3. Understanding the sources: Performance and recordings

¶1 If we are going to use recordings as a means of studying performances we’re going to need to be reasonably confident, first of all, that they represent performances sufficiently accurately to do duty for them. As with any source of evidence, their production and original function need to be understood before we can understand what they transmit: how were these sources made and why? Above all, we need to know what a recording is of. A necessary step towards a musicology of recorded performance, then, is to find out. And that means learning rather a lot about the history of recording, its economic, social and technological history, all of which have a significant bearing on what comes off a record when we play it.

¶2 Fortunately, some good histories of recording now exist, 1 though there is room for much more detailed work on the technology and the way it was used in studio practice. 2 Here it is necessary only to outline the story, focussing on information that enables us to assess the relationship between playback and original sound in recordings made by various means at different times through the past 100 years. 3

3.1. How early recordings were made

¶3 The first machine for reproducing sound, Edison’s Phonograph, consisted of a sheet of tinfoil wrapped around a cylindrical drum. The drum was rotated, and at the same time moved along a metal rod, by means of a screw attached to a handle which the operator turned. As the cylinder moved along the rod, it passed across a metal stylus, attached to one side of a diaphragm. On the outer side of the diaphragm was a small mouthpiece into which the operator spoke. The sound waves focussed onto the diaphragm caused it to vibrate, which in turn caused the stylus to press into the tinfoil. As the drum rotated and moved across the stylus a groove was embossed in the tinfoil consisting of undulations related to the pressure patterns of the sound waves. Playback involved placing the stylus at the beginning of the groove made during recording, and winding the cylinder along the rod once again. The undulations in the tinfoil caused the stylus to move in and out, and so the diaphragm to vibrate, which in turn moved the air in the mouthpiece, recreating the sound.

¶4 As Peter Copeland has written:

The extraordinary thing about Thomas Edison’s 1877 invention for recording and playing back sound was its apparent simplicity. There is no reason why it couldn’t have been invented at almost any time in the previous century―it used no new scientific knowledge, no new materials, and no new technology. 4

¶5 But as I suggested in chapter 1, the difficulty must have lain not so much in the technology as in the imagination. The idea of recording sound was so alien that it’s not surprising that it only occurred to Edison as a possibility after he had already worked for many years on telephony. From the electrical transmission of sound to the recording of it is an imaginable step; from speaking to recording is not. Edison’s presentation of the phonograph in his first patent as a telephonic repeater device shows very clearly the imaginative route he took.

¶6 Nor does anyone seem to have been able, in the early years, to imagine an industry of music recording. Edison lost his initial enthusiasm quite quickly, and turned his attention to electric light, so that, as Roland Gelatt has said, ‘Mankind gained the lightbulb, but posterity lost Jenny Lind and Franz Liszt.’ 5 It was only the appearance of competition that, ten years later, brought Edison’s attention back to sound recording. Only then were phonographs set up to record music on any scale, and then only as slot-machines for use in restaurants and bars. 6 The main serious application—from the viewpoint of the first manufacturers—was as a dictation machine, and this is how the early commercial phonograph was marketed, often with a typewriter at its side. Nevertheless, it was the success of the slot-machines, more than anything, that suggested that commercial music recording was viable. The main obstacle thereafter was the technical difficulty of manufacturing reliable copies of recordings, and this is the principal reason why it was another decade after Edison’s ‘Improved Phonograph’ before records and players were being sold to the general public in any quantity. Only in the very last years of the nineteenth century did classical music begin to feature in record company catalogues. Consequently, with only a very few exceptions, surviving recordings of classical music date from no earlier than 1897.

¶7 By then two competing technologies were available. Developing Edison’s invention and improving its weakest point—the tinfoil, which playback soon destroyed—Alexander Graham Bell, inventor of the telephone, and Charles Tainter showed how a hard wax cylinder could be used in place of the foil sheet; cylinders produced less background noise and lasted longer.

¶8 Simpler to play and much easier to reproduce were flat discs, developed by Emil Berliner as a way of avoiding Edison’s patent on cylinders. In Berliner’s ‘gramophone’, developed between 1887 and 1893, the process was as different from Edison’s as possible. The stylus moved across the recording medium, now a disc rather than a cylinder, and recorded on it by causing the vibrating diaphragm to cut a groove which moved in the lateral plane rather than the vertical. A track was made in a thin coating of grease that covered a metal (zinc) disc. When the recording was finished, the disc was place in an acid bath. The acid etched a groove in the metal where the recording stylus has exposed it. This resulted in a ‘positive’ recording that could in fact be played. However, early on Berliner realised that the individual recordings necessary with the wax cylinder process were a bar to mass exploitation. He found that copies were easily made by firstly electroplating the original disc to provide a negative version with ridges instead of grooves. This metal negative became a stamper for producing identical copies in a steam operated press. 7

¶9 Although gramophones and pre-recorded discs were available from 1894 it was another four years or so before enough recorded music was available for the gramophone to take off commercially. Berliner soon improved the reproduction process by abandoning etching and (from 1900) using a wax master capable of spawning multiple stampers from which many more discs could be produced for sale.

¶10 Crucial to the development of the catalogue were a series of tours, first in Europe, but soon extending to Russia, India and the Far East, by Berliner’s agents, brothers Will and Fred Gaisberg (by now working for the Gramophone & Typewriter Co.), who from 1902 onwards were bringing back wax masters, made often in hotel rooms, of opera arias sung by Caruso, Tamagno, Battistini, Patti, Chaliapin and others, resulting in discs that sold in relatively large quantities, as well as a very large collection of world music, all subsidised (then as now) by more popular discs. 8 By 1904 the gramophone had arrived, and when Patti made her first records in 1905 it was national news.

¶11 By our standards, the technical limitations of both technologies look crippling, but one has to remember just how remarkable recorded sound seemed to people to whom the idea was so new. A report in The Times on 5 October 1909 describes ‘A Gramophone Concert’:

Yesterday afternoon, in a room at the Savoy Hotel, the temerarious gramophone challenged comparison with the living voices of two famous singers. Mme. Clara Butt and Mr. Kennerley Rumford appeared before an invited audience to sing certain pieces and then to listen to the gramophone’s versions of the singing of them, taken, not on the spot, but on some previous occasion. Mme. Clara Butt sang Mr. Landon Ronald’s “Believe me if all those endearing young charms” to the composer’s accompaniment at the piano; then Mr. Ronald remained at the piano and, while the gramophone gave its reproduction of Mme. Butt’s singing of the song, accompanied it once more with the soft pedal down. Mr. Kennerley Rumford sang Maud Valérie White’s “King Charles,” so familiar in the voice of Mr. Plunket Greene, and the gramophone repeated it; and the singers joined in two duets, Mme. Lehmann’s “Snowdrops” and the “Night Hymn at Sea.” It must be admitted that the gramophone came very well through its ordeal. The instrument has improved so much of late that we have almost forgotten the days when it began every piece with an imitation of cats fighting. Had we not just listened to the singers viva voce we should certainly have declared the gramophone’s a wonderful imitation. Wonderful, indeed, it is—and it seemed more wonderful than ever under the conditions of yesterday’s performance—in the mere fact of its singing at all and reproducing so unmistakably the voices we had just heard. But with Mme. Clara Butt’s tones still ringing in our ears it was impossible not to notice that it had introduced a faint nasal tinge into her lower notes and tightened some of her higher notes so as to give the impression that her throat was not fully open. The volume of sound it could hardly hope to equal; but something remains yet to be done in the direction of quality. An odd little weakness on the instrument’s part is its inability to pronounce sibilants—it lisps. Otherwise the words were extraordinarily clear. Some of the other selections from the répertoires of Mme. Butt and Mr. Rumford which formed the rest of the programme showed a very unsatisfactory reproduction of the orchestral accompaniments, which suggested a wheezy harmonium. But it remains wonderful that any mechanism should do what the gramophone does. 9

¶12 Quite how listeners could have been so impressed is hard for us to imagine (even with modern replay equipment which reveals details recorded but never heard at the time). But for them, far more striking than the surface noise and the poor frequency response (not even mentioned here) was the fact that the voice was recognisable at all. 10

Later acoustic recording

¶13 In order to understand the way early recordings sound, and what that means for the kinds of things we can learn from them, we need to look at the whole process of recording in the early days, and then look at an example in that light. Rather than try to cover the huge range of slightly different techniques used at different places and times, let’s imagine a recording made onto disc in about 1920. 11

¶14 The recording medium was a form of hard wax, prepared at the factory in vats, filtered to keep the mixture as smooth as possible and, after cooling and hardening, turned on a lathe in order to produce a smooth surface. The resulting wax tablets were packed in cases and sent out to the studios or, for recordings elsewhere, along with the technicians’ baggage. (Fred Gaisberg reports taking 600 on a year-long trip to India and the Far East in 1902-3, and bringing close on 600 recordings back. 12 ) For recording, the wax tablet was placed on the turntable of the recording machine, where it rotated, in theory (though by no means always in practice) at 78 revolutions per minute (rpm). The whole turntable assembly moved sideways beneath the cutting head so that a groove was cut in the wax from the outer edge of the wax disc towards the centre. Because it was attached directly to the horn, the cutting assembly was fixed. The stylus was moved by a (usually glass) diaphragm, and, depending on how efficient the connections were, pressure was transmitted more or less accurately. Sound was transmitted to the other side of the diaphragm through tubular connections leading from the narrow end of the recording horn, while the horn itself captured and focussed as much of the performers’ sound output as possible. Horns tended to have resonances of their own, which were damped as far as possible by wrapping tape around them. Multiple horns could be used to capture sound from a larger group of performers or from different parts of a piano, for example, and these were connected up via Y-shaped metal connectors joined to the horn with rubber tubing. The horns were suspended from (or occasionally supported on) stands to minimise strain on the cutting mechanism.

  Plate 1: Edward Elgar and
                    musicians posing at a 1914 recording
Plate 1: Edward Elgar and musicians posing at a 1914 recording session

¶15 If we look at a picture of a recording session some of this will become clearer (Plates 1 & 2).

  Plate 2: An acoustic
                    recording session for the Victor
Plate 2: An acoustic recording session for the Victor label

¶16 The recording machinery, as in almost all the surviving photographs, is out of sight. Partly this was to cut out the noise, though given the insensitivity of the horn and all the noise generated along the way, it is unlikely to have been audible on disc; partly it was to protect company secrets. What we can see is the horn, suspended by a wire and wrapped with tape; and we can see the very unorthodox arrangement of the musicians in front of it. Although one might suspect that the crowding of the musicians is for the photographer’s convenience, in fact there is ample period evidence that this was normal and necessary in order to get the best possible balance of sound focussed into the horn. Stanley Chapple, in a 1928 issue of Gramophone, provides a floor plan of such a session. 13

  Plate 3: A two-horn
                    acoustic recording layout
Plate 3: A two-horn acoustic recording layout (1928)

¶17 Two recording horns are used, with the violins (which recorded least well) nearest to them. Squashed around them are the woodwind players, who would have been reinforcing the string parts. Behind them, but higher, were most of the brass, the French horns facing backwards in order to direct the sound from their bells into the recording horn, the players following the conductor in a mirror. The conductor is pushed out of the way to the side where he can be seen but doesn’t obstruct the sound. Bassoons reinforce the cellos, and a tuba and contrabassoon replace double basses, which would not have recorded adequately. 14

  Plate 4: An acoustic
                    recording session for voice and
Plate 4: An acoustic recording session for voice and piano

¶18 Plate 4 shows a typical studio layout for a recording of voice with piano accompaniment. The piano is an upright because its sound was more effectively focussed into one horn. The horn is hung right in front of the singer’s mouth, and the piano is set above and behind the singer at a height that ensures that the maximum amount of piano sound enters the horn. Pianists were instructed to play fortissimo throughout. 15 Singers, on the other hand, had to move towards the horn for quieter passages, and away for louder notes to avoid distortion. Inexperienced soloists were guided back and forth by an assistant (sometimes on trolleys). 16 One other thing noticeable from many photographs is the appearance of the men in shirtsleeves. Because the wax has to be soft enough to cut cleanly recording rooms had to be uncomfortably warm. It seems safe to assume that some of the appalling intonation one often hears on early recordings of strings and woodwind was caused by the heat.

¶19 Recording proceeded in takes as long as one wax blank took to fill: for early cylinders and discs this was about two minutes; for later cylinders three and then (from 1908) four minutes; for a ten-inch disc a little over three minutes; for a twelve-inch three at first, and later up to four and a half. 17 Pieces that played for longer than this had either to be played quickly (for which there is some evidence, 18 though not as much as is sometimes suggested) or—and this was much more common—had to be cut. 19 If a piece or movement were being recorded that lasted much longer than one side, recording stopped at a musically convenient point in the score, and then continued later when a new wax tablet was in place and everyone was ready to resume. Sometimes side-ends were composed by an arranger, for example adding a perfect cadence to what would otherwise have been an open-ended musical phrase; mostly the musicians closed a side with a modest ritardando. 20 (The consequences of this for performance analysis will become clearer later.)

