands) and the music has quite a bit of ‘delicate' detail, but I'm am not sure that it actually makes a difference that is detectable to anyone who might listen to the result.

It is possible to noticeably, or badly, damage a recording with automated declicking, but it is usually possible to avoid that. In most cases I do several steps of automated declicking (and possibly decrackling), then I go through once or twice, or sometimes more, to find what the automated declicking couldn't handle. I always have the unprocessed copy available to copy over a bit that was damaged by automated declicking, but I don't need it for very many albums.

The results of automated declicking will never be exactly the same as manual declicking. It is easy to make comparisons and see the differences in an audio editor. Hearing the differences, however, is not easy, or not possible, given "good" automated declicking (except for the more serious problem here and there).

I've done A/B comparisons, and ABX comparisons a fair number of times when I was unsure or anxious (do a section manually, compare that result with the same section done automatically). Most of the time either I can't tell any difference or I can only tell a difference through a very close comparison. In the later case, a decision has to be made: put in ten extra hours (or 48 extra hours) to get something I will never again be able to detect (because I won't be doing A/B comparisons after I'm done and just want to listen to the music) or go with the "good enough" result.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #22
However, "completely removed" vs "reduced to the point of inaudibility when listening with  closed back headphones" might be debated from a philosophical viewpoint (i.e. what is acceptable or desirable? how do you actually tell?). Often, many clicks are gone as far as any practical consideration goes. In the case of large clicks that are only partially removed, they are often enough made easier to deal with.

Andy, I think you and I are largely in agreement over many aspects of LP restoration. It was certainly not my intention to start a fight over this.

Of course, if some easy process produces as good an audible result as a more involved process, then the easier process should be used. There is no argument over that. I'm just surprised that you've found that mixdown to mono can actually reduce clicks sufficiently. In the case of damage to one groove wall only (ie. the glitch is only on one channel), then a preliminary mixdown to mono will result in that glitch appearing in both channels, but 3dB down. There are very few glitches which audibly disappear when they are reduced by only 3dB. I haven't done a huge number of mono records - most that I have were 78s - but I was very surprised by how many glitches were indeed almost entirely on one channel only.

The results of automated declicking will never be exactly the same as manual declicking. It is easy to make comparisons and see the differences in an audio editor. Hearing the differences, however, is not easy, or not possible, given "good" automated declicking (except for the more serious problem here and there).

Agreed. And if automated declicking gives acceptable results, then it is of course easier to use it. My own experience is that automated declicking is usually worth doing, but a manual tidyup is nearly always necessary. Automatic declickers usually miss low level "ticks", and often make a dog's dinner of large clicks and pops. They are also pretty helpless when it comes to things like "scuffs" and "thuds".

But we're getting away from the subject of this thread, which was specifically about mono records. And in that respect I am completely with you on the desirability of recording and restoring mono records in stereo, prior to a final mixdown to mono.

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #23
Rather than saying "it won't work", do you think either of you can try the foobar2k keep centre channel plug-in? Or post an audio extract I can test it with?


My home audio set up is currently in pieces, and I don't have a "mono record, recorded in stereo, declicked and decrackled, but not yet converted back to mono" to test.

From simulation (mixing in run-in groove noise and clicks with various panning), it seems that this plug-in does offer some advantage to a plain old "sum both channels" at the end of the job, but I suspect with a real recording and all its vagaries this advantage could reduce dramatically.


If you can test a real recording in foobar2k, or upload a 30 second extract of a mono restoration, just before the final mixdown to mono, so I can test it myself, that would be great.

If I do test, I'll upload my results for all to hear, but I need some source material!

Cheers,
David.

(I'm ashamed I haven't got a single archived track to test this on, but I have never saved the result at this stage!)

Digitizing mono vinyl: using phase-inversion to reduce noise

Reply #24
Automated declicking is never the end of declicking. I've done a few LPs where I found only a handful of not very evident clicks after automated processing, but generally there are enough to still be a distraction, sometimes very many.

I'm certainly not adverse to helping out in an experiment when the topic actually interests me, but I don't save intermediate results from a project once it is completed. However, I do have a mono recording on the computer that I'll look at sometime in the next few days, as time permits. It has so much in-the groove damage, producing distortion, that I abandoned it, but I did not delete it because the music seems interesting --  and maybe I will eventually persuade myself it could be worthwhile to proceed.