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: Transcribing 78's (Read 8493 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Transcribing 78's

About 10 years ago now I embarked on a year+ project to transcribe my collection of 78's onto CD. I spent quite a lot of time experimenting and deciding on the best approach before I attacked the pile of records. I was using a moving magnet cartridge with a stylus profile supposedly optimised for my records (mostly 1920's Columbias). The methods I tried were :


- Flat transfer in stereo
- Flat transfer in mono (cartridge output commoned at the headshell)
- Transfer in Stereo but using RIAA playback eq
- Transfer in mono using RIAA eq
- Transfer via a preamp using RIAA eq and a "scratch filter" and tone controls adjusted to get the most pleasing sound.

I ended up using mono transfer with RIAA eq as it seemed to eliminate a lot of clicks and other surface noise (presumably due to phase relationships) while the inherent treble cut and bass lift seemed to be the best compromise overall and meant I needed to do less DSP. However, there was one major drawback to this approach. That was mistracking distortion due to the damage that the old, heavy arms and steel needles had inflicted on the grooves. I should have realised this at the time but I recently found that in most cases the mistracking ditortion is most evident on one channel at the beginning of a record and on the other channel at the end.

I'm seriously thinking of revisiting the original project (most of the 78's in question are sort of family heirlooms so I want to do the best job I can) and recording them in stereo then crossfading or simply switching channels at some point to get the least damaged signal from the groove. However, before I start I wondered if anybody else here has done anything similar and might have some other suggestions to improve my recordings

Transcribing 78's

Reply #1 –
I think cliveb might have a contribution...

Also...
http://home.clara.net/rfwilmut/repro78/repro.html


An attempt at the correct transfer EQ (i.e. not RIAA or flat) in stereo is the "best".

Correct stylus is hugely important. One stylus for all 78s = wrong.
Correctly centred records are important.
Documenting the EQ curve used during the transfer is important (as it can be corrected later).
Clean records are important.
Correct speed is marginally important (can be corrected later).

Mono is wrong - both for the reason you've found, but also because some delickers work better in stereo with the signal summed to mono after processing. However, good restoration tools can work well enough with a mono source for this not to matter.

You should end up with two copies of everything - the raw stereo transfer, and a delicked denoised correct EQ correct speed mono version.


I collect 78s, but I'm not sure that these "family heirlooms" are always the best way to hear the music. Reissues from the original metal masters (where they exist) should be far better.

Cheers,
David.

Transcribing 78's

Reply #2 –
Quote from: 2Bdecided on
I think cliveb might have a contribution...

Also...
http://home.clara.net/rfwilmut/repro78/repro.html


An attempt at the correct transfer EQ (i.e. not RIAA or flat) in stereo is the "best".

Correct stylus is hugely important. One stylus for all 78s = wrong.
Correctly centred records are important.
Documenting the EQ curve used during the transfer is important (as it can be corrected later).
Clean records are important.
Correct speed is marginally important (can be corrected later).

Mono is wrong - both for the reason you've found, but also because some delickers work better in stereo with the signal summed to mono after processing. However, good restoration tools can work well enough with a mono source for this not to matter.

You should end up with two copies of everything - the raw stereo transfer, and a delicked denoised correct EQ correct speed mono version.


I collect 78s, but I'm not sure that these "family heirlooms" are always the best way to hear the music. Reissues from the original metal masters (where they exist) should be far better.

Cheers,
David.

Thanks very much for the reply.
My records are not family heirloms in the sense that they've been handed down from generation to generation. They are simply records made by bands that my great-uncle played with and I actually collected them myself in the 90's. The quality of the records varies a great deal but in cases where I had a poor copy I was, on the whole, able to borrow a better copy from another collector. I did my best to clean the records and my stylus was provided by a specialist firm and was supposedly a good match for the records. On the whole I don't think I have much scope for improving things there.

I got the impression from collectors that 78 was at best a notional standard and that all sorts of speeds were actually used. I tried initially to adjust to what I thought was a correct speed but faced with a recording playing back somewhere between Eb and E which do you chose? I think Eb is probably most likely for a band containing brass instruments but maybe it should really be in D or F. I'd be very grateful for any help you can give me on that. Similarly centering the records. I can't remove the spindle from my turntable so I can't see any way round living with the records as they are (most are not too bad in that respect actually). Do you have a method you use?

The earliest records are acoustic recordings so I had no idea what to do with them eq-wise. I opted for what sounded best to my ears. The later ones were recorded using the Western Electric system. The National Sound Archive gave me the eq curve for that process but as I had no way of implementing it in the analogue domain - other than crudely using a graphic equaliser - I tried to do it after the event in the digital domain. I didn't like the result very much - it seemed to be more successful at emphasisng surface noise than I was at removing it - so I gave up trying to restore the records to their original state and again went for what sounded best to me. I realise that, to a purist, this might be undesirable but I was quite surprised in my dealings with collectors to find that they invariably used modern-ish record players and amplfiers when listening, so I didn't think my approach was too radical. I do actually have some EMI LPs from the 60s/70s that contain some of the records in question.  Do you think I would I do better to aim for a similar tonal balance to those?