¶20 After the recording session was over the wax masters were returned to the factory for electroplating. 21 This produced a negative metal copy which was used to stamp a test pressing. If the musical results and sound quality were considered satisfactory, further negatives were made and nickel-plated for use as stampers. (An alternative method was to use the negative to make a positive from which harder negative stampers could be made, and there were other approaches, but the principle was the same.) In some cases (mainly reissues of very early discs) this metal master had the shape of its groove adjusted in the interests of better long-term results from the shellac copies, and this process must to some extent have altered the sound. 22 The negative metal stampers were then used to stamp copies in shellac, either pressed against a blank for single-sided discs, or against another stamper for double-sided. Once the metal negative was made the original wax master was returned to the lathe to be planed smooth and reused. The shavings, along with tablets now too thin to reuse, went back to the vats for recycling. Over time the stampers became worn, with consequent deterioration in the sound of the shellac discs. Eventually quality became sufficiently compromised for the metal stamper to be abandoned. Provided more stampers could be made from the metal master this wasn’t a problem; but if that was no longer possible then two options remained; either an existing disc had to be re-recorded onto a new wax master 23 —which meant all the original noise was recorded too, and added to the noise produced through the new recording—or the artists had to be assembled again and a new performance made.

Misrepresentations in early recordings

¶21 Reading through this description, it’s not hard to see that some quite serious compromises are made in order to produce sounds the machine is capable of recording. Both what went in and what came out were far from representative of ideal music-making at the time. Moreover, there were very significant distortions in the recording and playback of the sound, as we shall see in a moment. Clearly, if we are to use these recordings for study it is essential that we understand what these compromises and distortions are, how they arise, how they affect what we hear, and how we might be able to compensate for them.

¶22 Let’s begin with disc speed. Readers old enough to remember LPs or 78s, or anyone familiar with DJ techniques, will know that if the speed of a disc changes the pitch changes with it: a rise in speed brings a rise in pitch. Consequently if a record is played at anything other than the speed at which the master was recorded the pitch of the music will be wrong. Ways of establishing exactly what pitch the musicians were using, and of ensuring that the final discs played at the correct speed were, in theory, available but would have been impractical in most recording situations. In a very warm studio it was inevitable that pitch would rise during a session. The machinery driving the cutting turntable would have been hard to regulate, was subject to variations during manufacture, and could change speed in different atmospheric conditions and over time. Even the cooling and hardening in the wax during recording, slowing the progress of the cutting head, could make a difference. Consequently there can be not only problems in establishing the correct speed at which to play a disc, but there can also be changes of pitch during playback. Gramophones were equipped with devices for varying the playback speed. Few were precise, or even precisely marked. ‘Fast’ and ‘Slow’ were generally stamped on a guide, and one moved a lever in one direction or the other. Although most discs play somewhere around 76-80rpm, speeds can on occasion be as low as 60 or as high as 90rpm. To put it into perspective, at 78rpm a difference of c. 4.5rpm produces a difference of a semitone in pitch, so the differences may not be negligible. (Figure 1)

Figure 1

¶23 A singer speeded up by a semitone can sound very significantly different, as a comparison of modern reissues of Adelina Patti shows. Patti was one of the most famous singers of the 19th century, and yet, as we’ll see in chapter 4, what you think she sounded like depends on the speed at which you play her recordings. And about that, modern CD transfers have widely differing views. There is no scientific solution to this problem. All one can do is hear as many recordings of a performer as possible and gradually form an idea of the sound she makes, then test any one disc against that impression, adjusting the speed until she sounds ‘right’. Of course one’s impression may be based on so many mistaken speeds that one’s quite wrong. All we can do is discuss the results among experienced listeners and hope to agree, more or less. Similar problems occur in more modern recordings because of differences among tape machines, cutting lathes and playback equipment. To make things worse, pitch was not necessarily fixed by the musicians at A=440Hz, and singers frequently had pieces transposed to suit their voices: neither fact was noted on the disc. Nothing but musical instinct can guide us here. 24

¶24 Next let’s look at general considerations of sound quality. Cylinders were capable of surprisingly good sound quality, which may be partly why Edison stuck with the technology for so many years after every other manufacturer had gone over to discs, until 1929 in fact. 25 Both because the stylus moved up and down in the groove instead of side to side, so not interfering with adjacent grooves, and because the stylus moved at a constant speed (on discs it moves much faster nearer the outer edge of the disc than near the centre), cylinders at their best could provide more consistent results. But they were less successful commercially: they were harder to duplicate, so cost more; they took up much more storage space; they required a more complex playback mechanism; and, although it may seem trivial, they lacked the attractive and convenient centre label on which so much useful information about the disc’s contents could be printed. 26 So the technology that succeeded in the marketplace was in some ways not the best available (not for the last time). 27

¶25 On early cylinders and discs much the most successful instrument for recording (after the brass band) was the voice, and especially the coloratura soprano and tenor. The reason for this is the limited frequency response, which was confined at first between about 150 and 2000Hz, enough to give a good representation of the frequencies constituting vocal colour, but not enough to give a good approximation of most instruments. It helped that high voices could also generate a lot of energy. Closest in effectiveness were wind instruments, with strings far behind. Consequently, most of the earliest recordings are of singers, followed by military bands. 28 The results we have already seen in the layout of a typical session ensemble. Pieces had to be rescored, replacing lower strings with wind—hence the ever-present tuba pumping the bass in opera arias. In addition, violins, when used at all, were routinely replaced (from about 1902) by Stroh violins, a device invented by Augustus Stroh that played like a violin but replaced the body with a diaphragm attached to a horn, which amplified the upper frequencies. There are several in Plate 2. As we have seen, pianos had to be played fortissimo, and right into the singer’s ears; and we have seen how in ensemble recordings all the musicians were playing loudly and right on top of one another. Singers were required to exaggerate vowels, roll their Rs, sing S as Sh, 29 and had to maintain the most even tone possible, at the same time as moving towards and away from the horn as required. In these sorts of conditions, anything like a normal musical performance must have been extremely hard to produce. All this we need to bear in mind when listening to the results; it may have important consequences for the kinds of things we can safely say about details of performances.

¶26 Sound File 4 (wav file) is a transfer without noise reduction from a 1921 recording of Frieda Hempel singing Schubert’s ‘Ave Maria’ (Ellens Gesang III). 30 The disc has been transferred at 78rpm because it gives a plausible key (A major) for her voice as we know it from recordings (always bearing in mind the circularity of the argument): Schubert’s notation is in B-flat, but this recording sounds less satisfactory there. Schubert’s piano accompaniment has been rescored for harp, harmonium, violin, and cello, not in this case because piano would have recorded less well but simply because in 1921 it had long been customary to see this song as extremely sentimental, and this ensemble offers the opportunity to squeeze every last ounce of saccharin from the score. The violin part is a study in unrestrained portamento, the harmonium piety in sound. With singer and all the instruments crowded round the horn the balance is curious, the violin a little distant (perhaps fortunately) and the harmonium a little closer than one might wish. Most striking of all for a modern listener is the very high level of surface noise, much higher than we’ve come to expect listening to CD transfers which are invariably treated with digital noise reduction. This noise had several sources including problems with the recording machinery, the nature of the materials of which the disc is made, imperfections in and damage to the disc surface, and the misfit of stylus to groove. 31 Sound File 5 (wav file) presents a much earlier recording of the same song from 1905, 32 with Albert Piccaluga accompanied by a 2-manual Mustel harmonium with a celesta stop for the right hand, producing a sound somewhere between piano and harp. One might otherwise have been tempted, given the poor sound reproduction at this date, to assume it a piano, which just goes to show how careful one has to be. (The identification in this case comes from the original record label.) 33 Karsten Lehl, who has kindly provided the transfer of a disc in his collection, has speeded up the turntable by as much as 7.3% in order to pitch the performance in A and to give the voice what he feels is the right timbre (Piccaluga was a baritone). Where Hempel had a full-length 12-inch disc at her disposal, and used all four and a half minutes of it, Piccaluga has a 27cm (10¾ inch) disc and has to make do with two stanzas. The German poem set by Schubert has been replaced with the Latin Roman Catholic ‘Ave Maria’ to make the piece fully religious. The surface noise is of a slightly different type here, with a swish that’s very characteristic of early acoustic recordings (and very hard to remove today). Sound File 5a (wav file) offers the same transfer, this time with digital noise reduction. Because the musical signal is so limited in frequency range at this early stage of recording a lot of noise can be removed without damaging the music. Even so, the result is still fairly grim. All in all, there is little about these recorded performances that we would recognise today as ‘correct’ or acceptable, either in the arrangements, the performances, or the recordings. Listening to them with understanding, not just amusement, requires us to know a surprising amount about how they were made.

¶27 In fact there is much that could be said about these performances, and in the light of later chapters readers may like to come back to them and work on them for themselves. But first we must focus on the shortcomings of the manufacturing process, since without understanding that we have little hope of commenting usefully on the music it obscures. 34 Imperfections in the composition of the wax tablets could produce unwanted resonances. Any irregularities in the progress of the cutting stylus through the wax produced undulations in the wax, in addition to those made by the sounds being recorded, and these would generate unwanted noise at playback. The problem was exacerbated by a worn stylus or a fractionally incorrect cutting angle. The various components of the recording machine could combine to emphasise certain frequency bands (and we’ve already mentioned horn resonances). Irregularities could develop during the electroplating of the wax master, and surface imperfections during the pressing and cooling of the shellac copies. Moreover the shellac—a compound of various fillings bound together by a lacquer made from ground Indian beetle—needed to consist of particles of varied sizes in order to ensure strength, and the roughness that resulted helped to wear the tip of the steel needle to the shape of the groove, ensuring better response to the details of sound recorded along the groove walls; yet at the same time the roughness in the shellac compound produced further noise. 35 Finally distortions of various kinds were inevitably generated by the playback mechanism through the connections of stylus, diaphragm and horn at the output stage. Only through understanding these various sources of surface noise and distortion (as George Brock-Nannestad has persuasively argued) can we begin to think about how to compensate for them.

¶28 The process and problems I’ve described so far apply especially (but not by any means only) to acoustic recording, those systems in use until 1925 where sound pressure generated by the performer is applied through mechanical connections to the stylus cutting the wax. The situation was transformed almost overnight, with remarkable benefits for musical recording, when electrical systems were introduced commercially in America in February 1925. Three inventions were essential, the microphone, the vacuum-tube (‘valve’) amplifier, and the electromagnetic cutter-head. For the rest it was a matter of redesigning the machinery and training recording engineers to use it. Electrical recording brought several major advantages: first, frequency response improved dramatically, immediately doubling the range that could be captured up to 5000Hz (which had almost doubled again by the end of the 78 era), and extending it down as low as 60Hz; secondly, mechanical resonances were largely eliminated, so the sound was coloured less by the equipment; thirdly, the electrical systems allowed studio control of many parameters of the sound, including balance; and last but perhaps most important of all, it suddenly became possible to record almost any kinds of sound in any environment. Replacing the horn with the microphone meant that performers no longer needed to shout or play unnaturally loudly; they could be spread out over a much wider area; and it was no longer necessary to re-orchestrate. 36 Recording could take place in concert halls, opera houses, cathedrals, or even in the open air. Not every problem went away, however. Discs were still cut in wax and stamped in shellac; pitches and speeds were still variable; there was still surface noise. 37 But nevertheless the increase in naturalness and detail is very striking, even to a modern listener. Sound File 6 (wav file) presents a 1935 recording, without noise reduction, of Schubert’s ‘Die junge Nonne’ sung by Germaine Martinelli in French (‘La jeune religieuse’) accompanied by an orchestra conducted by Eugène Bigot. 38 (Sound File 6a (wav file) is a version with digital noise reduction.) Two things will strike you immediately: the surface noise is considerably reduced—in fact this was not a by-product of electrical recording but simply the result of improved manufacture—and the orchestral sound is far more realistic. Unlike the ‘Ave Maria’ recordings, this time we can hear a full orchestra with something like a natural sound: even the timpani and double basses are clearly audible, and all the instruments are plausibly balanced. 39

3.2. Transferring and listening to early recordings

¶29 I think enough has been said about the making of early recordings, especially on 78rpm discs, to make it pretty obvious that the performances they carry need to be interpreted with great care. We can’t listen to them and hear the sounds an audience at a live performance would have heard then. What we may be able to do, knowing enough about what happened between performer and disc, is to begin to imagine the sorts of sounds that would have given rise to what comes off the discs, given that means of recording. It’s very hard to do. It would be considerably easier if we could listen to modern recordings made simultaneously on modern and on early equipment. Experiments have been done, but as yet none of the results has been published on disc. 40 It is probably, however, one of the most useful experiments that could be done at this stage in research, and an adequately resourced research program that brought together a range of original equipment with the few people in the world who have any idea of how to use it would be hugely valuable. Looking at the problem from the other end, it is not difficult, using digital editing, to introduce surface noise from shellac discs and speed fluctuation into modern recordings, and while it’s a poor substitute for properly set up recording sessions it’s quite informative. 41

¶30 Lacking a reliable measure, then, of the effect that early recording processes have on sounds, we have to fall back on our imaginations and try to listen through the noise and through all the other distortions of pitch and colour. While wholly unscientific, it’s surprisingly helpful. I can’t demonstrate it, so I can only seek to persuade you that with practice one does begin to find early recordings much easier to listen to; and it becomes simpler to understand how it was that so many millions of discs were sold, and records reviewed with such enthusiasm. Indeed, this was the experience of the original listeners: an article in The Times from 1922 commented on just this phenomenon.

