78s - Groundbreaking discovery...

[here]

I think this is the first time anyone's managed to do anything quite like this...

Below is a graph which represents, to what I believe is a reasonable degree of accuracy, the frequency response of the recording equipment used by EMI in 1932 on a specific classical recording in Abbey Road Studio One. Frightening, isn't it?



But having devised a means by which this EQ can be determined, it becomes a relatively simple process to reverse this and cancel out the distorted response. The results are nothing short of amazing, creating a realism of sound literally never heard before from these recordings. (BTW it works the other way round - apply the above curve to a modern recording and it sounds frighteningly like the original old one...)

I've written more about the theory behind this idea and uploaded a 27 minute illustrated podcast interview to our website here. As a result of this discovery I'm in the process of remastering all our 78rpm recordings - the homepage of our site includes an up to date list of those currently available. Each has extensive streaming samples available to listen to.

The technique I'm using was discovered by accident and involves the use of modern reference recordings of the same piece to determine the expected frequency response averaged over an extended period of music - preferably the whole piece. Matching the older recording to this modern reference can have a dramatic effect - there appears to be a point where everything snaps into focus, almost as if by magic. The EQ only has to be slightly out for it simply not to work - and your ears instantly tell you when it's spot on.

This was all quite unexpected, in every respect, when I began the experimentation which led to this discovery...

Can you tell if that is using the new microphone designed by Blumlein about then, or the older (Western Electric?) one?

Paul

I don't have that information - and I suspect that the microphone is only a part of the problem, which appears to be unpredictable both in frequency response and in severity.

the use of modern reference recordings of the same piece

So - Andrew... was the modern reference recording made in the same Studio with the same acoustic placements and setup?  If not, the the curve also contains the differences in room response, which can be substantial as you already know. 

It is good that the correction curve improves things so much.  I notice that (excluding the extremely low HF roll off) the curve is mostly contained in a 10db range (i.e. +1db +/- 5db).  This 5db variation is substantially less that the normal response variation of many modern loudspeakers, especially anything made by Bose!

Excellent, Andrew.  Today is my father-in-law's 79th birthday, and this morning having just read this I have feverishly downloaded the Beethoven 1 / Toscanini restoration, burned it, printed out the notes, and will shortly be wrapping it and taking it round to him - and if he doesn't shed a few tears I shall be surprised.

If I confess that I've reflected my personal taste by running it through a touch of RaySpace you'll probably be appalled, but then I tend to put salt on my food in restaurants too, with complete disregard to the feelings of the chef.

I'll also confess that I hadn't previously actually visted your site before... but shall certainly do so again!

the use of modern reference recordings of the same piece

So - Andrew... was the modern reference recording made in the same Studio with the same acoustic placements and setup?  If not, the the curve also contains the differences in room response, which can be substantial as you already know. 

It is good that the correction curve improves things so much.  I notice that (excluding the extremely low HF roll off) the curve is mostly contained in a 10db range (i.e. +1db +/- 5db).  This 5db variation is substantially less that the normal response variation of many modern loudspeakers, especially anything made by Bose!

I suggest you apply the curve as shown to a modern recording and listen to the results! My research suggests that over the course of an entire symphony, quartet or whatever, the averaged frequency response of a particular work has more relevance than the finer nuances of acoustic placement and set-up.

I also found that it doesn't take much adjustment to screw this effect up. There really does seem to be a pretty narrow 'sweet spot', a bit like focussing a camera's lens, when suddenly everything becomes clear. Over and above this there is the restorer's judgement over broad EQ - treble/bass/mid - and perhaps the notching out of specific problems (lathe whistle, hum, etc.). But this is an effect I've simply never come across before...

(BTW there's now a short excerpt of that Beethoven before this treatment. It's also worth saying that on the scale of things this was by no means the worst - I've seen discrepancies of over 15dB on some recordings.)

[correction curve]

It is good that the correction curve improves things so much.

Just to be clear - the curve displayed on the Natural Sound page is the inverse of the correction curve. Note too that the variations shown here were not designed into the frequency response - they really do make things sound horrible!

Andrew, I'm sorry but I don't understand anything about the above posts but one thing did shoot into my head. I hope it's apt and not cynical. If something was so groundbreaking, I would have held back a bit from announcing it (I understand your excitement) done a little more research, posted my theories in an envelope, which I wouldn't open for copyright purposes, and then approach speaker companies and software companies and offer the studies for loot.
Is it possible to get a plug-in developed to convert the final mixdown aka Ozone etc and one mention of Bose and I'd be out the door trying to sell this to speaker companies but then, as I've said, I don't understand the above posts.
Well, why don't you make some money and get a little credit instead of the big guys.

