78rpm Restoration Tips and Tricks

[Anubis]

Hello Everyone,

I've just recently discovered this program which seems to be a great tool for properly restoring 78s.

I have spent the past few days scouring these forums for tips and advice on getting the most from my transcriptions, but I still feel like I am missing some steps. Notably, my efforts seem to result in muffled, or uncomfortable sounding tracks.

It has been mentioned that the CNF is the cause of this, which I agree with, but I am unable to find a more satisfactory way in removing the background "hiss" noise without using the CNF. Is there perhaps better methods of applying it in small doses?

My hope it to produce results similar to this;

http://www.youtube.com/watch?v=P4OQNEmRucg

Non-muffled and little to none background noise. His set-up is likely much more advanced than mine, so it is a vain hope.

With that result in mind though, could anyone give me some step-by-step (small details and all for a newbie!) guide to how you would restore a typical 78rpm record? I would really appreciate it!

-Ben

[Dan McDonald]

Hi Anubis -
If you're getting muffled sounds, then you're getting too aggressive with the cnf.
Basically (everyone can add to this list)
1. Start with the best sound you can - transfer with a cleaned record and good stylus is first step. Flat transfer is best. Higher sampling (96kHz) is best because of the information in the impulse (clicks) noise.
2. De-click first. Don't be too aggressive as that can also introduce distortion, but don't be afraid to try various settings. The presets are just starting points, and every record has a unique history that affects how it's cleaned.
3. After you are satisfied with the de-clicking process, work on noise reduction. If there are multiple spots to take samples, you can kind of 'average' those spots in your head before you actually run it. Also, try different techniques here, such as artifact reduction or some of the "automatic sampling" techniques. Also try Doug's method, which I've had good luck with also. It involves a multi-pass using different fft sizes. I can find that link if you don't have it. Or Doug can post it.
4. Don't get rid of all the noise. If you're getting rid of all of it, you're also getting rid of some of the signal, and it can sound a little flat. If you leave some of the noise, it will sound better.
5. Experiment with the bypass button to help settle on different methods.
6. When you think you have it just about perfect, take it down a couple of notches so that it's not quite perfect. That will work better, in my experience.

Anyway, everyone else is better than me at this, so they can add more information and correct anything that may be wrong.

Here's Doug's method:
http://www.diamondcut.com/vforum/sho...tion-Questions



And this is from a note Doug sent me a while back (hope that's ok to post, Doug):

Smaller FFT bins are better for reducing low freq noise due to the improved frequency accuracy, however, at higher frequencies it induces artefacts. So the idea was the best of: (1) have a high FFT (4096 or 8k) and run it on frequencies from 10Hz ~1kHz then use a second lower FFT (512 04 1024) and run it on 1 - 20+kHz. The alternative approach (2) was to use a high FFT (4096) to increase the separation between noise and desired sounds by say <6db, then use a lower (or series of lower 1024, 512/256) FFT to further separate noise from content with less induction of artefacts. Both methods produce vastly different effects from single FFT runs.

Maybe also have a look at this image I generated a while back: http://www.78rpmrestorations.com/tec...3/allTests.jpg

That image was from Doug, not me.
Dan

[Craig Maier]

Just a point about de-clicking - - - I prefer the EZ Impulse filter compared to all the rest because of its effectiveness and ease of use. I use the EZ Impulse filter over the EZ Clean Filter and the Expert Impulse filter for my 78 work. Also, I reverse the RIAA curve (if an RIAA Preamp was used for the transfer) first before de-clicking with the EZ Impulse filter.

You can hear some examples of what the system is capable of at the Diamond Cut homepage by way of the CD snippets at this location:

http://www.diamondcut.com/store/inde...897496a7d4f34a

Just click on an album there and the songlist will pop up and you can listen. All of those tunes were cleaned with Diamond Cut.


Craig

[Marc Hildebrant]

Another thing that helps me.

Perform the impulse noise on the two channel version of the song, then convert to mono.

Also, if using the EzImpulse filter , set the crackle setting quite low.

Marc

[Anubis]

Thanks for the great info so far everyone.

Dan, the info about doing multiple runs with the CNF had something I didn't quite get. How do you "run it on frequencies"? Could you elaborate on this point in a bit more detail? These tools are still new to me, so most of this stuff is likely going over my head.

I have read on some other posts that using the EZ Impulse, then reversing the file, then running the Impulse filter again, finishing with another reverse is good a technique. Assuming this meant to physically reverse the direction of the file, I tried this and resulted in more than a few persistent errors. Any more info on this method would be welcome.