But the very defects of the gramophone record have their compensation. They encourage close listening. If, for example, one has intensely enjoyed the perfect ensemble of the Flonzaley Quartet playing Schumann, the first feeling on hearing the gramophone reproduction may be one of disappointment. The magical tone has fled and the constant burr of the revolving disc is very present to the ears. It is extraordinary, however, the extent to which one can forget the burr as one concentrates the mind to catch the balance of phrase with phrase, the interweaving of the instruments, the details which distinguish line from colour, to use the phraseology which music has borrowed from another art. The gramophone record used in this way is, in fact, a close analogy to the faithful photograph, a thing which acquires its value from our knowledge of the original, and serves not only to recall but to deepen appreciation of the original. It seems at present that only a very small proportion of the great originals of music have been provided with these photographs; but their number is being increased daily, and old ones are being scrapped in favour of better ones as makers realize more insistently that musical people are looking for the best. 42

¶31 Listening on original equipment, then, is not, from an artistic point of view, such a crazy idea, nor necessarily (though it sometimes is) a social mannerism. A horn gramophone, if it has been expertly maintained, can be a very satisfactory reproducing instrument from a musical point of view; and it’s well worth trying to hear one. 43 Portables are still reasonably cheap and surprisingly good, not to say astonishingly loud. But there are excellent conservation reasons for not using valuable discs on them, since discs wear out far more quickly on original than on modern equipment, and to make matters worse few non-specialist users are inclined to insert a new steel needle for each side played, still less to make their own thorn needles which are a safer and musically preferable alternative. Very few people still use original equipment, then, and in most cases it’s probably just as well.


¶32 The vastly increased interest in early recordings which has characterised recent years comes, of course, from reissues on CD. There are good reasons why this has happened— early recordings preserve wonderful performances that, once one’s got used to their very different style, can give as much pleasure as (often more than) modern performances—and there are much less admirable reasons. With the market for new recordings of classical music in steep decline, it’s far easier for the record companies to make money if they simply reissue recordings they already own. In Europe (though this may be about to change), recordings are out of copyright 50 years after the date of issue (provided that the music performed is not still in copyright: the law there covers the composer or arranger until 70 years after their death). In the US copyright law is far less favourable, and copyright not only has a much longer duration, but can be extended indefinitely by making reissues. For companies operating in Europe, then, it became extremely cost-effective to transfer recordings issued originally on cylinders and 78rpm discs, especially of out-of-copyright classical music, and reissue it on CD; and many small companies have produced more or less impressive results alongside the big hitters like EMI exploiting their own recording past.

¶33 So most of the early performances we encounter today are available easily and cheaply on CD. And on a single CD one can get up to forty two-minute cylinders, or at least twenty single-sided twelve-inch discs, each of which might have cost as much as £1 a hundred years ago, about a third of a teacher’s weekly wage at the time, and around a thirtieth of a barrister’s, which gives us an idea of the market for the most expensive celebrity discs. 44 There’s not much doubt that we are fortunate to have so much available for so little. Even if you cost a double-sided 78 at 50p from the local charity shop, a bargain-priced CD reissue with its ease of use, an accompanying booklet, and all the work that’s gone into the transfer is a pretty good deal.

¶34 But it’s ‘all the work that’s gone into the transfer’ that’s the problem. Consumers of classical music, which is more or less how the record companies view us, expect realistic sound, steady speeds, a quiet background, and continuous music from start to finish. For transfer engineers, extracting performances from cylinders and 78s, that means a lot of intervention and editing: in other words, the sounds get changed. Let’s look at the process and try to identify the sorts of modifications that get made and consider their consequences for us when we study the performances. 45

¶35 It’s important to say right away that transfer engineers are mainly freelancers, enthusiasts for early recordings who have spent many years working with discs and cylinders, and have accumulated a huge amount of knowledge of the sources. Many have a professional recording background. Each one has a toolkit of procedures, developed through informal learning—gained from reading the professional journals, bulletin boards, books, and from talking to colleagues—and through experience, the latter closely tied to the particular collection of equipment they prefer to use. Each has their own musical preferences that lead them to aim towards particular kinds of sound. While rough transfers can be done in no longer than it takes to play the original, most work at this level is done with great care over many hours of meticulous trial and error. It’s a complex mixture of technique and taste. 46

Choosing the disc

¶36 The first step is to find the best possible copy of the original issue; that means the least worn or damaged. Sometimes, if the disc is rare or unique, there’s no choice; on other occasions there may be thousands of copies surviving, and in that case it’s a matter of drawing on reliable collections and selecting the best copy that can be found. A few companies have retained at least some of their metal masters, and in that case an in-house transfer engineer has access to the best possible material (EMI is the prime example). National sound archives and broadcasting organisations generally have very large collections, though few will allow users to make their own copies, so on the whole transfer engineers will avoid them unless there is no alternative. If they must, they will ask for a ‘flat transfer’, one with as little modification to the sound as possible: a set speed, no filtering or noise reduction. More commonly copies are borrowed from collectors—the world of early recording enthusiasts is generous and cooperative—or made in the few archives that allow users to handle originals. Some transfer engineers—though fewer than one might imagine—have their own collections: on the whole the collecting bug and the need to make a living from recordings are not economically compatible.


¶37 Next the disc (for simplicity I shall say disc from now on, but similar points apply to cylinders) is cleaned. Over 50 to 100 years a certain amount, often a great deal of muck accumulates in the grooves and gets smeared across the surface of a disc, and getting rid of it makes a huge difference to the quality of the sound, allowing the stylus to track the groove rather than the film of grime adhering to it. There is, needless to say, much debate about the best ways of cleaning 78s. Many professionals use a record cleaning machine, looking something like a record player, that uses a brush to spread cleaning solution over a disc spinning on a turntable and then an arm to suck it off, bringing the dirt with it. What solution should be used on different materials is a matter for intense debate. For 78s a solution of triple distilled water mixed with tiny quantities of washing-up liquid is recommended by many, and produces impressive results, but there are concerns among those who have examined the whole question scientifically about the longer-term damage that residues may do. This isn’t a matter we need solve here, but it is characteristic of the whole field that it remains an unresolved issue. 47 More serious cleaning may be required for discs growing mould or fungus.


¶38 When the only available disc is damaged repairs must be made. This may involve temporarily gluing together the pieces of a broken disc, in which case clicks made as the pickup stylus tracks across each join in every groove will have to be removed from the sound later on. (An expensive but interesting alternative is a laser player, which in theory can play broken discs. Results, however, are mixed and as yet they are not widely used.) 48 Other kinds of repairs include rebuilding damaged groove walls, which can be done under a microscope with suitable materials, and filling in chips in the edge of the disc. In the latter case music may well be lost, and it will need to be supplied from elsewhere—copied from another point in the disc if available, or from another performance if not.

Playing the disc

¶39 Nowadays most playing of early recordings is done on modern electrical equipment, either a suitable turntable, arm and cartridge; or the equivalent for cylinders, now made in much smaller quantities but to very high standards. 49 A few engineers and companies, however, use original equipment as a matter of principle, aiming to produce the sounds expected by the original engineers and listeners. Here the technique at its most purist involves nothing more than using a microphone to record the sound produced from a horn gramophone, though more complex setups are possible. 50 More typical is to use a modern motor-driven turntable with standard speed settings of 33⅓, 45 and 78rpm, and a wide-ranging variable speed control with digital readout so that exact speeds can be set or speed can be varied by exact amounts during the playing of a disc. As we’ve seen, the transfer engineer will have to decide on a musically plausible speed and pitch, and will have to ‘correct’ any changes in pitch during the playing of the disc, changes that occurred originally because of cooling wax, or winding down drive mechanisms, or otherwise imperfect equipment or operation. Speed fluctuation in LPs may be caused by accidents as simple as an operator brushing against a moving tape reel during the cutting of the master disc.


¶40 The choice of stylus is crucial, and makes most difference to the sound after cleaning. Because there was no standardisation of cutting styli or angles, and also because discs wear in different ways according to the kinds of needles used to play them in the past, grooves can be different shapes and sizes, and the least damaged part of a groove may be accessible to smaller (if it’s near the bottom) or larger styli. Transfer engineers therefore have a dozen or more different styli available: bi-radial, conical, and in various sizes. The skill with which they are chosen can have a very significant effect on the results. Sometimes different styli track different parts of the disc most effectively, and in that case the end result will be made from several transfers edited together. There are no rules about the choice of styli: the engineer simply chooses whichever sounds best to her. 51


¶41 From the stylus/cartridge assembly the electrical signal will usually go to an equalising pre-amplifier. 52 As with any equipment, the construction of the amplifier can colour the sound to some degree (valve (tube) equipment especially), but far more important is the nature of the equalisation. Equalisation involves increasing or decreasing the amplitude (loudness) of certain frequency bands, usually gradually over a range of frequencies, in order to compensate for reductions or increases introduced during the cutting of the master. Because bass frequencies made the cutting stylus cut a groove with more sideways movement, the stylus on playback has more difficulty tracking it, and fewer grooves (less music) could be fitted onto a disc. It was the practice, therefore, to reduce the volume level at lower frequencies when cutting, and for the playback system to restore the relative levels via suitable circuits in the amplifier. At the other end of the spectrum, high frequencies produced less signal strength compared to the amount of noise, and again the playback equipment had to be set up in such a way as to compensate.

¶42 In theory, acoustic recordings need no equalisation, since the original engineers had no means of altering the balance of frequencies during mastering. But as we have seen, the nature of the equipment, especially the horn, introduces changes to the balance of frequencies, and modern equalisation can be used to compensate for that. If so, it can only be done by ear as a matter of taste and experience, supported by knowledge of the original recording process.

¶43 For electrical recordings, however, the situation is quite different. Electrical amplification between microphone and cutting stylus offered engineers the opportunity to modify the recording characteristic, as I’ve described, and the advantages in disc length and sound quality were considerable. On the other hand, this left the manufacturers of playback equipment with the obligation to introduce equalisation in order to compensate. The main difficulty for them, and for us, is that no two companies applied the same characteristics, nor were the discs from one company necessarily consistent. Even two cutting machines could produce different results; and engineers could decide, either on the spot in the studio or as a matter of company policy, to modify the equalisation in order to produce the sound they wanted. Consequently, although a great deal of knowledge has been accumulated through experience, and attempts have been made to standardise and publish some of it, 53 choosing equalisation settings when transferring electrical recordings is as much an art, subject to taste, as choosing playback speeds or selecting styli.

Noise reduction

¶44 Once the signal has passed through the equalisation stage, the next consideration is whether and how to apply noise reduction in order to remove as much of the original surface noise as possible. There can be a very great deal of it, both the result of the recording and stamping process and through wear and damage to the disc. The background noise from a 78—a continuous mix of hiss, crackle, and clicks—is usually relatively loud and sometimes overwhelms the musical signal. To most CD buyers it would be totally unacceptable. Noise can be reduced somewhat by using a filter to remove all the frequencies that could not have been recorded on the disc. For practical purposes, for most 78s until quite late on, any signal above about 5000Hz is almost certainly noise and not content. Cutting it out should produce an immediate improvement in the sound. Unfortunately, it doesn’t, partly because the brain tends to assume that if it can hear high frequencies in the hiss it can hear them in the music too. So when one cuts off all high frequencies the music sounds dull. To an extent it’s an aural illusion, 54 but disturbing nevertheless. Some hiss can be musically beneficial, therefore.