Just a thought

You're not the first to ask me this - the truth is I blabbed before I even considered whether this was in any way protectable by copyright, which I suspect it isn't.

After all, there's a 'match' button in Har-Bal which effectively does what I'm doing - all I did was found an unusual application for it. I also don't want to be (a) left having to re-EQ every pre-1950 recording in the world, or (b) trying to explain myself any more than I already am.....

(Actually I do it by hand most of the time, but that's because pressing the button merely moves you towards an EQ match - they really don't want you to do this, which in just about any other circumstance, you shouldn't.)

I've looked for patterns between the EQ curves I've derived for different recordings, especially where it's worked well, and can't find any correlation - I was hoping there might be some generic settings one could try, as it's a laborious process which involves the purchase or ownership of at least one other copy of each recording you're working on, and preferably more than one.

To try and explain in more simple terms:

I found that if I analysed in Har-Bal the complete frequency response of a modern (2006) recording of Beethoven's 7th Symphony, and then matched a poor-sounding but brilliantly-played 1936 recording to this, that it had a dramatic and unexpected effect on the sound quality, giving it a realism I'd never heard from a 78 disc before.

I then repeated this with three other recordings I was working on with pretty much the same remarkable results.

I've now added another 21 recordings to the list, all of which have benefited tremendously from the technique.

I am keeping one or two essential details and tricks up my sleeve, you'll be relieved to know, but in the suspicious world of historic recordings, the answer "I'm not telling" is simply not allowed, nor commercially viable!

Pity about that indeed. It's just a shame that all the knowledge, blood, sweat and tears can be reduced to people stealing all that know-how. I wouldn't even tell anybody I use Har-Bal
I bought years ago a plug-in for Photoshop and when I installed it I said, "Why did I buy this, I could've done that myself" All the program did was speed it up a bit but had I known, I would've saved £100.
Even last night on Coast (TV Docu) during the war, a Polish guy, working for the British, invented the handheld minesweeper which saved thousands of lives for the allies clearing German minefields. It was kept a secret 'til well after the war and by then, nobody wanted to talk about the war so he didn't get any money or recognition. He did however get the Polish equivalent of the Iron Cross but that doesn't get you lobster on a Caribbean Island every day.

Best 'o' Luck anyway Andrew

Indeed, Har-Bal is extraordinary.  Since I began using it a year and a half ago I've discovered my ears could not tell me everything necessary to get the most out of my files.  There are just too many subtle sonics to discern.  What I thought I'd done a good job on in 2004 and prior I took through Har-Bal and found appalling graphs of the frequency responses.  Once harmonic balance is achieved correctly, the listening fatigue goes away and detail not previously heard comes forth.  Surprisingly, often only a slight change is needed -- as opposed to trying to get the curve to be absolutely flat, which I've found usually does not achieve pleasing results on the music I deal with. I consider it my most indispensable tool.

I was listening last night to some 1935 recordings of Bach preludes and fugues on the piano by Edwin Fischer, which I've known very well now for the last 3 years. In one of the preludes there's an inner line of repeated notes which holds the melody, as the upper and lower notes weave around it. Playing this and bringing it out requires great skill and subtlety. Until the recording had been through this process, and in this case it was relatively good to start with (as most from Abbey Road Studio 3 by the mid-thirties tend to be), it's a line I'd never noticed before - and yet it's the crucial part of the piece, and one of the most beautiful and subtle pieces of keyboard writing and playing I've ever heard...

I've just added to the Natural Sound page an example of a 2006 recording with the 'Emperor Concerto' 1932 EQ applied to it - you might find it somewhat alarming!

I couldn't hear the difference with the curve applied to the modern recording on the website.  Perhaps the wrong file is presented for listening. 

Nonetheless, the theory checks out logically and is an exciting development. 

It reminds me of convolution applications. 

I've just added to the Natural Sound page an example of a 2006 recording with the 'Emperor Concerto' 1932 EQ applied to it - you might find it somewhat alarming!

It might help a little if you present a before/after of exactly the same short passage, both for a modern recording and a restored one. Whilst I can certainly hear very clearly just how 'wrong' the sound of a modern recordingi is when you apply the EQ from an unprocessed 78 to it, a lot of people who are not as used to what to listen out for might find the whole process somewhat easier to comprehend, perhaps?