Another point I might be missing... Should I be using all these filters at once in the Multi filter? Would that produce better results than doing them one by one?

Thanks again for your time and consideration!

[Dan McDonald]

Hi Anubis -
I think you are referring to Doug's method, where you take a sample in the CNF, and then move the sliders in the graph all the way down to the bottom so that they are doing nothing, except for selected frequencies (such as above 1kHz). Leave them alone or adjust them as needed. Then you run the next one with essentially a mirror image, with sliders above 1kHz down all the way, and you are just working with the bass end of things. He also has one where he runs it for all frequencies and then the second pass is more selective.

I've tried the backwards runs also, but haven't had great luck with it. Some people have better luck. I have found it does sometimes get rid of more persistent clicks but it's much more difficult to tell whether you're doing damage to to music or not (since it's reversed). My guess is that if you're getting damage to the music, then the filter is too aggressive.

I use a multifilter whenever I can because it cuts down on quantization errors and some other types of errors. It may also give you a little more headroom to work with since it's all math until the final output.

Dan

[Craig Maier]

I think that another element of Dougs CNF method involves the use of a high fft count for the low frequencies (say below 500 or 1000 Hz - - - fft of 8192 or 16384) and a low fft count for the high freqnencies (above 500 or 1000 Hz - fft count of 1024). Thus, he is achieving high frequency resolution on the lower notes (thus eliminating more rumble) and producing less artifacting on the higher frequencies due to the lowered fft count.

Craig

[Craig Maier]

Oh - - - and do not forget the interpolate "I" key on your keyboard. If there is a stubborn click or pop, zoom in on it, highlight it and then hit the "I" key and it will be replaced with a calculated signal that should have been there.

[Dan McDonald]

Yes, Craig answered your question better than I did. I was just talking about how you select those frequencies, but Craig is talking about how you go about with the sampling, etc.

[Anubis]

What settings should I use for frequency reduction? IE: Attack, release, and attenuation settings. I think I may have these set wrong...

I've spent the better part of two days trying various methods. Always using Doug's "Final 78RPM DC8" at the end. The results are still rather bland, with obviously muffled music, or ugly background noise depending upon my CNF usage. I am quite amateurish at this, my only previous experience in audio editing was with lesser programs. (sorry!)


Would anyone mind sharing their complete rundown for restoration? The more detailed (small settings for each filter) the better!

To demonstrate my progress so far, here is a restored Decca side for your listening pleasure.

https://soundcloud.com/ben-hughes-63...-decca/s-w4Boz

[Dan McDonald]

I'm not sure what you mean by frequency reduction. Setting the attack and release is very tricky, and all you can really do is experiment with material. I tend to leave those alone except when I have really tricky problems, such as different "layers" of hiss, etc.

[Marc Hildebrant]

I checked out that Decca song you referenced.

I would say that you have taken a first step, but that much of the original "lively-ness" is now missing.

I would NOT get discouraged though. My early work was really bad until I started to learn more about both Diamond Cut Software and what music should sound like.

Could you post your settings for the various filters you are using ?

Marc

[Dan McDonald]

Also, have you gone through the tutorials? They will help a lot. There's a "super easy" one and another one that's more advanced, I believe.

Dan

[Anubis]

I begin with the EZ Impulse filter. Scratch set to 50, Crackle to 35, and Narrow Crackle to 75 to 90 depending. I also usually check the Solo/Brass filter. I use this once.

Then I use the Paragraphic EQ. Not sure how to describe the settings, but I adjust the blue dots to add some bass and treble. I use this once.

Next is the CNF. Attack is set to 80, Release set to 160. This seemed to work, though frankly I don't detect much difference when moving either of the settings. Attenuation ranges from 10 to 30 depending. I usually take a second-long sample of plain background noise. 2048 FFT size is what I often use. I use this only once, since I can't seem to figure out how to properly reduce selected frequencies with the CNF.

Last (but not least) I use Doug's "Final 78RPM DC8" multifilter, which seems to be very beneficial.


Dan, I had not checked out the tutorials. Being the type of person who thinks that I can master something by just messing with it enough, I never even looked. I'll go through them right now.

Thanks again for your replies everyone. I appreciate all the help! :]

[Dan McDonald]

Hi Anubis -
Good luck. I've been using the software for 14 years (just realized that!). I learn something new with every project.