¶45 Much of the noise we’d like to remove, however, lies within the same frequency band as the musical sounds on the disc. 55 There are essentially two approaches at the moment to reducing it. One is to edit out the worst of it in a digital sound editor, cutting out the clicks one by one: usually they are so short that the ear cannot tell that the musical note has been minutely shortened. The other, and more commonly used method, is to employ automatic noise-reduction, either through a software program or through dedicated hardware such as CEDAR. In either case an algorithm attempts to identify noise and to remove the frequencies that constitute it. Naturally this can only be done with digitised sound, so a further stage is necessary in which the output from the equalising pre-amp is converted, by an A/D converter, to the digital domain for further processing. Clearly noise-reduction changes the information taken off the original disc in a big way. It’s very easy to overdo, producing a relatively silent background but a thoroughly unnatural sound, and there are some truly awful examples available on CD. A promising alternative approach currently under development involves photographing discs and converting the shape of the groove into digitally generated sound, but at the time of writing it remains to be seen whether this will work on a commercial scale (bearing in mind that, in an industry with such small print-runs but a constrained retail price, ‘commercial’ means low-cost production). 56

¶46 Once the output from the noise reduction stage is saved, which at the moment is likely to be on computer disc, a transfer should have been achieved that meets modern expectations as closely as can be managed, given the nature of the original. But, as we’ve seen, it will differ from the original in many respects. Further editing may be done, usually on computer, in order to improve it, removing the noise of the stylus hitting and leaving the disc (‘topping and tailing’) and fading in and out the background noise at the beginning and end, for example; or perhaps more radical intervention. Speed/pitch can be corrected at this stage, digitally, if that seems more practical than at the turntable.

¶47 Remember, too, that we still have just one side of one disc to deal with. If the performance being transferred was originally spread over several sides, a further editorial task lies ahead as the separate sides get edited together into an apparently continuous performance, which may require not just the joining of the sides but also the editing of background noise and pitch levels where, as often happens, these are not consistent across the side-change. This, of course, is a major misrepresentation of the original source material, but one that seems to be regarded as absolutely essential by record companies and consumers. One might have thought that there would be a strong case to be made for issuing 78rpm discs as a sequence of separate sides, exactly as they were originally published. At least, unlike the original users, we wouldn’t have to get up and turn over the disc after each side had played. But clearly this is a minority view, and so one of the essential tasks of the transfer engineer is to stitch those sides together. What this involves depends on what happens at the side ends. As we’ve seen, musicians tended to slow down for the last few beats to make an artistic close; sometimes cadences were inserted, as in Sound File 7 (wav file), which collects the side breaks in the first movement of Schumann’s Piano Concerto as recorded on HMV with Alfred Cortot and the LPO conducted by Landon Ronald in 1934. 57 In such cases the transfer engineer has to cut the cadences out and join together the last chord that’s in the score with the beginning of the next side, provided that the result is a fair representation of the composition. If it’s not, or if, as occasionally happens, there is a gap in what was recorded, it may be necessary to supply the missing notes from another performance of the same piece. With the ritardando it is usually possible to join the sides without further editing, simply because the original musicians did their best to place side ends at cadences, and slowing down for cadences is musically acceptable anyway, even expected. Nevertheless, the slowing for a side end is very often greater than for a cadence, and consequently much larger ritardandi are present in the artificially continuous performance made by the editor than the musicians would ever have made in those same places in a concert. Equally, editors can cut them short by cross-fades. Students of tempo variation in early recorded performance can easily be misled. 58

¶48 Comparison of different transfers of the same recordings—of which there are now many instances easily available—shows at once just how much difference the transfer engineer makes. Speeds/pitches, tone colour, background noise, side-end joins can all be significantly different, and none need represent at all closely the sound of the 78s played on original equipment. 59 Sometimes they will sound better. Modern equipment can extract sonic information from the grooves that was possible to record but impossible to reproduce via early record players. Sometimes, though, that can be a disadvantage, as in the case of cold wax chatter, noise at ca. 10kHz produced by vibration from the cutting machine in wax that was cooling too fast, and which therefore tends to occur towards the end of a side cut in a cold studio, for example in Winter. Cold wax chatter wasn’t audible on their equipment because it’s above the frequency limit of gramophones, but it is audible on ours. However one looks at the issues, then, it’s never going to be possible to say, for any transfer, that one has achieved the original recorded sound. One can’t even say what that is, let alone know that one has recovered it.

¶49 How much does this matter for our purposes? It depends on what we are trying to say about the performances. If we’re studying tempo change, as we’ve just seen, it can matter a lot at certain moments. The same goes for studies of instrumental or vocal tone colour, which are essential to any work on changing ways of using instruments, or any work on individual voices. The ways transfers are made can have a marked effect on the detail with which instruments can be distinguished in larger ensembles. Textures can sound clearer or more muddled not because of things the conductor and players did but because of choices made by the modern transfer engineer. Parts of the music—individual notes, longer sequences, or even whole sides—can have been edited-in from other performances. So, while CD reissues can be extremely helpful for identifying those things we want to study more closely, and (given enough data) can support conclusions about performance style with widespread validity, any really detailed work on an individual performance can only be properly founded when we have gone back to an original disc and have played it ourselves, with various choices of equipment and settings. Our situation is not unlike that facing the musicologist who wishes to comment in detail on a piece of medieval or renaissance music. One can rely on modern editions, or one can go back to the original manuscripts and prints, check, and make up one’s own mind. Except that modern editions come with critical commentaries telling us what the editor has done and CD reissues don’t, or only, tucked into a paragraph at the end of the booklet, in the most general terms.

¶50 Equally, just as the editor of a manuscript may be exceptionally knowledgeable, far more so than the user, so a really good transfer engineer—a Roger Beardsley, a Ward Marston, a Mark Obert-Thorn—may be far better able than most academics to make the best judgements about speeds and equalisation. Not all modern transfers are that good, but in cases like these it can sometimes be appropriate to value the best CD reissues as highly as the original disc. 60 It depends very much on the kind of work one wishes to do. In the end, though, it’s impossible to know what to think unless one is in a position to hear what has been done. And that ideally means comparing any reissue one may wish to use with a copy of the disc, played in various ways in a well-equipped studio. Responsible work with early recordings, then, involves getting to know where to find the original discs, what can be learned from them, and how. Only then can one know whether the research one wants to do with recordings can safely make do with CD transfers or not.

¶51 So far I’ve talked about the musical information on a disc, and what happens to it when we try to recover it. But there is written information there too, and that can also be useful.

3.3. Discs and discography

  Plate 5: HMV B 2686,
                    matrix Bb12053, label
Plate 5: HMV B 2686, matrix Bb12053, label

¶52 Plate 5 shows a typical label from a 78rpm disc. The background colour is a reddish brown, known at the time as plum, and the writing is in gold. These were the colours designating The Gramophone Company’s B Series 10-inch discs, about nine thousand of which were issued between 1912 and 1958. 61 The Gramophone Co. used complex but logical schemes of numbering for its discs and series, reflecting content and country of origin. Series B contained mixed repertoire, tending towards the popular, or what we might now call light music or easy listening, but with a sprinkling of popular classics, mainly vocal or solo piano.

¶53 The company name is printed around the edge at the bottom: ‘Record manufactured by THE GRAMOPHONE CO., LTD., Hayes, Middlesex, England.’ At the top is the tradename ‘His Master’s Voice’ above the trademark, the painting by Francis Barraud of his dog, Nipper, listening to a gramophone record (originally an Edison phonograph but painted over in 1899 with one of the recent Improved Gramophones). 62 Across the centre of the label is the title of the item on this side—in this case, two items, since both were short enough to be fitted onto one side: ‘(a) WHITHER (b) HARK! HARK! THE LARK’. On the next line comes the composer’s name in smaller type and in brackets, ‘(SCHUBERT)’; then the singer’s name in larger type, ‘JOHN GOSS’; and beneath that the details of the accompaniment: ‘with Piano accomp. by Kathleen Markwell’. Next, on the left, comes the voice type, ‘BARITONE’; language, ‘In English’; and accompaniment again, ‘with Piano’ (this to help record shops file it correctly). Beneath this is the identifying number for this side of the disc, ‘(6-2939)’ which was habitually used within the company for identification purposes. In the centre at this level is a pasted-on stamp indicating the appropriate purchase tax for this type of disc. 63 To the right is the recommended playback speed, ‘SPEED 78’ (which may or may not be correct); and beneath that the catalogue number of the disc as a whole (embracing both sides), ‘Cat. No. / B / 2686’. This is the number by which the disc will generally be identified today in writing and discussion, and is properly called the coupling number (though more commonly these days the catalogue number), since it covers the coupling of the items on the two sides of the disc, one with the other. (The same performance could be issued in another series with a different coupling, in other words with a different item on the other side of the disc.)

¶54 Further and for specialists more interesting information is stamped into the shellac around the label. This is the identification data engraved into the metal stamper and reproduced inevitably thereafter. At the top is the company number for this side, ‘6-2939’; to the left is a ‘1’, the number of the positive from which the stamper was made; and to the right a ‘G’, which numbers the stamper. Gramophone Co. stampers were identified, in order of use, G R A M O P H L T D. (Popular records might go round the clock more than once, in which case the letters were understood as numbers 1-9 and 0, hence GD would be 10, GG 11, and so on.)

¶55 Most important of all is the matrix number at the bottom, since it is this alone that indicates, by uniquely identifying the wax master, the particular performance carried by this disc: Bb12053IΔ. Bb was a series of matrix numbers that the Gramophone Co. used from 1921 to 1930 for recordings made in the London area for Head Office at Hayes. 64 Thanks to the research of Alan Kelly in the EMI Archive, working from company documents, we know that 12053, and 12054 on the other side of the disc, come within a sub-series recorded between November 1927 and January 1928 in the C studio of the Small Queen’s Hall, London; these two sides were both recorded on the 24 November 1927.

¶56 The matrix number up to this point identifies the recording event: this pair of pieces recorded in one session. But a session could (usually did) include more than one attempt, or take. First there would generally be a trial (unnumbered), whose wax record would be played in order to check on sound and performance quality, and by being played, since it was only wax, would be destroyed. Only then were takes made for potential publication. As many were made as it took to get two acceptable performances (one for issue and one as a reserve in case of production problems). Each was numbered in sequence, and the number, as an uppercase Roman numeral, was stamped superscript immediately after the main matrix number, and before the triangle (the delta symbol) which indicates the electrical recording process used (here the Western Electric system). In this case the take number is a ‘I’, so it’s the first take, although Kelly’s research in the company archives reveals that three takes were made for each side of B2686. 65 While take numbers are occasionally found in double figures, and numbers up to four are not uncommon, most published takes are numbered I or II. It’s true that at this stage in the history of sound recording, before editing had become possible and perfection had become the expectation, performers and engineers, and indeed consumers, were content to accept small blemishes in a performance provided that the performance as a whole was engaging; and so an adequate result on the first or second attempt was not as surprising as it might seem today. But equally, we need to bear in mind that when major performer mistakes occurred, or if there was a visible technical problem, recording would stop and the wax would be set aside for recycling without a take number ever being allotted to it. The numbered takes in the company archives and on the discs, therefore, are those that were not obviously unusable; and so I and II may not necessarily be the first and second attempts.

¶57 In the case of expensive recordings involving, for instance, a symphony orchestra with vocal soloists, it was customary to record each take on two or even three sets of equipment in order to be sure of a technically acceptable take, and in that case the simultaneously recorded wax matrices will be stamped with the same matrix and take numbers, but with the addition of a letter after the take number indicating which set of additional equipment was used. Frequently the lathes were simply fed with the same signal, but sometimes, especially later on, different sets of microphones would each have their own lathe, which provides us potentially (if the metals survive) with, in effect, a different recording. Whichever produced the best sound was chosen for issue. Thus Albert Coates conducting Beethoven’s 9th Symphony merited two sets of equipment for the first three movements and the beginning of the fourth (CR759 to CR769), but three sets for the sides with vocal soloists and chorus (CR778 to CR782).

¶58 I’ve gone through this trainspotter’s trip to heaven mainly for the sake of the take number, for although it is of course important to understand all the information printed on a disc it’s the take number that, for anyone studying performances, is potentially really interesting. In order to understand why, we need to go back to the manufacturing process described earlier. You’ll recall that as the stampers were used over and over to print discs in soft shellac they gradually deteriorated. While fresh stampers could be grown from the original metal positive, this was not a problem. But where this was no longer possible through damage to the master, either accidental or from over-use, a decision then had to be made whether or not to keep the title in the catalogue. If it was decided to continue supplying it there was a choice: to remake the recording using the same performers or to attempt to make a transfer from a standard pressing. The latter was more cost effective and so was generally the route taken. A famous exception concerns the legendary ‘Hear my prayer’ sung by Ernest Lough with the Temple Church choir (HMV C1329). The original best seller, and we are talking of a million copies, was made in April 1927. By the end of the year it was obvious that the original masters could no longer be used to make satisfactory stampers. Thus in April 1928 both sides were re-recorded: the same matrix numbers were allocated, but with incremented take numbers, presumably to allow even experts to assume that they came from the original sessions. That later recording remained on sale as a 78 until the demise of the format in the early 1960s. 66 For students of performance instances such as this are, of course, fascinating, because they give us the possibility of comparing two or more performances made by the same performers in the same (and in this case even in a different) session.

