78 RPM records: EQ, getting rid of crackles and surface noise.....

Unquestionably, the most important technical advance in the history of recorded sound was made in 1948 with the introduction of the plastic, long-playing (LP) record by Columbia Records. Earlier discs were recorded and replayed at anywhere between about 65 to 85 RPM (revolutions per minute) and are now universally referred to as "78s". (See below for more information about rotation speeds). These earlier records were not made of PVC plastic, but of shellac; the resin secreted by the lac bug, on trees in the forests of India and Thailand. A ground-slate filler was added to the shellac to make the discs hard enough that they could support an acoustic-gramophone tone-arm weight measured in ounces1.

The decision to support replay with acoustic machines well into the electrical recording period is something of a shame since it is these particles of slate (and not the shellac itself) which contribute the harsh, surface-noise that we all associate with pre-vinyl records. It's worthwhile remembering that it was less due to the new, plastic medium as it was to the decision that the new format would only support lightweight, electrical tone-arms that underwrites the low-noise of the LP medium.

Nevertheless, the recorded legacy of eighty years of planet Earth exists on "78s", so we need to take them seriously and see how the information might best be teased from their rough grooves!

1 One ounce is approximately equivalent to 28 grams.

Performance of 78 records

An understanding of the performance of 78 records is greatly enhanced by the existence of test discs which may be hunted-down at reasonable cost. One, belonging to Pspatial Audio is illustrated right.

With a good modern cartridge and the right stylus, the frequency-response derived from DECCA Test disc K.1803 (pressed in 1956) is somewhat chastening. It's easy to dismiss these pre-vinyl records as "stone-age", but we obtained a frequency response within ±1dB from 30Hz to 14kHz! These results were derived using the Phædrus Audio PHLUX cartridge described here and Stereo Lab.

Looking at the spectrum of a 1kHz test-tone on the sister-disc (DECCA K.1804) we can see that distortion is mostly second harmonic at about 1.5% and third harmonic at about 0.5% (a creditable performance). And, whilst the spectrum of the noise from a 78 RPM record rises from about 500Hz to very high frequencies as illustrated below, the mid-band noise is relatively low: in the case of the illustration about 60dB below the 1kHz test tone - a very respectable figure.

Interestingly, the spectrum of the unequalised groove-noise rises at about 3dB/octave (not 6dB/octave) meaning that the level of the noise is 10dB higher at 10kHz than it is in the mid-band. No clearer demonstration could be given of the advantages of pre-emphasis.

As an example of just how good 78 records can sound with pre-emphasis (just, sadly as they became obsolete) is given in this example from a 1956 recording. This recording is made with RIAA pre-emphasis (no noise-reduction was applied). Seventy-eight discs made in Europe at this time often use Coarse groove (BSI 78) (EQ-C in Stereo Lab).

Neither of these equalisations are suitable for earlier 78 discs of the thirties and forties or earlier and the equalisation of all shellac-era records is discussed below.


Equalisation of pre-LP (or shellac-era) records

There appear to have been a plethora of recording characteristics used during the shellac period (and possibly even more opinions on those characteristics!) The free digital audio workstation program Audacity, made freely available under the terms of the GNU General Public License, has many contributors to their Wiki page on early recording characteristics. It may thus be taken as a good synthesis of older references and modern, investigative work.

On the strength of this Wiki, it seems an armoury of different equalisation curves are required for the reply of shellac era discs. However, we advise caution regarding choosing equalisation on the record label alone such as those given in the Audacity Wiki.

Record companies and their labels

The marketing of gramophone or phonograph records has been slick since the very earliest days of the medium and many of the plethora of record labels were simply sub-brands of the larger companies for the sales of older or alternative product through different sales channels. (See the DNA of record labels.) Top, mid and budget price records from the same manufacturer have existed since before the turn of the 20th century and often the labels were adopted to distinguish these categories (and perhaps protect the value or the sales-channels of the master-brand). As a vivid example of this, consider that, by 1939 only two major record companies existed in the UK: the mighty EMI and Decca. Yet records were available with a vast range for labels.

Other labels were owned by mail-order companies or "value-brand" outfits who obtained their masters from a variety of sources. And there are even many examples of paste-overs in which the records themselves were bought (rather than the masters) and a new label "pasted-over" the original. Clearly, when records or masters were sourced from a variety of sources, there is no hope that consistent equalisation was applied and to advise that all records from such and such a label should be equalised with a particular characteristic is often misleading.