Thanks again for your help

Transcribing 78's

Reply #3 –
It would be a great project to write a software that interpolates from several takes (at best from different records). The algorithms are already there and are used, for example, in video forensics where you interpolate high resolution images from several successive low resolution frames. Filtering of both digital images and digital audio is very cognate.

Transcribing 78's

Reply #4 –
Quote from: 2Bdecided on
I think cliveb might have a contribution...

Actually I don't have much experience of digitizing 78's. I did about a dozen for a neighbour. I've also done some mono LPs.

For what it's worth, my main piece of advice is to always record in stereo. Some damage will be on one side of the groove wall only, and by recording in stereo you have the opportunity to pick the undamaged channel and paste it over the damaged one for each individual glitch. And as you mentioned, if you've got bad mistracking on one channel at the start, then the other at the end, of course you can do some sort of crossfade between the channels to minimise the effects. But do the individual impulse noise edits first.

The benefits of mixing down to mono (a typical 3dB reduction in surface noise) accrue even if you leave it to the very end and do it in software.

Transcribing 78's

Reply #5 –
Quote from: rpp3po on
It would be a great project to write a software that interpolates from several takes (at best from different records). The algorithms are already there and are used, for example, in video forensics where you interpolate high resolution images from several successive low resolution frames. Filtering of both digital images and digital audio is very cognate.
I don't think video super resolution and combining the signals from different pressings of the same record (presumably what you meant?) have much in common at all.

I've tried both. I've never made the latter work!

Has anyone ever made it work properly? It sounds "easy" at first, but sub-sample syncing of two fundamentally different signals isn't easy.

"1 good copy of a record + the best restoration tools currently available" gives a great result.

"A ?non existent? method + multiple copies of the same record" gives a few dB less noise for each copy of the record: double the number of copies = half the noise. So for 18dB noise reduction, I need 8 good copies of the record, and I need the syncing algorithm to work!

In comparison, I get easily get 20dB noise reduction out of several prosumer restoration tools - better than that if the noise is mostly impulsive.

Cheers,
David.

Transcribing 78's

Reply #6 –
delete duplicate post - how on earth did that happen?!

Transcribing 78's

Reply #7 –
Quote from: rpp3po on
It would be a great project to write a software that interpolates from several takes (at best from different records). The algorithms are already there and are used, for example, in video forensics where you interpolate high resolution images from several successive low resolution frames. Filtering of both digital images and digital audio is very cognate.

Unfortunately I think that most of what needs to be fixed is due to damage to the surface of the groove and thus will be present in every take.

I recall a brief discussion recently of the possibility of taking more than one sample of a record and combining those in a way that would eliminate most surface flaw type problems. The only problem would be to adequately synchronize the multiple takes.

Edit: David beat me to it.

Transcribing 78's

Reply #8 –
Quote from: 2Bdecided on
I don't think video super resolution and combining the signals from different pressings of the same record (presumably what you meant?) have much in common at all.


Mathematically it is almost exactly the same problem. A digital image's digital representation is a sequence of discrete frequency domain values, the same representation that you transform digital audio into before filtering. You have the exactly same relationship between bit depth, signal to noise ratio and quantization artifacts, even the counter measures (dithering and noise shaping) are comparable.

I don't know what exactly you have "tried", but synching & averaging or subtracting is not sufficient (or the best possible), video doesn't work that way, either. You need pattern recognition algorithms to extract inter-frame (on the time axis) correlation to be able to subtract non correlating artifacts.

Of course, it really only makes much sense with different records at hand. It could show some effect with only one record and several takes, but probably not much.

Transcribing 78's

Reply #9 –
Quote
For what it's worth, my main piece of advice is to always record in stereo. Some damage will be on one side of the groove wall only, and by recording in stereo you have the opportunity to pick the undamaged channel and paste it over the damaged one for each individual glitch.
Clive is modest, but his own software Wave Repair ($30 USD) has a tool that makes this easy (once you find & zoom-in on the defect).  Frequently, this works quite well on stereo records...  A few milliseconds of mono with the audio from one channel missing is usually undetectable.

Transcribing 78's

Reply #10 –
Sorry to bump this one but I'll be starting soon.