¶59 The difficulty, as ever, is finding copies of the different takes. B2686, the John Goss disc, is listed in the company archives as having been issued only in this take; confusingly the discography of HMV Series B published by Andrews & Bayly in 2000, 67 gives the take number for this side as -2, which may be a typo or may indicate that they (or their source) had seen a disc with a different take number. These sorts of discrepancies can be tantalising, but on the whole most of us have better things to do than to research them. However, where different takes are easily available, or enough research has already been done, comparisons can be well worthwhile: we shall make some in chapter 8.

¶60 The alternative solution to the problem of worn out stampers was to make a transfer, as mentioned above. In acoustic days this was achieved by connecting the replay stylus directly to a cutting stylus, which worked surprisingly well: many copies of Adelina Patti’s are mechanical dubs, identified by the lack of her signature in the wax. Caruso was another whose records were popular enough to have had to be re-mastered in this way (identified by an ‘S’ over ‘8’ symbol). 68 With the introduction of electrical recording and replay, transfers became much simpler. A pressing was played with an electrical pickup and the signal fed to a cutting-lathe as the source rather than the usual microphone. Transfers done this way became quite common, and by 1930 were generally very successful.

¶61 Apart from the need to replace worn-out masters, transfers were frequently made to alter the sonic nature of the record, especially to enable it to pass the wear test. This required that a record should last at least 50 playings. Electrical recording placed great strains on acoustic replay equipment, and over-loud records would be chewed up within a few playings. Reducing the overall volume or just cutting bass could enable a record to pass the wear test. 69 Other reasons were not uncommon. Sometimes one sees in the recording sheets notes such as ‘transferred to increase volume of oboe solo at 1’43”’. Up until 1931, transfers were marked with a ‘T’ in the matrix number. 70 Later, a common indicator became an increase in the take number. Many of Gigli’s popular records show, say, a ‘take 3’ when only two were originally made, leading unwary collectors to mistaken ‘discoveries’. On other occasions a fresh matrix number would be allocated. 71 Victor electrical transfers are more difficult to spot and may be less common. However a small ‘R’ next to the take number is a definite indicator. Before its merger with HMV, Columbia would usually add an ‘R’ to the catalogue number but this indicates an electrical re-make of an acoustic recording, and not a transfer.

¶62 All these are sources of potential confusion to students of the performances, and emphasise how necessary it is to have a good understanding of record company procedures in order to see why performances with identical catalogue and matrix numbers may be different, and vice versa.

¶63 In the very early days, before moulding cylinders was possible, multiple copies were made right away by feeding the sound through numerous tubes to a bank of cylinder recorders. If more were needed, though, there was no alternative to getting the artists back into the studio and recording again. For this reason, what appear to be several copies of the same original may in fact preserve different performances, and it is therefore always worth listening to every copy of an early cylinder. Here, too, dubbing was common. 72

¶64 It is also worth being aware of the ways in which the same performance could appear in more than one series, so that the same side number could appear with different couplings and therefore with different coupling numbers. The Gramophone Co., in particular, tended, in the case of popular discs, to issue one performance in several series for different international markets. For example, the Albert Coates Beethoven 9 mentioned above was issued by the Gramophone Co. in six different series—4-, D, AB, AW, EJ and ES—while a further matrix, made using one of the other sets of equipment (not always the same set, and sometimes both), was sent to the USA for issue by the Victor company, with whom the Gramophone Co. had a very long-standing arrangement to exchange recordings for sale in one another’s territories. All these publish the same takes, so that if you can’t find a decent copy in one series, one may still be found in an archive under one of the other catalogue numbers. The exchange agreement with Victor, of course, resulted in the same performances being issued not just with different catalogue numbers but under the names of different companies.

¶65 A final common cause of multiple issues was the habit of reordering the sequence of sides in a multi-disc set so that the discs could be played in sequence by an auto-changer. This was a mechanism built into later gramophones that allowed the user to put a pile of discs on the machine, and have the machine play them in turn, minimising the inconvenience of having to change sides. Thus for a 4-disc set (a convenient quantity for a nineteenth-century symphony) the sides would be paired 1-8, 2-7, 3-6, 4-5; the machine would play sides 1-4, the user would turn the pile of discs over as one, and the machine would play sides 5-8. These reorderings would be reissued with a separate set of coupling numbers, usually in a separate part of the numerical series. Thus HMV Series DB reserved numbers in the 7000s for autocouplings (the main beneficiaries at first being Gilbert and Sullivan operettas previously issued in the 3000s). Again, this offers us the opportunity to find further copies of discs in which we may be interested.

¶66 None of this information can be translated directly into dates, though much of it can be extremely helpful if enough interlocking information is available. Numbers were generally allocated in sequence (though sometimes in blocks), so if one has a date for some matrices in a series it will often be possible to narrow down the dates on which others were recorded. For the main labels much of the research has already been done in the surviving paper archives, and a good discography will be able to provide dates not only for the recording of each take but also the dates of issue and deletion from the sales catalogue. 73

¶67 As with any kind of research or collecting (which is psychologically much the same thing), details can be assembled because they are useful or because they are there. And it can sometimes seem, when one goes through the kinds of information that can be recovered about early recording, as if much of the data discussed is of the latter sort. I can only emphasise, as a musician with a very limited interest in technology, that all the information I’ve detailed here can make a difference to the kinds of things we can sensibly say about performances on record. And that’s why we need access to them, and why we have so much to learn from professional discographers. If you don’t understand how your sources were made you’ve much less chance of understanding the sounds they transmit.

¶68 At this point some advice on using published discographies would be appropriate, but in fact it’s been so well covered recently in Simon Trezise’s chapter for the Cambridge Companion to Recorded Music that I can do no better than recommend readers who need guidance to begin there. 74

3.4. Piano rolls

¶69 There is another early recording technology we need to know about; in fact, in this case it’s even more essential, because if we don’t know how piano rolls are made we’ve no hope at all of knowing what to make of the performances they encode.

¶70 The first thing to understand is that there are two crucially different types of player piano and music roll, those that encode information about dynamics and rubato and those that don’t. The first type play on what are now known generically as reproducing pianos, and second type on player pianos, the latter being the foot-operated ‘pianolas’ more commonly found today. All types are operated by suction, generated at the pianola by the player pumping two large pedals, in the reproducing piano by an electrically driven pump. In both types the piano action is played by a series of small pneumatic motors or bellows, one for each note. The mechanism is controlled by a roll of paper, usually 11¼ inches (28.6cm) wide, and it is these rolls that, by one means or another, encode the original performance.

¶71 Rolls for pianolas were not, strictly speaking, recorded at all. They encode only pitch and duration, which was notated by hand, copying from a score. Expressivity had to be supplied by the operator at playback stage, and this was done by using the pedals to control dynamics and accents and a system of hand levers in the keywell which allowed the performer (a word that under the circumstances seems fair) to modify the playback speed continuously during the playing of the roll and thus to generate phrasing and rubato. The sustaining and una corda pedals were controlled by two further hand levers. Without any intervention by a performer, pianola rolls produced sounds hardly more musical than a modern MIDI encoding of pitch and duration. 75 (I say ‘hardly more’ rather than ‘no more’ because the irregularities in the live piano provide, albeit accidentally, a semblance of musical shaping that a computer does not.) But with the participation of an experienced player a pianola can produce miraculous results. What is so fascinating about it from our point of view is that, uniquely among musical instruments, the pianola separates out ‘getting round the notes’ from ‘musicality’. The machine takes care of the right notes, leaving the performer to do everything that makes a performance expressively musical, or not. It is thus, potentially (though as yet this potential seems entirely unrealised), an extraordinarily valuable teaching tool. In a totally different way from the reproducing piano, therefore, the pianola offers fascinating possibilities for students of performance style and musicianship. Every teaching and research institution should have one.

¶72 In the case of reproducing pianos, the pianist played as normal, and what he (rarely she) played was more or less automatically (according to the system) marked in ink onto a roll by the recording mechanism, assisted by an operator, the extent of whose musical (as opposed to technical) contribution remains unclear. Later, the inked roll could be edited to remove imperfections in the performance or the recording, then converted by hand to a perforated roll, which in turn, and having been checked by the original pianist, was reproduced by machine in multiple copies for sale. The mention of editing should already be raising concerns. More will appear as we look further into the details. Nevertheless, reproducing piano rolls encode enough of performances and pianists we have no other means of hearing to make perseverance worthwhile.

¶73 There are three main systems of reproducing piano, the Welte, the Duo-Art and the Ampico. I don’t propose to go into the differences here, 76 although it’s important to understand that none will play rolls made for another, and each—rather as with the record companies—aimed to build up their own roster of pianists who would record exclusively with them. Most owners bought only one instrument (and incidentally a huge number did so), and so were restricted to the rolls designed for it. Each system built up a catalogue of several thousand rolls during the 30 years or so in which reproducing pianos were made; it is estimated that about 20,000 rolls of classical and light music were made in all, and perhaps a similar quantity carrying popular music. 77 The Welte-Mignon was the first, introduced in 1904. In 1912 the Ampico system appeared, continuing to be developed until the end of the era in the 1930s. 78 And in 1914 the Aeolian Company in America introduced its Duo-Art system.

¶74 What can we learn from reproducing piano rolls? While the original pianist’s tempo was encoded inevitably by the length of the perforations, there remains more doubt as to the role of the assistant in fixing the values of other parameters. Editing after the fact clearly had little bearing on what was actually played first of all, except in so far as editor or pianist could remember or re-imagine. Moreover, the mechanisms were hamstrung in their ability to reproduce dynamics precisely both by their ability to distinguish only two dynamic levels at once, one for the lower notes and one for the upper, and by the modest number of dynamic levels that the suction mechanism was capable of differentiating. 79

¶75 When we listen to a recording of a piano roll we are of course several stages further removed from the original sound. First of all, the instrument has to be in perfect working order. Few specialists exist who have recovered the skills to restore them adequately, and not all modern recordings use well-prepared instruments. Another approach, rather than to restore the piano with a built-in mechanism, is to use a ‘push-up’ or ‘Vorsetzer’, in which the roll mechanism operates levers (‘fingers’ in the trade) which, when placed in front of a piano, play the keys. In either case the sensitivity of the result depends very much on the expertise of the restorer, particularly so since in the case of Welte as many details of the original recording equipment as possible were kept secret and thus much modern knowledge has been acquired by trial and error, the results shaped by modern expectations. Secondly there is the question of the piano itself. Any performance at a piano, and most of all performances by the finest pianists of the sort who recorded on rolls, has its details determined by the way that particular instrument sounds. If you replace the instrument with another it’s inevitable that some of those details, on which the expressivity of the performance depended, will no longer work in the same way. 80 Consequently, piano rolls are not necessarily best reproduced on the finest modern pianos; and whatever piano is used, it is never going to give the results heard in the recording studio. The huge benefits of recording rolls on modern pianos—that here we seem to have early twentieth-century pianists playing as if in a modern hall with modern sound reproduction—are illusory, therefore. It sounds as if Rachmaninov is present, but he’s not.

¶76 A comparison of a reproducing piano roll and a performance by the same artist recorded on disc will give us some help in putting all these points into perspective. Of course it’s not the same performance, but it’s the best we can do to assess what piano rolls can and can’t pass on. Sound Files 8 (wav file) and 9 (wav file) contain extracts from recordings of Grieg playing his own Norwegian Bridal Procession Op. 19 no. 2, and Sound Files 10 (wav file) and 11 (wav file) extracts from Raoul Pugno playing Liszt’s Hungarian Rhapsody no. 11. The first of each pair is a Welte roll transferred by Denis Hall, whose reproducing pianos are probably as carefully restored and maintained as any today; the second in each case is a transfer from a Gramophone & Typewriter Co. 78rpm disc. 81 Both discs were recorded in 1903, the Pugno roll was made in 1905, the Grieg in 1906, so neither disc is exactly contemporaneous with the corresponding roll. In general terms the performances within each pair are very similar; only a few details differ. Grieg on disc tends to accent the second beat of the bar in the opening theme, whereas his roll does not. He may have changed his approach, although it would be quite a significant change to make to the principal theme in a character piece; or it may be that the assistant normalised what he perceived as an irregular accentuation. With this roll a further comparison is possible, because a playing of it was also recorded on an Odeon disc using a Welte piano in 1930, in other words at a time when technicians trained in maintaining the instruments were still available. According to Eliot Levin, ‘As Welte-trained service engineers were still working then, there is every reason to suppose that these transfers were made in optimum conditions; the Odeon people would have made certain of that. There is every reason to believe that the venue was Welte’s own studio, and that the master-rolls themselves, rather than duplicated rolls, were used. So although as an elderly gramophone record, the sound may not be wonderful, I do believe that we are hearing a roll in optimum circumstances.’ 82 It would be reassuring to have some firmer evidence. Certainly the speeds and therefore the timings are significantly different from Hall’s, probably due to a combination of different roll and turntable speeds, and there is a pitch difference between the 1930 roll and the other two recordings. Despite Levin’s belief, the 1930 Welte seems to articulate less clearly than either Grieg on disc or the modern-maintained instrument, 83 and there are tiny differences in relative loudness in which the Hall recording seems slightly more subtle, all tending to cast doubt on how ideally the 1930 Welte had been set up. Advocates of the reproducing piano will find much to discuss here. But for us, much the most significant point is that neither playing of the roll can match the subtleties of Grieg’s rubato and dynamics. They are good, but not good enough.