Labels issuing their own, independent material were not immune to variations (developments?) in equalisation either. Because the recording characteristic (and the precision of the characteristic) applied to a recording was, and remains, a function of the mastering studio and not of the record company itself. Only the majors operated their own mastering facilities.

We have tried to be mindful of these issues in Pspatial Audio's Equalisation Guide where we have included notes about the origin and operation of the labels where this might be appropriate to the choice of equalisation.

For a deeper discussion of the details of the recording-characteristics used in Stereo Lab, read on.....

Turnover Variations

A good place to start, when considering how best to process the fifty years of shellac-era recordings, is to develop an understanding of the response-variation which results from a mismatch of recording and replay time-constants.

With this understanding we can make sensible choices about how many practical replay equalisers we require to cover the variation in the recorded characteristics.

Historically, turnover (the breakpoint which determines the shift from constant-velocity cut to constant-amplitude at the bass end of the response) pre-dated pre-emphasis. So let's start looking at the effects of a mismatch in the turnover frequency in the record and replay circuits.

The transfer function of a response with a falling bass characteristic (a one-pole, high-pass filter) such as in the filter we would find prior to the cutter amplifier is,

    H(ω) = jωt1 / (1 + jωt1)

where t1 is the time-constant of the filter.

The transfer function of a circuit with a rising characteristic at bass frequencies (such as we find in the replay equaliser) is,

    H(ω) = (1 + jωt2) / jωt2

Unsurprisingly, provided t1 = t2, for any given value of ω (frequency), because one expression is the reciprocal of the other, if we multiply one by the other, we always get unity: a perfectly flat frequency response. Taking into account that t1 might not equal t2, the full expression is ,

    H(ω) = [jωt1 + (jωt2 . jωt1)] / [jωt2 + (jωt1 . jωt2)]

Now, because the expression (jωt1 . jωt2) is going to be very small compared with either jωt1 or jωt2 and because it shows up on the top and bottom of the equation, we can ignore it to a first approximation and thereby simplify this rather unweildy expression to,

    H(ω) ≈ jωt1 / jωt2

Where t1 is the turnover time constant of the cutter amplifier and t2 is the turnover of the replay equaliser. This is a very convenient finding because it means that the simple ratio of one time constant to the other (or one turnover frequency to another), it's possible quickly to derive the maximum frequency response variation due to the mismatch. Thus,

    max variation = t1 / t2 or f1 / f2

where f1 = 1 / (2.π.t1) and f2 = 1 / (2.π.t2) .

Thus, with a mismatch of 12% in the turnover frequencies, we will get a maximum mismatch of,

    20 log (f1/ / f2 ) = 20 log (1.12/1) = 1dB.

By the same principle, a 26% variation will give a variation of 2dB... and so forth. The graphs below indicate the simulated variation in the ratio of t1 to t2 by 25% in both directions and confirm the reasoning above.

   

Given that recording characteristics during this period were specified to +/- 2dB, it seems reasonable to take the 26% variation figure as a practical limit on acceptable variation of record vs. replay time-constants (or frequencies). Thus a record cut with a 250Hz turnover but replayed with a 300Hz turnover is acceptable because,

    20 . log (300/250) = 1.58dB.

But the same record played with an equaliser with a 350Hz turnover would result in a response variation of,

    20 . log (350/250) = 2.9dB which would not.

Pre- and de-emphasis

At the high-end of the frequency-scale, our job is made easier by the preferred annotation of the recording characteristic during this period. By expressing the upper time-constant in terms of loss at 10kHz in decibels, we have data upon which we may make our +/-2dB judgement without further calculation.

Recording characteristics

Below we have regrouped the list on the Audacity Wiki page into more mangeable groups.

American 78 electric-era practice

  • Western Electric: Very early Columbia and Victor recordings (1926) used a bass turnover frequency of approximately 250Hz
  • American 78 – 250N-6/250N-8: Common settings for many American shellacs.
  • Columbia 78 – 300N-16 American CBS-Columbia shellacs (1938 - 1948); although we question this, see below.

Using our above technique and criteria for deciding about the seriousness of recording characteristic mismatch we can say the following of these three characteristics.

