Making transfers from 78rpm sources - the CHARM Transfer Engineer's approach
CHARM is fortunate to have a collection of 78rpm discs to hand. The King’s Sound Archive, consisting of around 150,000 discs, encompasses the breadth of recorded music available to the British public for half a century or more. Once everyday items, 78rpm discs are now hard to find, and require some expertise and a lot of specialised equipment to replay. Only a very small percentage has ever been reissued on CD, and many of those have been significantly adapted to suit modern taste. In choosing recordings to make available through CHARM we have focused first of all on the CHARM research projects, which is why there are a lot of Schubert song performances here and most of the NGS label, as well a wealth of recordings from the early electrical era. Thereafter we have concentrated on performers and recordings that have not been well covered in commercial reissues, and especially on the HMV C series (one of the so-called ‘plum labels’), which aimed to cover lesser-known performers and to satisfy a home-grown audience. We hope in this way to offer a more representative experience of what most record listeners enjoyed in the 1920s, 30s and 40s than one can currently get from CD reissues. Making a good transfer of a 78rpm recording is not an automated, mechanical process. Interrelated practical difficulties and historical and technical considerations need to be disentangled before the audio content is revealed at its best. My task as CHARM’s transfer engineer is to manage the practicalities so that the original audio content is reproduced with as much fidelity as possible in a way that is accessible to those without specialist skills or expertise.
Choosing the best copy
To describe the entire transfer process one has to begin in the archive where the discs are stored. As King’s is fortunate to have several copies of some discs it is important to choose the best copy. By best, I mean undamaged and least worn. Some of the discs are chipped and cracked, scuffed and scratched and, in the worst cases, broken. The choice in such instances is usually unambiguous although it is less easy to discern which discs are badly worn. In general this afflicts the most heavily modulated discs because loud music, in causing the stylus to move eccentrically, usually manifests the worst erosion. Nevertheless, any disc that has been played often will be worn. To compound this difficulty, signs of wear and heavily modulated grooves do look rather similar in artificial light. Sometimes, the least worn disc is damaged in another way; chipped at the edge for example. In such cases I have to decide whether to choose the un-broken disc and tolerate the surface noise, or to use the better copy and edit-in the beginning from the unbroken disc.
Cleaning the disc
Having made the choice the next stage is to clean the disc. After years of sitting on shelves in open-ended paper sleeves, the discs are certainly dusty. More particularly though, the mixture of mineral powders bonded together (originally containing shellac resin), from which the disc’s surface is made, leaves a sticky residue in the groove which clogs the stylus if it is not first cleaned away. We have a machine made by Keith Monks specifically for this purpose that washes and brushes the discs with double-distilled water before sucking away the dirt and residual water with a vacuum pump.
Stylus selection and tone-arm set up
The record label and year of manufacture give a good indication of the groove dimensions and therefore the most appropriate stylus. Generally speaking the earlier the recording, the broader the grooves: the broader the groove, the larger the stylus will need to be. Even so, the only real way of determining which is the most suitable is by listening and experimentation. For example, I’ve just completed a transfer of a complete symphony spanning five discs for which I judged a .0025 elliptical stylus to give the best results. One disc however, sounded distorted in places, probably as a result of wear. By changing to a slightly larger stylus (.0028) for this side only, I was able to lessen the distortion. If I had decided that because of this I should make all the transfers with the larger stylus there would have been a small yet significant reduction in the high-frequency response. We have seven styli from which to choose and one additional microgroove LP stylus (Shure N44s retipped by the Expert Stylus Company).
Other factors need to be determined at the same time and in conjunction with choosing the stylus: the weight of the tone arm, which determines how heavily the stylus rides in the groove, and the necessary equalisation – which I’ll come to in a moment. As both of these factors depend to some extent upon the choice of stylus, all three are interdependent. Having said this, it’s usually possible to hear which stylus renders the best reproduction before equalization, but as the un-equalised bass is rather weak, and the treble very prominent, one cannot always be certain to make the best choice at the first attempt. Besides, the best sonic reproduction is often disguised beneath a correspondingly excellent reproduction of the surface noise.