¶77 The same story is told by Pugno’s Liszt performances, particularly interesting in those passages that imitate cimbalom playing, involving very fast repeated chords. Compared to the disc, the roll’s performance is notably even in its dynamics, and as a result seems slightly artificial; this is even more the case at the end of the piece where the roll is unable to emphasise the melody notes that appear from time to time at the top of the repeating chords. Pugno’s performance on disc, although the sound is terrible and the texture, as a result, is a complete mess, sounds plausibly musical in a way that the roll does not. That said, for pianists not recorded at all on disc, transfers of rolls as good as Hall’s are well able to tell us about general style. We must just be careful not to draw any conclusions from their details. 84

¶78 Others transfers are less satisfactory, and so it’s clear that the way the instruments are restored and set up for performance is crucial. Comparing two recordings of Duo-Art 6566, which encodes Paderewski playing Chopin’s Mazurka Op. 24 no. 4 in November 1922, recorded with different pianos, one played by a push-up, the other with the mechanism built-in, maintained by different specialists, and recorded in different acoustics, a host of details in the performances we hear are significantly different. One, on the Nimbus label, takes eight seconds longer than the other, on Aeolia; 85 although it speeds up at first, the Nimbus roll then starts to slow, and falls fairly consistently behind thereafter, though not at an entirely regular rate. Or if you prefer Aeolia moves ahead: it’s impossible to know. 86 As far as the sound goes, Nimbus’s modern Steinway, played by a push-up (in Nimbus terminology, a robot) is much brighter and more clangourous than Aeolia’s 1926 Steinway Duo-Art, and as a result the details of timing from note to note produce quite dramatically different effects. Many of the things one might say about the pianist’s response to the score at any specific moment would be different if one used the other recording.

¶79 There is no doubt that, carefully played on a well-restored piano, a reproducing piano roll is capable of producing remarkably lifelike results, considering the limitations of the process. But there is no way of knowing which details one hears faithfully reproduce what the pianist played. Consequently, for detailed work they cannot be relied upon. That said, we can still learn things from them that transform our understanding of a piece. The classic example, much discussed in recent years, is the Welte roll recorded by Debussy on 1 November 1913 of his Prélude ‘La cathédrale engloutie’, on which he doubles the speed for each passage notated in minim chords, and halves it again when the notation returns to crotchets. 87 The minim = crotchet is not indicated in the score, and the roll is thus the only evidence of Debussy’s intention, but the difference is so radical (and so much an improvement) as to be impossible to set aside as a momentary whim. Consequently the new Debussy edition includes the requisite indications of speed change, 88 and the very many recordings made by other artists between Debussy and modern times, which play with a notionally constant beat, are now considered to be mistaken, even though they include recordings by pianists otherwise considered as among the greatest Debussy interpreters. 89

¶80 No less extraordinary, I think, is the slow movement of Mozart’s Piano Concerto K. 537 (the so-called Coronation concerto) on Welte 237, recorded by Carl Reinecke in 1905. Reinecke was born in 1824, somewhat after Mozart’s death, but before Beethoven’s. He grew up absorbing, presumably, performance styles current in Denmark in the later 1830s and early 40s. How much of that was still present in his playing in 1905 is impossible to say. As we shall see in chapter 7, 90 while a majority of artists seem to play broadly consistently throughout their careers, albeit absorbing something of current trends as they pass, some change more fully in line with changing tastes. We can’t know where Reinecke lay along that spectrum. But with all those provisos, his Mozart playing is truly astonishing, and utterly unlike anything we have ever imagined in modern times of historical Mozartian playing, but also—significantly—much more radically expressive and more highly decorated than any other Mozart piano playing on early recordings. By all the standards we have become used to, whether in modern or early recorded playing, it is exceptional, and thus of the greatest interest. A sample is given in Sound File 12 (wav file). 91

¶81 Before leaving pre-Second World War recording it is important to mention the beginnings of radio. Particularly relevant to us is broadcasting policy of the BBC which from 1922 aimed to raise musical taste by broadcasting what its Music Department considered the best music both old and new. 92 Music was broadcast either live—and in due course enough was required to sustain a national BBC Symphony Orchestra as well as several regional orchestras—or from records which were bought in multiple copies by the BBC Gramophone Library and sent out to regional stations for broadcast. 93 Very little recording, however, was done by the BBC until the outbreak of war, when recording broadcasts onto acetate discs became much more useful. Recording as a matter of course, however, began only with the introduction of tape recorders into programme production in the early 1950s.

3.5. LPs

¶82 We have cinema and the Second World War to thank for the long-playing 33⅓ rpm disc, which was to be the standard format for recorded sound from the late 1940s until the introduction of the digital compact disc in 1982. 94 Sound tracks for films, introduced in the late 1920s, could either be printed on the film or on a separate disc to be played simultaneously with the film projection. In order to make a shellac disc run for the same amount of time as a 1000-foot reel of film—about eleven minutes—the disc had to be run at 33⅓ rpm. Adapting this format for recorded music, however, posed several new problems. While techniques were gradually introduced for assembling film sound tracks from several smaller recordings, musical performances aimed at record buyers needed to be continuous, the sound quality needed to be good—and for both reasons dubbing from smaller takes was not suitable—and performers had to be able to make acceptable takes that were between two and three times as long as for a 78rpm disc. Consequently, although there were attempts at introducing long-playing discs early in the 1930s, they were not especially successful.

¶83 At the same time, in the later 1920s, recording machines began to be developed that stored sound on magnetic wire instead of on wax and shellac discs. The wire was fed from a reel through the machine to a take-up spool. Electrical signals from a microphone fed an electromagnet (the recording head) that magnetised the wire in proportion and frequency to the original signal as the wire passed by the head. Running the recorded wire through the machine allowed the sound to be played back via an amplifier and loudspeaker. The wire could be stored and played, or reused many times; it provided a longer running time than conventional discs; and if necessary it could be edited by cutting and soldering the wire, removing unsatisfactory material and soldering in something better. The main disadvantages were that the wire tended to break, with potentially dangerous consequences for the operator; wire and tape recordings were far more expensive and time-consuming to duplicate on a mass scale than discs, and on wire the sound quality was no better.

¶84 Magnetic recording improved, though, with the invention in Germany in the early 1930s of magnetic tape, which was safer than wire to operate (though more flammable) and easier to edit. During the Second World War, recognising the propaganda value of continuous recording and almost instant replay, work continued on improving the system, and German scientists developed a technique for storing a much wider range of frequencies on tape. When the Allies requisitioned German radio stations in 1945 tape recording was found to have advanced way beyond pre-War capabilities. Among the material recovered by the Russians were tape recordings in stereo, a few surviving examples of which resurfaced in the 1990s including an astonishingly clear recording of Bruckner’s 8th Symphony under Karajan recorded in Berlin in 1944, which, although it lacks the first movement, is well worth searching out. 95 Compared to performances on contemporary 78rpm disc it is truly remarkable. 96

¶85 Another product of war work, this time in Britain and the USA, was the ability to store a much wider range of frequencies—up to 14,000Hz—on disc, useful for a variety of secret purposes including anti-submarine warfare. Decca was at the forefront of this work, culminating in its ‘Full Frequency Range Recording’ (ffrr) system, enthusiastically promoted over the following decades. So by the time the recording industry began to recover after the War all the technology was available to produce long-playing discs with a much better frequency response, carrying recordings that had been edited to give the cleanest possible performance. The consequences for anyone studying performances on record are very far-reaching. Once it was possible to edit easily, by cutting and splicing magnetic tape, it was possible to record in takes of any length. Because the tape ran, typically, at 30 inches per second, cutting out a few inches of tape would remove tiny blemishes; sounds fractions of a second long could be spliced in from another take; or much longer passages could be replaced. The end result could be a tape made up of fragments of many separate performances, assembled by an editor into what sounded like a coherent whole. With the introduction of tape, then, we see the end of live performance on record, save for special occasions marketed as ‘live recordings’. From this point on, records aimed to transmit something more perfect than a live performance, more perfect than any performer could achieve in one unedited take—more perfect, and yet less real. 97 Everything we say about performances from now on has to be said with that in mind. We’ll return to this question a little later in order to try to decide how much it matters.

¶86 The final piece in the technological jigsaw was the introduction in 1948 by Columbia of the microgroove 33⅓ rpm record. Pressed in vinyl instead of shellac, and with its grooves smaller and closer together than the 78, the microgroove disc could carry 23 minutes of music, which was enough for all but the very longest symphony movements or continuously running operatic acts. RCA Victor responded almost immediately by introducing the 45rpm ‘single’, a seven-inch disc containing the 4½ minutes or so that had, through the 78, standardised the duration of popular music titles. It was on singles that most pop music sold throughout the glory days of rock ’n’ roll; it was probably the most influential medium for the dissemination of popular culture ever invented. 78s continued to be produced, in parallel with LPs and singles, especially in Britain and the Commonwealth countries, right through the 1950s and into the early 60s, but were bought in declining quantities and only by those unwilling or, beyond the developed nations, unable to invest in new playback equipment.

¶87 Stereo sound on disc and tape was introduced commercially quite soon after the LP. Again, experiments from before the War were taken up afterwards, leading in due course to multi-track tape, in which several tracks were recorded in parallel along the same length of magnetic tape. Two tracks, of course, were all that were required for stereo, but the same principle, allied to wider tape in order not to compromise sound quality, was extended to four, eight and sixteen tracks in order to allow performances to be assembled a part at a time. The balance between the parts could easily be adjusted later in the studio before the whole was mixed down to two tracks for cutting to a master disc, one channel on each side of the groove. It would be a mistake to think of multi-tracking as a technique used only for popular music. While it is true that it became (and remains) the standard way of assembling performances in popular music, so much so that live performances are often difficult to bring off with the same success, 98 in classical music, too, multi-tracking gradually replaced mixing straight to stereo. Even without multi-tracking, though, radical manipulation of balance became possible early on. Sometimes the results are intentionally audible, as in the famous Decca Ring cycle in which the producer, John Culshaw, mixed sound effects with the instrumental and vocal performance in a virtuoso display of studio technique. 99 Most of the time, though, we as listeners are quite unaware, unless very familiar indeed with the sound of these same pieces in concert, of just how much has been changed in the arrangement of instruments and voices in the so-called sonic stage. The term itself is revealing, for it’s clearly not supposed to be the same as the concert stage. In the same way the title of Culshaw’s book about that recording, Ring Resounding, with its pun on re-sounding, celebrates the remaking of Wagner’s score into something for the imagination rather than for the theatre. From now on, then, we can no longer trust the balance of instruments on a disc as a faithful representation of the balance heard in the studio. Soloists could be brought forward, other groups pushed back, in order to make a performance that balanced the materials in the score in a new way. I put it like this to stress that while the results may not have been natural, in the old sense, they were not necessarily unmusical. And the fact that we find recordings made this way musically satisfying, often gripping, is one with which we shall have to engage later on. Artificiality in the making of a performance doesn’t necessarily produce a less satisfactory musical experience. On the contrary, for most people the results are preferable.

¶88 By quite early in the 1950s, then, what people heard from record players was radically different from what was heard before. Measured by sound quality it appeared to be far more lifelike that anything that came off 78s and cylinders. Yet older listeners continued to prize their 78s, and critics continued to refer back nostalgically to performers long gone. Although in the 70s and 80s one heard less of that, especially once CD made 12-inch discs and their playback equipment seem awkward, and 78s fit only for the dump, the torrent of reissues in recent years had unquestionably reawakened the sense that there is a directness of communication through early recordings that was later lost. There seems little reason to doubt that editing is what makes the difference. But is it because an edited performance sounds unnatural, or is it rather because of what editing did to the way musicians thought about performing? We’ll return to this later, but for now it is worth bearing in mind, as we look briefly at digital sound, how the increasing perfection of sound reproduction, coupled with increasing expectations of critics and record buyers, has encouraged musicians to focus much more attention on being accurate and on being typical than on being innovative or on being moving.