So, other than recording in stereo and and doing my DSP in stereo does anybody have anything to add? I'm especially interested in flat transfer Vs RIAA but any other hints would be gratefully received

Transcribing 78's

Reply #11 –
If you do RIAA in hardware there are 2 possibilities, it is the right correction or it isn’t.
If you do a flat transfer you don’t have any RIAA at all.
I think it is easier to find software applying the right RIAA to a flat transfer than finding software correcting the wrong RIAA

Maybe a link like this is of use: http://www.channld.com/pure-vinyl.html
TheWellTemperedComputer.com

Transcribing 78's

Reply #12 –
Quote from: rpp3po on
Quote from: 2Bdecided on
I don't think video super resolution and combining the signals from different pressings of the same record (presumably what you meant?) have much in common at all.
Mathematically it is almost exactly the same problem.
It really isn't! With real video super resolution, by definition you must have aliasing in the image, meaning each frame is a sub-set of the actual Nyquist sampled data. If you can "synchronise" the images, and if you are lucky, you can get back all the data - because as long as there is some movement, then in the spatial domain there is data in one frame which is completely absent in another. In the classical maths, there are no "errors" to correct - though in real world cases there are, due to M-JPEG, MPEG etc etc artefacts, sensor noise etc.

A transfer of a 78 is already Nyquist sampled by any half-decent sound card. There's no data "missing" - you have absolutely all the data present, but you also have noise on top. If you can synchronise two transfers, you can delineate the noise from the data, but it's a different mathematical problem from video super resolution.


Quote
A digital image's digital representation is a sequence of discrete frequency domain values, the same representation that you transform digital audio into before filtering. You have the exactly same relationship between bit depth, signal to noise ratio and quantization artifacts, even the counter measures (dithering and noise shaping) are comparable.
This is basically true, barely useful in practice, and irrelevant to the current problem.

Quote
I don't know what exactly you have "tried", but synching & averaging or subtracting is not sufficient (or the best possible)
That's true. In the audio case, some kind of median filtering across the different synchronised recordings is probably better than a straight mean average. The problem I've found is that you don't just have clicks and noise - you have distortion too. Some conventional digital NR can tame this is a bit on 78s, whereas multiple copies will usually have near-identical distortion.

Quote
video doesn't work that way, either. You need pattern recognition algorithms to extract inter-frame (on the time axis) correlation to be able to subtract non correlating artifacts.
There's two stages with both audio and video: firstly synchronising, secondly doing something with the synchronised information. Advanced algorithms have a feedback from the second to the first, but these are two discrete stages.

In audio, the second stage is just a fancy way of saying "find a good way to remove the differences and only keep the common parts". There are lots of algorithms for that. There is no comparable process in video super resolution - the second stage needs to integrate the differences into the final signal, because in video super resolution, it's the fact that the different signals contain different real information that makes it useful.

Quote
It could show some effect with only one record and several takes, but probably not much.
That would only remove the noise from the replay and capture hardware - which in the case of 78s, can easily be made insignificant. You could try slightly different sized / shaped styli to track slightly different parts of the record groove, and then process that. You'd still get a lot of clicks common to the different recordings, but not all of them.

Cheers,
David.

Transcribing 78's

Reply #13 –
Audacity has a predefined EQ for Columbia 78's.  Might be worth checking out.

Woody

Transcribing 78's

Reply #14 –
Quote from: 2Bdecided on
A transfer of a 78 is already Nyquist sampled by any half-decent sound card.

This made me think. Given that there's little or no "real" signal above 7kHz on most of my 78s would I gain anything in terms of noise suppression by simply using a sample rate of 22050 or maybe 16000? Or would that cause problems with the subsequent click removal, NR etc?

@woody-wooward
I have the eq requirements for Columbias but they are for playback : IE reversing the eq that was applied for the recording. Is this what the Audacity eq setting is for? I don't have Audacity, so can't check

Thanks

Transcribing 78's

Reply #15 –
Quote from: botface on
@woody-wooward
I have the eq requirements for Columbias but they are for playback : IE reversing the eq that was applied for the recording. Is this what the Audacity eq setting is for? I don't have Audacity, so can't check

Thanks

Yes, digitize your records flat (without eq) then apply "Columbia 78" EQ using Audacity.

Woody

Transcribing 78's

Reply #16 –
Quote from: botface on
Given that there's little or no "real" signal above 7kHz on most of my 78s would I gain anything in terms of noise suppression by simply using a sample rate of 22050 or maybe 16000?
No, that's a very bad idea.

You can trivially remove everything above a certain frequency using a software filter. Why use your sound card's ADC to do it, while simultaneously forcing any sound card filter imperfections right down into the audible range, and delivering a file that's not suitable for CD?

If you do decide to remove high frequencies, you should do so using a low pass filter with a reasonably gentle roll off. A brick wall, which what what you're likely to find in an ADC, will sound horrible.


The ear often falsely associated some of the noise as part of the music - if you brutally filer the top-end, it can sound like you've removed part of the music, even if you can prove that there was nothing real there to start with. Of course analogue systems themselves roll off (comparatively) gently - it won't be the case that you have a good signal at 6.9kHz, and nothing at 7.1kHz.

Cheers,
David.