The most appropriate weight to apply to the cartridge is the least necessary. In this way the stylus will be free to respond to the movements dictated by the groove wall. But, as always, there are compromises to contend with. Warped discs for example, necessitate the application of greater weight simply to prevent the stylus skipping out of the groove. The trade off here is sometimes manifest by a duller, less responsive sound and an increase in resonance.
The turntable we use is an EMT 948, with a speed control unit made for us by Roger Beardsley who explains how replay speed and pitch are fundamental to successful transfer in the resources section of the CHARM website.
At this stage it’s already possible to make a ‘flat’ transfer (no EQ, filtering, or noise reduction) for archiving. This I do as a matter of course so any additional processing beyond this point is, to this extent, un-doable. But as my remit is to make the audio most accessible and useful to musicians and musicologists I need to make some effort to minimise the noise that’s the perennial accompaniment to music on 78s.
Re-mastering engineers, whose job it is to prepare recordings for commercial release on CD, have the rather difficult and sometimes thankless task of matching what are often very different sounding originals in order to render them aesthetically appealing to the modern ear. Compilations are particularly difficult in this regard because it’s assumed that listeners will be disturbed by sudden qualitative changes between tracks. What’s more, the sound changes quite considerably even during a single side of a 78 owing to the gradual diminution in quality, and a proportional increase in surface noise, as the linear speed of the stylus slows as it approaches the centre of the disc. (Other factors such as wax whine, and inconsistencies in the speed of the cutting lathe compound this problem.) The transfer engineer who is expected to join the sides seamlessly is cast into a hopeless quandary from which it is impossible to escape unscathed. My task is incomparably easier. For me each side is a separate entity that will end up as a single audio file, and whilst I attempt to maintain contiguity between sides - at least with regard to speed, styli and equalisation – I am not impelled to make any effort to match the sonic characteristics of surface noise and tonal balance for aesthetic purposes. Indeed, many of these characteristics are of great interest and usefulness to certain areas of study.
As I said at the beginning, a number of interrelated practical difficulties need to be understood to achieve good transfers. Many of these difficulties arise from the physical and mechanical properties of the recording medium and the problems faced by the cutting engineers at the time of recording. One of the handicaps of acoustic recording technology was the obscurity of low frequency sound: the acoustic horn was incapable of capturing long wavelengths with sufficient intensity for them to be registered by the recording device. The invention of electrical recording in the mid-20s resolved this problem but in doing so created another for the enterprising engineer. The practice of engraving electrically amplified sound waves of constantly differing intensity and pitch into the narrow walls of a closely wound helical groove is really incompatible with the physical laws of kinetic energy. This is because amplitude (or intensity) is linked to frequency (or what we perceive as pitch) in the following way.
The distance traveled by a particle in one cycle (wavelength) is four times the amplitude. From this we can show that the distance traveled per second is 4 times the amplitude multiplied by the number of cycles per second (or Hertz), which is therefore also the average velocity of the particle. From this we can see that amplitude is inversely proportional to frequency. Given that this is the case, the problem for the cutting engineer is that when using a constant velocity cutter, the amplitude doubles each time the frequency is halved. In other words; the lower the frequency, the higher the amplitude.
The conundrum then, is that a recording with sufficient amplitude to reproduce the high frequencies adequately well would render the low frequencies at such absurdly high amplitude that the cutter would deviate wildly from its helical path slewing through the groove wall into the adjacent groove. Indeed, it’s quite common to encounter discs that have been over-modulated to such an extent that the groove walls are so thin that the reproducing stylus is affected by the adjacent groove, resulting in pre-, or post-echoes. To alleviate this problem recording engineers were obliged to employ equalisation to reduce the low frequency amplitude. This explains why a flat transfer of a 78rpm recording sounds rather light in the bass.