The mismatch of 250 and 300Hz turnover in record/replay is below our threshold since 300/250 = 1.2. Thus 300Hz will serve as a compromise turnover frequency for all American discs and give a frequency-response error below 2dB for any disc.

That the same cannot be said for the pre-emphasis hardly needs saying: with an apparent variation of 16dB at 10kHz, three de-emphasis curves are required: none; approximately 6dB; and 16dB.

To cover all the records, Stereo Lab has three dedicated shellac-era characteristics:

  • 300 - 0 : called EQ-G
  • 300 - 6 : called EQ-H
  • 300 - 16 : called EQ-J

European 78 electric-era practice

Once again, we have regrouped the list of European characteristics on the Audacity Wiki page into more mangeable groups.

  • Blumlein 300 - (300N-0): A British traditional for Gramophone Company, Decca, Columbia and EMI (1930s – 1944).
  • 500 Flat - (500N-0): Apparently used by British Columbia, EMI, His Master’s Voice, MGM and Parlophone between 1931 and 1953 (although we dispute this).
  • Decca 78 – (300N-5.5): Decca and London shellacs featuring the ffrr (full frequency range recording) system.
  • European 78 - (250N-0) : a common setting for European shellacs (1926 – ca. 1944), especially for Columbia and His Master’s Voice produced by EMI (UK), Cetra and Cetra-Soria.
  • Telefunken (400N-0): used by European Ultraphon, Supraphon and Turicaphon from 1929. Also used by Telefunken – after the takeover of Ultraphon – until mid 1950s.

Apart from the obvious comment that it appears odd that EMI/HMV are attributed to have used three different characteristics concurrently (which seems very unlikely), we have here two groups: the 250Hz/300Hz turnover group (a ratio which is below our 2dB threshold) and a 400Hz/500Hz group, a ratio too which too is below our threshold if a 500Hz turnover is chosen.

We can therefore cover all these curves with respectable compromises with a turnover at 300Hz (which we can use for all American recordings too) and another at 500Hz; the latter being the basis for more modern characteristics like RIAA as well.

Thus we can say:

  • Blumlein 300 - (300N-0) recordings should be equalised using EQ-G.
  • 500 Flat - (500N-0) recordings should be equalised using EQ-E.
  • Decca 78 – (300N-5.5) recordings should be equalised using EQ-H.
  • European 78 - (250N-0) recordings should be equalised using EQ-G.
  • Telefunken (400N-0) recordings should be equalised using EQ-E (500/400 = 1.25 which is below 2dB error.)

Outsiders - Acoustic discs and the 630Hz turnover

In the interests of completeness, a couple of other characteristics are annotated on the Audacity page under Table 2: 78 rpm Shellac Labels and Their EQ.

The first involves an annotated 800Hz turnover. These entries are all related to acoustic or re-issued acoustic recordings and behind this value of turnover is a serious mistake. Discs of the acoustic period, the author(s) claim are cut with a "clean constant-amplitude characteristic". Curve 800N-16” EQ curve is ths recommended for these discs because it is close to a characteristic which falls throughout the whole audio range by 6dB/octave.

This is incorrect: an ideal acoustic disc recording displays a constant velocity characteristic throughout its recordable range and does not require equalisation; except possibly on subjective/artistic grounds as discussed here.

The 800Hz turnover is therefore discounted from the list of necessary equalisations. In any case, Stereo Lab incorporates EQ-S (Displacement) which generates an accurate amplitude response from a velocity pickup should this be required.

The second outsider is 630Hz turnover which only the Brunswick label seem to have employed; and then only from the end of World War II to about 1954.

Two points about this value. Firstly, this number is suspicious. It's such an odd value: yet surpisingly exact. And it's remarkable that 630μS is the time-constant which is the equivalent of a frequency of 250Hz. It seems entirely possible that, somewhere in the murky past, turnover time-constant has got misinterpreted for the turnover frequency.

History bears this theory out. Brunswick Records had been sold to to Warner Bros. in 1930 and Warner had discontinued the Brunswick label by the end of the decade. In 1941, Warners sold the Brunswick label to American Decca along with all masters recorded prior to December 1931. Decca revived the Brunswick label, but mostly for reissues of recordings from earlier decades; particularly Bing Crosby's early hits of 1931 and jazz items from the 1920s - a period during which we know Brunswick were using 250Hz turnover.