3.6. Digital sound

¶89 Digital recording was available to professionals from the late 1960s, and during the 70s record companies gradually changed over to digital tape. Combined with Direct Metal Mastering (in which the recording lathe cut straight into metal, instead of into lacquer followed by electroplating) digital recording produced very good sound from the last generation of LPs. Indeed, it is useful to understand just how good LPs got during the late 70s and early 80s, because the first CDs were not really very good at all. While a huge amount of marketing hype was generated by the release of CDs and CD players in early 1983, the really surprising thing about them at the time was not the quality of the sound, which was grainy and unnatural, but rather the silence of the background. Even the best analogue recordings included tape hiss, and even the best vinyl discs generated surface noise. It was the absence of that that was so striking.

¶90 In principle digital sound is simple. The sound wave is sampled a large number of times per second—44,100 for CD, more in some other formats—and the numbers are used to record its shape. The sound is thus represented by a sequence of numbers which can be stored in any format you like; at first it was on tape, later on computer. In order to edit the sound one simply changes the numbers. There are snags, though. Because a soundwave is curved and numbers are discrete, at the most detailed level the wave is represented by steps and not a curve. The edges of the steps have to be smoothed out by filtering. In among the numbers that encode the sound waves is a lot of complex coding that records the position of the data within the file, information about track numbers, and increasingly now information about the content, including titles, composers, and other information that can be displayed during playback. There is really no sense, then, in which a digital sound file is at all like an analogue recording on 78 or LP. But it has two overwhelming advantages. It offers a very convincing representation of the original sound, and (given a lot of very complex software) it is extraordinarily easy to edit. And done carefully, digital editing is impossible to detect. It’s become normal, therefore, for record producers to edit far more frequently than before. What this means is that it’s reasonable to assume that most pages in a score that’s been recorded have at least one edit in them; sometimes there are many. 100 Moreover, working in the digital domain, and editing at a computer, it is remarkably easy to change any detail of a recording. What was done for several decades with a razor blade and sticky tape, substituting slices of the complete recording, has now been refined to the click of a mouse, operating on as much or as little of the whole sound as one wants to alter, changing it in pitch, speed, colour, amplitude, while juggling, overlapping and combining as many takes as one needs to in order to get whatever result one desires. Clearly, CDs, for all their apparently lifelike sound, do not carry performances in the sense in which performers and audiences normally understand them, nor in the sense in which early recordings, despite their compromised sound, preserve them from the past. But to what extent does this matter?

¶91 Before we try to answer that, let’s look back over the technologies described here and try to list their pros and cons as evidence for the sounds originally produced. 101

Cylinders and 78rpm discs (acoustic era)
  • do preserve single unedited takes
  • can transmit accurate relative timings and relative pitches (subject to speed change during disc cutting)
  • cannot transmit precise data on speeds or pitch
  • cannot be used as sufficient evidence for the colour of a voice, still less an instrument, because they don’t transmit the higher frequencies that fully determine it, and the recording equipment tends to alter the balance of those frequencies that get through to the disc
  • do not represent concert-practice scorings in instrumental music, nor concert balances between instruments, nor venue acoustics
78rpm discs (electrical era) improve on acoustic era recordings in these respects:
  • can transmit more information about colour
  • do represent concert scorings, and are somewhat more likely to represent concert balances, and are much better at representing recording venue acoustics
LPs and analogue tape
  • do provide a relatively faithful representation of sound colour
  • do provide reliable data on speeds and pitch
  • because of tape editing cannot be relied upon to represent one continuous performance
  • do not consistently represent the balance of instruments/voices heard by a listener at the recording session
  • may or may not represent the balance of instruments/voices heard by a listener at the recording session
Digitally recorded LPs, CDs and subsequent media
  • may or may not provide a faithful representation of sound colour
  • may or may not provide unaltered data on speeds and pitch
  • are unlikely to represent one continuous performance for more than a few seconds at a time
General notes
  • ‘Live recordings’, whether on LP or later, may have been assembled from recordings of several continuous performances
  • LP or CD transfers from 78s may not represent faithfully the content of the original discs.
  • Digital ‘remasterings’ of analogue recordings may not faithfully represent the sound of the original tapes.

3.7. Editing and its implications

¶92 With all these limitations in mind, and having a clearer idea, I hope, of what kinds of documents recordings are, we can now turn to the key question for the study of performance from recordings: is a recorded performance a useable substitute for a real performance? This can be a philosophical question if we want it to be, and indeed philosophers of music have examined it. We’ve already seen Godlovitch comparing recordings to photographs, 102 and I suggested that film would have been a better, though still flawed, analogy, for at least film uses events ordered in time so as to create meaning. And as José Bowen has said, 103 we recognise that film is edited, and that things happen as a result that could not be achieved in a single take, and we’re perfectly happy with that. But film editing is not really very like music editing. Film editing is more like performing—realising the script 104 —while editing musical recordings is more a matter of tinkering with a performance (which may explain why film-making has long been regarded as an art while recording is still generally considered a craft). 105 You can falsify a musical performance on record in a way that you can’t falsify the acting of a film, because acting a film is never intended to be done in one take: illusion is the whole aim. A better analogy is film of a theatre performance; but nobody does that, because film can always make the same story more thrilling, since theatre implies visual contexts, and film is much better at making those real: film replaces the theatre-goer’s imagination. Fortunately, there isn’t an equivalent for listening to music (except opera, and the singing gives film an impossible task in creating an illusion of reality). So we can’t quite decide what recordings are by looking to film.

¶93 For Davies one of the most significant disadvantages of recordings is the thing that for us makes them essential, their repeatability. 106 Live performances (performances at which one is present) benefit from their unpredictability and their creativity; recorded performances gradually lose those qualities the more one listens to them, and any quirks eventually become annoying. (That belief, incidentally, which has done so much to inhibit dangerous performance on record, may well be at least partly mistaken if experience of early recordings is anything to go by. Davies quotes Alfred Brendel: ‘The studio is ruled by the aesthetics of compulsive cleanliness’, which seems about the right analogy. 107 ) Davies sees the perfection of studio recordings as compensation for the lack of spontaneity, and concludes that studio recorded performances are not worse, but simply different.[313]

¶94 Clearly, if one considers recordings from a production point of view, reconstructing in one’s mind, as one listens, the studio and the procedures that made the performance, then Davies is absolutely right. But do we listen to them that way? It’s possible to, and if one were a record producer or sound engineer one very probably would, at least when choosing to do so. But I don’t think many people do, and it’s certainly not what records are made for. They’re made to sound like performances, and on the whole, most especially when we’re engaged in the performance, that’s how we perceive them. So it seems more realistic, and more appropriate, to evaluate recordings and their relationship to live performance from a perceptual point of view.

¶95 From this perceptual point of view, one dimension is of course lacking, and that’s the visual. Hearing music without seeing it being performed is a different experience. When we can see what the performers are doing as they make the sounds we gain a visual experience that’s at least closely tied to the sound. But what sort of tie is it, and it is always helpful? Performers make two kinds of movement (considered conceptually), movements necessary to produce the sound (for example moving a bow over a string) and movements that express visually something that they mean the sound to express. A movement of the bowing arm may be accompanied by a movement of the whole upper body, or a visibly indrawn breath, for example. Those kinds of visual signals can contribute to a significant degree to the meaning of a performance, and on record they are entirely lost. As compensation in the studio, a player may decide to make the sound gesture larger to compensate for the lack of video. One can’t as a listener recreate the original visual gestures, though an experienced listener will undoubtedly invent gestures of her own in her mind as she listens, in so far as she is imagining the performers performing. So there is a downside to recordings in this respect, but it’s not as great as is sometimes made out. Moreover, much of the visual information that one takes in during a live performance is, however normal, distracting. The horn player draining spit out of his valve—which, as Davies reminds us, 108 fascinated the young John Cage more than the music—is not conducive to a sublime musical experience. Nor is the soloist mopping his brow; nor even the Evangelist’s stagey solemnity. Listening without seeing can very easily gain in intensity.

¶96 More important shortcomings of recordings than this include the absence, in most domestic circumstances, of the physical impact of the sound hitting one’s body in a three-dimensional soundspace. New technology is doing something to address this, and the option of playing recordings at a realistic volume does help. Although a believable facsimile of concert-hall sound is much closer than it was, we’re not yet in a position to use a recording to talk usefully about that in an academic study (at least, I’m not). The other major shortcoming, which has yet to be addressed although the technology is available, is the predictability of recordings once one gets to know them that Davies and so many others have noted. Recordings with programmable options for randomly varied sequences of takes would be interesting to try, and, for reasons that will become clear in a moment, might not be as unsatisfactory from a musical point of view as we might think.

¶97 What this seems to suggest, then, is that recordings are by no means inherently unsatisfactory as substitutes for being present at a concert. Certainly they are different from a cultural point of view: they are not just experiences but artefacts that can facilitate experiences, but so are concert halls and orchestras. They make use of illusion, but provided one doesn’t insist on taking a moral stand opposed to illusion of any sort, the illusions they produce can be just as fascinating as the real thing. What we need to do, then, if we are to answer this question about the acceptability of recordings as research data on performance, is to make a practical distinction between what recordings are (or can be argued to be)—which is something culturally very complex that involves all sorts of illusions both on the production side and in the listening—and what they are perceived to be when one listens to the music. The brain treats them as performances, and is content to overlook all the tricks that went into making them and all the illusions that we could say are present in the sound as heard. And we're happy to do that first because in terms of style recordings operate within the range of possible performances that people currently find acceptable at the time of their production, and secondly because any unrealities in the sound are not perceived as problematic provided that the performance is interesting enough in itself. I don't mean to suggest that knowing how recordings were made is of secondary importance. The discrepancy between the 'original' and what is perceived is one of the things that's so interesting about them, which is another reason why we need to know about how they’re made (as well as the need to keep tabs on producers, and the inherent interest of the facts). My point is rather that we shouldn't allow our historians’ respect for the facts to blind us to the value of the illusion that we hear, because in the end it's the illusion that makes the music in our minds.

¶98 So having attempted an answer to our question in principle, let’s now go back to the practice and try to understand how it looks from the studio perspective. What is the producer trying to achieve? The short answer is that that depends on the producer, but each is, in the end, employed to produce something that will sell. So the producer, to a considerable extent and whatever her artistic aims, responds to what is perceived to be current taste among record buyers, as well as—more directly—whatever she believes the record company expects of her. But accepting that, what are the producer’s artistic aims? A very interesting case study is offered by Robert Philip in his observation of a recording produced by Andrew Keener, a freelance producer working on this occasion for Hyperion. 109 Keener emphasises the need for cleanliness and repeatability, as one would expect a modern producer to. But beyond that he sees editing as a process of ‘collecting all the jewels’: a jewel may be ‘a nuance, ... a group of notes which are most beautiful, which that person didn’t play in quite that most beautiful, spontaneous, coloured way in any subsequent take... And you owe it to the artist to mine the ... takes for that kind of thing.’[55-6] ‘But’, as Philip explains, ‘the starting-point is complete takes.’ For Keener, then, the aim is to discover, from listening to complete performances at the start, what the performers are trying to achieve, and then to create a more perfect representation of that than they can achieve in any one take.

¶99 Other producers, sharing Keener’s general artistic aim, may seek to reproduce what they perceive to be the performers’ sense of overall structure, what producers sometimes call 'the span', and in order to do so may choose takes that are not necessarily the best in every way but that work together cumulatively to create the sense of structure that the performers create in a single successful run-through. 110 Others may prefer to focus on the moment, and in that case will use the most beautiful and accurate takes they have. Keener’s approach is an idealistic composite of those two somewhat opposed options (albeit one requiring a great many edits). And any producer one might interview would come up with slightly different aims again. But few classical music producers have any interest in producing performances that don’t sound like very good, real, performances.

¶100 Producers may also play a considerable role in shaping the performance as conceived by the musicians—the ‘span’ itself—through comments and (diplomatically phrased) advice. So their musical role isn’t confined to post-production work, to being creative with the takes. But how much input a producer can have at that level depends very much on the performers. While an experienced and widely respected producer can intervene quite a lot with young performers, stars make their own judgement of what’s successful and, aside from requesting additional takes to deal with creaking chairs, noisy page-turns, planes overhead, and so on, the producer may not be able to get musical decisions changed. A lot depends on personalities, and one can’t define the role of the producer on any particular disc without being there oneself or at least interviewing a lot of people involved. Moreover, performers experienced at recording may be very good indeed at picking up speeds and styles (moods, atmospheres, levels of concentration) when recording patches or sections of a movement so that they match exactly what’s already been put down. Thus the producer in sticking together takes is not having to do anything much to select those takes that fit together best. The performer’s conception survives being recorded in bits. In general, then, the influence of the producer can be exaggerated easily, but on those occasions when it is considerable it can be extremely creative.