Clearly then, it’s important to be able to reverse the equalisation process upon replay in order to reconstitute the original amplitude of the sound wave. Notwithstanding the fact that we have documentation showing the equalisation that each recording company claims to have used, it’s imperative for a number of reasons to evaluate each individual recording aurally, not least because cutting engineers were without doubt more adaptive than company specifications might suggest.
Equalisation (or eq) always affects a range of frequencies so it is usual and useful to think of it as a curve, because this is how it appears when expressed visually on a graph. Most people who have experience of audio recording terminology will be familiar with the terms usually associated with equalisation: Hertz (Hz) describing frequency, decibels (dB) amplitude, and Q bandwidth. However, eq curves relating to disc cutting are more often expressed as time constants (μ) for the reasons I have described. Ted Kendall’s Front End pre-amplifier that we use has only 3 eq controls (calibrated in microseconds) covering the audible frequency spectrum. These are specifically designed and craftily calibrated to maintain the same shape of curve regardless of where the curve is deployed in the audio spectrum. Faced with a conventional equaliser it would be all but impossible to realise the same curves because of the interrelatedness of amplitude and frequency and the consequent interdependence of each of the controls.
The final stage of the transfer process involves passing the signal through two devices specifically designed to eliminate some of the most intrusive noise. Cedar Audio’s ‘decrackler’ and ‘declicker’ do exactly what their names suggest: they go some way to removing certain types of rather specific, un-musical noises and, unlike some other audio noise-removal processors, they do so without adding unwanted artifacts and without colouring the sound. Although they have very simple threshold controls it’s important to set them up carefully to tackle the individual characteristics of each transfer.
Although these processors perhaps make the greatest contribution to enhancing the listening experience of early recordings they unfortunately act upon the recorded sound in a way which is irreversible, no matter how carefully the settings are noted down for future users to reverse. (The clicks and crackles are removed by digital interpolation which cannot be undone). For this reason the procedure is not universally popular amongst archivists, and we don’t use it in our own archive copies; but as the equipment is far too expensive for each and every individual to use prior to listening it plays an important role in the preparation of early audio recordings for dissemination.
I generally make no attempt to remove any other type of surface noise from the recordings because such noise is invariably broad band and impossible to tackle without adversely affecting the audio content.
Encoding and mastering
The transfers are recorded in high-resolution audio onto a digital audio workstation in order to facilitate mastering (Lynx L22 digital sound card, Sequoia digital audio workstation, WAVES plugins). This process usually involves high and low pass filtering at frequencies beyond the limits of the audio content. For example, 78s commonly include a significant amount of low-frequency noise. Rumble from the cutting lathe and electrical mains hum are common causes and although probably comparatively insignificant when the discs were originally released, once the surface noise has been cleaned away these other kinds of noise intrude much more on the listening experience, masking the audio content and clouding the potential transparency of sound. Moreover, many early recordings, particularly those made with an acoustic recording horn, exhibit unwanted resonance at quite specific frequencies. As these are quite distinctive and localized, it’s usually possible to ameliorate them to a certain extent using a digital equalizer.
The mastering stage concludes by adding fades to the ambience – or, to be more precise, surface noise - at the beginning and ends of each 78 side. Several different files are then created: 24bit stereo wav and 16bit mono wav from which is derived a lossless FLAC file for downloading. These are all stored on various networked hard drives, and DVDs, together with jpeg image files of the 78 labels. The transfer documentation (metadata) in Excel format and the workstation files are also included. Furthermore, the metadata is also encoded as xml files which allow for every aspect of the transfer process to be annotated in the smallest detail. At the end of CHARM’s work (by the end of March 2009) all this data will be accessible online linked to the sound files.
If you have questions or comments on this article or the procedures it describes please send them to Andrew Hallifax.