BBC 2dB/octave characteristic & Western Electric Type 361 condenser microphone compensation

Other odds and ends include the putative BBC 2dB/octave characteristic & Western Electric Type 361 condenser microphone compensated characteristic.

Both these empirically derived equalisations, derive from work by Peter Copeland who proposed 3 special equalisations to compensate for the overall transfer functions of certain historical shellac-era recordings.

In the case of the BBC 2dB/octave case, Copeland proved (to himself at least!) that the BBC equalisation filter used for transcription recordings made between 1926 to 1933, (which ought to have been "Blumlein", thus 300-0) was over terminated and this resulted in a small loss of HF relative to the theoretically flat response above 1kHz. Because this equipment was used to master records released by small, British labels (Aco, Broadcast, Linguaphone, Vocalion), there is an argument these too should be equalised according to this putative BBC 2dB/octave characteristic.

In the case of the Western Electric microphone characteristic, Copeland proposed compensating for a peak in the response of the early Western Electric Type 361 condenser microphone (of about +7dB at 2.9kHz).

Whilst Copeland's experience is beyond question, these compensations go considerably beyond the role of providing a complementary characteristic to the network employed on the recording side. These empirical characteristics are thus justified on artistic, rather than engineering grounds. In Stereo Lab, we have confined our ambitions simply to engineering a unity transfer function for any given recording-characteristic filter and thus we have taken the BBC curve and the WE characteristic both to be straightforward 300-0 characteristics best treated with EQ-G.

NAB / Columbia 78s with 16dB pre-emphasis

Stereo Lab includes EQ-J with 16dB de-emphasis at 10kHz because of the multiple entries in the Wiki page which refer to this equalisation. However, we believe that this equalisation was rarely, if ever, used for 78 RPM shellac discs.

This NAB recording characteristic has its origins in equalisation of transcription discs used at radio stations. True, this appears to be within the period of commercial shellac discs, but the NAB curve only ever applied to lateral-cut 33⅓ transcription RPM discs recorded for later broadcasts (this was before magnetic tape was available for these duties). The NAB recording characteristic is of the period, but it was not applied to 78 RPM records.

A clearer narrative

Stripped of much complication (and frankly some muddle), we can see a much clearer narrative emerging from this story. We can see that, far from the impression of hectic chaos given by the tables of different characteristics, the turnover frequency of gramophone records stayed more or less constant at about 300Hz (give or take 20%) from the earliest days of electric recording to the introduction of the LP whereupon, with the introduction of this much more fragile medium with a much finer groove-pitch, the turnover frequency jumped to about 500Hz, from which it has never strayed in all the years since.

Similarly, pre-emphasis is a technique which we can usually ignore until after WWII at which point it began to be adopted enthusiastically in America and with caution elsewhere. It is very unlikely that pre-emphasis ever exceeded about 8dB at 10kHz except after the introduction of the LP.

The truth is, the world of recording equalisation wasn't really chaos at all: it was actually very stable over decades. Only in the last ten years of the life of 78 RPM discs did the situation regarding equalisation become really cloudy, and this is due to the incompatible views of the role of the 78 in the dying days of the medium. For example, the British continued to make 78s for markets where acoustic machines still flourished (eg. in India) and this meant sticking to simple, turnover-only equalisation. Whereas in America, the preference was to align 78 recording practice with that of vinyl LPs with a raised turnover frequency and plenty of pre-equalisation. It's perhaps the confusion of this last-gasp-decade of the 78 medium which has left the enduring impression that equalisation of 78 records is a highly variable and complicated affair.

In all cases Stereo Lab provides the appropriate equalisation curves for shellac discs of all eras. These filters are digitally implemented versions of real-world, analogue filters which possess a conjugate phase-response of the original, hardware recording filters. This ensures that, not only is the amplititude vs.frequency response of the replay-chain is maintained as level, but - unlike the FIR filters available in Audacity which are non-causal types - the phase-response is corrected as well.


Groove sizes and styli for "78" records

Groove dimensions changed during the era of the shellac record, and thus so did the dimensions of the styli best shaped to fit them. The experts here are the Expert Stylus Company (Tel: +44 (0) 1372 276604. Email: info@expertstylus.co.uk) and their recommendations for records made within the following periods are as follows:

Manufacturing Date Conical stylus Elliptical stylus
Pre 1920 0.0040" 0.0040 x .0012"
1920 - 1939 0.0035" 0.0035 x .0012"
1939 - 1966 0.0028" 0.0028 x .0009"

Speeds

Before the introduction of electric recording in about 1925, records were recorded with a wide range of speeds. Early records by the tenor and prototype superstar Enrico Caruso replay correctly at 68 to 70 RPM; a full 13% below nominal "78" speed. If these records are replayed on a straightforward record-player with no adjustable speed, the degree of mismatch between recorded and replay speeds is unacceptable, the pitch being one whole-tone high. As Roger Beardsley1 picturesquely puts it,

"... to play [Caruso's records] at the standard 78rpm produces a travesty of the the original recordings, with the great tenor sounding more like Demis Roussos."

Many examples of other records exist from this pre-electric period which replay correctly well above the standard 78 RPM. Speeds in the mid 80s are quite common. These records will play a semitone to a whole-tone flat when played on a standard 78 deck.

The coming of electric recording brought some standardisation, but it was not complete, and the correct speed for electrical recordings can still range between about 73 RPM to 83 RPM, a variation of 6% or about a semitone either side of the correct pitch2.

These issues influence the correct choice of record deck for replaying "78" records.

Cartridges for 78 RPM records

Do not be tempted to buy a special mono cartridge for recording needle-drops of "78" records as explained here.

Phædrus Audio's PHLUX active moving magnet cartridge may be fitted with a general-purpose 3 thou' 78 stylus. This is an ideal combination for needle-drop recordings for treatment in Stereo Lab and is suitable for needle-drops of all "78" records made from 1920 onwards.

Crackles!

Needle-drops of 78 records reveal a persistent and unique problem. Quite unlike LPs, which betray occasional "clicks and pops", shellac records exhibit what can only be described as "crackles" which are pops and clicks of a very short duration and which are present almost all the time. Presumably this difference is because the 78RPM medium is spinning faster, so the stylus encounters dust or damage more often than is does when it reads an LP. In addition, the medium is that much older; so it has been exposed to dust and other vagaries of fortune over a much longer period of time. For example, if the disc was played with steel needles - and most were - the abrasive action of the slate upon the steel means that minute particles of steel remain embedded in the groove walls; these are amongst the more recondite foreign-objects which add to the surface noise.

The Stereo Lab click-pop removal filter was originally developed for processing LPs and it needs to be modified to deal with the crackles from 78 records. That is the function of the check-box to the right of click-pop filter in the Phono EQ preferences dialogue. When this is selected, the click-pop removal filter is adjusted to deal with the crackles of 78 records. Note than, when dealing with all 78 RPM records, the phonograph regime setting in the Phono EQ preferences dialogue should always be set to MONO.

The Stereo Lab crackle remover is automatically adaptive and is very effective. It is a unique process developed by Pspatial Audio and, unlike so many other, competitive techniques, the crackle-reduction has no effect on the high-frequencies so that the sound is not "dulled" when it is cleaned-up. The proof of this lies in the spectrum of a pre- and post-processed track of a 10kHz test-tone from disc DECCA K.1804; illustrated below (the results of the post-processed file are superimposed over the results of the original).


Spectrum of pre- and post-processed 10kHz test-signal from DECCA test disc

Note how the level of the 10kHz is totally unaffected but the wideband noise floor is reduced some 10dB. (At 10kHz, the signal-to-noise is about 40dB improving to about 60dB in the midband.)

Like all audio processes, the proof is in the listening. The following audio file demonstrates the difference between equalised needle-drops with and without Stereo Lab crackle removal.

Click here to hear a demo of Stereo Lab crackle-removal algorithm. In each case, the original is played first and the treated ("de-crackled") version follows. The musical examples (of various genres: chamber, orchestral and "popular") are:

  1. Tchaikovsky. Andante Cantabile. Fritz Kreisler (violin). 78 RPM record. HMV (D.B.3443)
  2. Chopin arr. Les Sylphides Ballet. LPO, Malcolm Sargent. 78 RPM record. HMV - The Gramophone Company Dum Dum (C. 7465)
  3. Night and Day (C. Porter) Vocal refrain: Fred Astaire. 78 RPM record. HMV (B.D. 5761)
  4. 10kHz test tone. DECCA Test disc. 78 RPM record. (K.1803)
  5. Blue Rain (Mercer - Van Heusen). Glenn Miller Orchestra (Vocal Ray Eberle). 78 RPM record. HMV (B.D 5927)

These examples were carefully chosen to be representative of 78s in "normal condition" from private, domestic collections. In every case, the equalisation applied is identical for both pre- and post de-crackle examples.

As is the case with LP pop-click removal, the Stereo Lab algorithm doesn't interpolate or predict the audio. Instead, a robust click-detection algorithm is coupled to an adaptive attenuator which lowers the intensity of clicks rather than attempting to interpolate valid audio. In this way, audio is never lost or synthesised.

Needle-scratch or surface-noise removal (GRAδIENT)

The spectrum of the noise from a 78 RPM record rises from about 500Hz to very high frequencies: in fact, to frequencies way beyond the recording technology of the nineteen-thiries and nineteen-forties, which was limited to about 7kHz after which the recording characteristic was said to be constant acceleration (probably a polite way of saying that the response of all the recording gear was in free-fall after this frequency!) Investigations reveal that very little, if any, information exists above this frequency (at least with older 78s), so nothing musical is lost by filtering them out. The surface noise however is drastically reduced.

The results of cascaded crackle-removal and simple low-pass surface-noise removal are demonstrated in this audio file. The original is played first (with the correct EQ), then the treated version.

Even better results are possible with an algorithm which is able intelligently sort wanted high-frequencies from the noise signal and this is the basis of the advanced noise-reduction introduced in Stereo Lab known as GRAδIENT, which stands for GRAmophone DIsc Enhanced Noise Treatment.


Turntables for playing 78 records

As interest in analogue records revives, there are more choices for turntables which play 78 RPM records. The Austrian company Pro-Ject Audio Systems, who have done so much to furnish equipment for the analogue revival, even make 3-speed turntables "out of the box".

For the real enthusiast, a modified Technics SL1200 is a very good solution. The Technics is a high quality, Direct Drive turntable which ensures both the ability to vary the speed around the 78 RPM and maintain rotational speed with great accuracy. Various aftermarket speed modifications are available for this turntable, some of which permit sufficient variation around the standard 78 RPM speed to replay almost any gramophone disc of any era.



References and Notes

1 Beardsley R. (1999) Speeds and pitching of 78rpm gramophone records. Classic Record Collector magazine, Winter edition.

2 The whole issue of using pitch to determine the correct speed of a gramophone record is a bit fraught. Not only were gramophone speeds not standardised pre 1948, but pitch wasn't either! International standard pitch wasn't defined until 1955 (and as ISO 16 in 1975), and whilst the mighty American entertainment industry had standardised on tuning the A above middle-C (A4) to 440Hz from as early as 1926, other countries stuck to their own, much older, standards. For example, much of Europe tuned A4 to 435Hz.

In summary, Beardsley (op. cit.) informs us that a pitch variation of A4 from 425Hz to 445Hz is the range a 78 RPM record collector should expect. Given that this adds a further 5% uncertainty on top of the 13% speed uncertainty and thus a total of ±19% variation, the situation seems bewildering.

Fortunately, as he explains, the ear is a discerning instrument and the general consensus is that gramophone speeds should be set so that the performance - and especially voices - "sound right".

3 Peter Copeland P. (2008) Manual of Analogue Sound Restoration Techniques. The British Library London; available on line from the British Library site


Links

Home page

For all support issues, go here.

For Pspatial Audio sales, email: sales@pspatialaudio.com



© Pspatial Audio 2015 - 2017. All rights reserved. Apple Certified Developer. Stereo Lab, Aria 51, Aria 20, Head Space, Groove Sleuth, iLOOP and FRANCINSTIEN T-Sym are trademarks of Pspatial Audio. FRANCINSTIEN and Bride of FRANCINSTIEN (BoF) are trademarks of Phaedrus Audio. Macintosh and the Mac logo are trademarks of Apple Computer, Inc.

The image of the 78 record at the top of the page is supplied under a Creative Commons Attribution-Share Alike 3.0 Unported license: the picture's author is Mediatus. All the recorded examples in the de-crackle demonstration were first published over 70 years ago and are therefore no longer in copyright.