¶101 What are the consequences of this relatively benevolent view of the role of the producer for our view of what a recording is? The first thing to understand is that performance can be flexible in several dimensions (frequency, amplitude, duration) and still be convincing. That is to say, a performer can play a phrase one way, or another, and still make good sense with it within the context of the performance so far. Editing can therefore change details of a recorded performance, just as a live performer can be unpredictable (even to herself) in what she’s going to do next. And that means that whatever one may feel in principle about editorial intervention in the studio one can’t argue that it’s undesirable on musical grounds. A performance produced after editorial changes can be just as persuasive as one produced without them.

¶102 One can test the limits of this flexibility by editing existing recordings, changing, for example, the lengths of pauses or cadential preparations to see how much they can be stretched or contracted before one arrives at a musically unacceptable reading. Everything depends on context, but in many cases there is no difficulty about accepting a wide range of note lengths in such situations. Of course the consequences will be subtly different, and each would have tended to encourage the performer to play what follows in a slightly different way, just as each would, in a real performance, be the consequence of slightly different preceding passages. To a certain extent the performer is improvising these effects and their consequences as he plays, taking a chance on holding a note, or speeding up a little, or playing a fraction louder, waiting for the right moment to move on. The results can be measured and thought about, and that in turn may offer ways of studying the process, for example by getting this kind of work on music-making to interact with empirical work on perception and brain-hand coordination in performance.

¶103 How far can the length or loudness or some other quality of a note be changed before it no longer works or before it comes to mean something else? In a previous study I illustrated a ready-made example from a Deutsche Grammophon disc of Schubert songs sung by Kathleen Battle accompanied by James Levine. 111 It’s a phrase-join in Nähe des Geliebte, and ‘offers a rare but useful example of an editorial change that definitely doesn’t work because it’s not plausible, it wouldn’t happen in any imaginable current live performance.’ 112 And of course that is the crucial test. If we can imagine it happening live then we can live with it. This DG disc is very heavily edited. On the minus side that means one has to use it with great caution, but on the plus side it means we can learn something of what’s gone on. DG has high production standards, so it seems safe to assume that this particularly awkward moment survived onto the disc because, for whatever reason, the session ended without everything perfectly ‘in the can’.

  Figure 2: Timings in
                    Battle/Levine, Nähe des Geliebte, bb.
Figure 2: Timings in Battle/Levine, Nähe des Geliebte, bb. 4-6

¶104 Figure 2 shows the notation of the relevant passage, annotated with the length of each quaver. (The offending edit comes right at the end.) What’s happened is that the final quaver of b. 5 [1’ 09”], setting ‘In’, has lost 0.2–0.25s from its end. The expected length can be worked out by what leads immediately up to it. In the preceding bars there’s a pattern set up of longer sixth quavers: have a look at ‘-nen’ of ‘fernen’, then ‘der’, then ‘sich’; the end of the bar seems to be setting up the same ritenuto, but sooner—the quaver before ‘In’ is already as long as the quaver at ‘sich’, though it’s not yet the end of the bar, so the last quaver must be expected to be longer still, probably about 0.6–0.65s, certainly not 0.36. And it sounds as wrong as it looks. This example illustrates just how small a departure from expectations can disrupt the effect of a performance. This seems paradoxical when one considers how wide the range of possible variation is in timing and amplitude at any moment in a performance. Clearly some kinds of changes are artistic and some are disastrous, and it’s not a matter of how great they are but rather of how they fit into their context. Although irregularities in note-length can be extreme and still musical (in the sense of ‘acceptable to musicians’), they have to belong within a trend, and one that is related to the musical structure (by which I mean the notated pitches & durations and the ways in which experienced musicians believe that they interrelate).

¶105 Now we’re in a better position to deal with this question of the extent to which a studio-edited recording can be considered as a performance, one from which we might safely draw conclusions about how performance works. Because of the enormous range of possibilities for convincing performances, even within the style-criteria of one place and time, we know that at any moment in a performance the performer has many different options; and provided that they are reasonably competent, whichever option they take when deciding how to play or sing the next note can be made adequately convincing by what follows. This is why recordings assembled by an editor in a studio from hundreds of different bits of digitally encoded sound work. The producer is a performer and performs the piece with the fragments of the performer’s recorded performances. And a recording made this way is thus every bit as much a musical performance as an unedited recording made in a single take.

¶106 So I don’t believe that we need be as concerned as we sometimes have been about recordings falsifying the performance or the act of performance. It’s nice to hear recordings that faithfully reproduce what a great performer did, but it won’t necessarily be better than a performance edited together in the studio: sometimes it will, sometimes it won’t. 113 What’s lost in recording—the sense of being there as it happens—is lost however the recording was made.

¶107 I don’t, therefore, entirely accept the argument that studio recordings are not performances. Of course, in an obvious and literal sense they are not. They are (nowadays) compiled from bits of many (usually incomplete) performances. Earlier they were performances recorded whole but very imperfectly. But when we listen to a recording we listen not to its construction but to its effect. And so the argument that a recording isn’t a performance (however it was made) misses the point. If it didn't sound like a performance it wouldn't be issued on disc, and if it sounds like one then it is one. What matters is that it’s experienced as a performance. Because it’s the sound that offers the proof of the pudding. Performance sounds, and if it sounds like performance then that’s the appropriate measure of whether it is. Recordings do what performances do well enough that we hear them as performances. This should encourage us to take the details of a recorded performance as the material for studies that seek to explain how music works, even though they may not represent any reality that could exist in a live performance. Provided that they work for us, they work, and that tells us something important about what music is. It's what we perceive as music, not what, looking at the production processes, we might say it 'really' is.


On the early years (pre-electrical) see Walter L. Welch & Leah Brodbeck Stenzel Read, From Tinfoil to Stereo: the acoustic years of the recording industry (Gainseville: University Press of Florida, 1994; rev. ed. of Oliver Read and Walter Welch, From tin foil to Stereo (1976)). A very good brief introduction to the whole period is Peter Copeland, Sound Recordings, (London: The British Library, 1991). A good medium-length survey up to the end of the LP is Roland Gelatt, The Fabulous Phonograph 1877-1977 (2nd rev. ed., London: Cassell, 1977). There are also several excellent treatments of recording and its cultural impact, including: Evan Eisenberg, The Recording Angel: music, records and culture from Aristotle to Zappa (New Haven: Yale University Press,1987 (London: Picador, 1988); 2nd rev. ed. 2005); Michael Chanan, Repeated Takes: a short history of recording and its effects on music (London: Verso, 1995); and Jonathan Sterne, The Audible Past: cultural origins of sound reproduction (Durham, NH: Duke University Press, 2003). Sterne is outstandingly good on the pre-history of sound recording and its cultural influence. Back to context...
Peter Copeland’s ‘Manual of analogue audio restoration techniques’ published posthumously by the British Library online, is a magnificent contribution that appeared just as this book was nearing publication. I am grateful to Nigel Bewley of the BL Sound Archive for allowing me to see three chapters in advance. Back to context...
An important and relevant study (sadly unknown to me when I wrote this chapter) is George Brock-Nannestad's 'The critical approach to sound recordings as musicological sources', Musiikki 1-4 (1989), 423-35. See also his 'Zur Entwicklung einer Quellenkritik bei Schallplattenaufnahmen', Musica 35 (1981), 76-81. Some of the following has been reused in Roger Beardsley and Daniel Leech-Wilkinson, A brief history of recording (published online by the Centre for the History and Analysis of Recorded Music, 2006). Back to context...
Copeland (1991), 7. Cf. Chanan (1995), 2. Back to context...
Gelatt (1977), 32. Back to context...
Welch & Read (1994), 32 Back to context...
I owe this description of Berliner’s process, and many other details here, to Roger Beardsley’s meticulous editing of an earlier draft. For a full description of the process, of which this is a simplification, see Beardsley’s online essay ‘Waxes, shells and stampers’. Back to context...
On some of the difficulties see F[red] W. Gaisberg, Music on Record (London: Hale, 1946), 104-5, 109; also Miller & Boar (1982), 44-55. For a detailed study of Fred Gaisberg see Jerrold Northrop Moore, Sound Revolutions: a biography of Fred Gaisberg, founding father of commercial sound recording (London: Sanctuary, 1999). Back to context...
Tuesday, Oct 05, 1909, p. 11; issue 39082; col. D. More examples are quoted in Welch & Burt (1994), 146; and (on Edison’s Diamond Discs demonstrations with Maggie Teyte) Eisenberg (1987/2005), 90-1. Warm thanks to Alison Bullock for the Times report. Back to context...
Gelatt, plate 2, reproduces a delightful advertisement for Edison’s phonograph which boasts that ‘It will Talk, Sing, Laugh, Crow, Whistle, Repeat Cornet Solos, imitating the Human Voice, enunciating and pronouncing every word perfectly, IN EVERY KNOWN LANGUAGE.’ Back to context...
My main sources for what follows are two fascinating papers by George Brock-Nannestad: ’Authenticity in the Reconstruction of Historical Disc Recording Sessions’, Preprint of paper delivered to the Audio Engineering Society’s 105th convention, 1998 (AES Preprint 4829); and ‘The Objective Basis for the Production of High Quality Transfers from Pre‑1925 Sound Recordings’, Preprint of paper delivered to the Audio engineering Society’s 103rd convention, 1997 (AES Preprint 4610). Back to context...
Gaisberg (1946), 52-65. The technicians were then known, very fairly, as ‘experts’. There are rich details and fascinating photographs of Gaisberg’s travels in Moore (1999) passim, some photographs also in Miller & Boar (1982), 45-55, and a less detailed description of some of their procedures in Peter Martland, Since Records Began: EMI, the first 100 years (London: Batsford, 1997) 44. Back to context...
Stanley Chapple, ‘In the Recording Studio’, ed. Roger Wimbush, The Gramophone Jubilee Book (Harrow: Gramophone, 1973), 64-70, reprinting an article originally published in the December, 1928 issue of The Gramophone. The diagram is reproduced in Brock-Nannestad (1998) fig 2. For an alternative, American layout drawn by Fred Gaisberg in 1907, with more space between the players and three recording horns, see Brock-Nannestad, (1997), fig. 8, p. 37. Back to context...
A description of the in-house Victor orchestra, its instrumentation and rates of pay, is quoted in Brock-Nannestad (1997), 27-9. Back to context...
For some pianists’ memories see Day (2000), 10. Back to context...
Gaisberg recalls doing this to Patti: Gaisberg (1946), 86-7. Back to context...
Side lengths depended also on how loud the music was: loud music produced wavy grooves which took up more space. The engineers did have the ability to vary the lathe’s transit speed resulting in closer groove spacing and longer playing time. In the early 1920s, HMV experimented with sides lasting over 8 minutes, but for the most part side-lengths of commercial recordings remained similar through the latter part of the 78-era. Timings relative to groove sizes are listed in ‘Columbia Graphophone Company Limited (London): Recording Expedition Instructions’, p. 18. I’m very grateful to Roger Beardsley for a copy of this rare pamphlet, now available online through CHARM at http://www.charm.kcl.ac.uk/history/p20_4_1_2.html. Back to context...
Philip 2004, 36-8, offers some cautionary tales and shows why they need to be checked against the recordings. A hilarious example, however, is Lev Sibirjakov’s recording of Schubert's ‘Der Leiermann’ which he sings extremely slowly on a three-minute disc, forcing the pianist to race through the interludes. Curiously it remains a magnificent performance. (Matrix 14815 b, recorded by Fred Gaisberg on 11 September 1910, issued on HMV 4-22135. CD reissue on ‘Schubert Lieder on Record, 1898-1952’ (EMI 5 66150 2), disc 1, track 22.) Back to context...
This point is made in more detail and with examples in Mark Katz, Capturing Sound: how technology has changed music (Berkeley: University of California Press, 2004), 34-5. Back to context...
For contrasting period views of this practice see Day (2000), 8. Back to context...
In this section I am particularly indebted to Brock-Nannestad (1997), Appendix 1, 20-22. There is a very fascinating 1942 film of the whole process as operated at RCA Victor downloadable from http://www.archive.org/details/CommandP1942. I am most grateful to Roger Beardsley for drawing my attention to it. Back to context...
Brock-Nannestad (1997), 10. Back to context...
The processes for this are outlined below. Back to context...
For a general survey of the problems see Roger Beardsley, ‘Is it or isn’t it...?’, International Classical Record Collector 5 (1999), 62-5, also available online at http://www.charm.kcl.ac.uk/history/p20_4_3.html. Back to context...
Cylinders continued to be used by ethnographers in the field until the early 1940s, due to the machine’s portability and its not needing an electricity supply. Back to context...
On the comparison between cylinders and discs see Welch & Burt (1994), 111-26. Back to context...
Exactly the same happened with videotape in the 1970s. Back to context...
Gaisberg (1946), 78. Back to context...
Copeland (1991), 10. Back to context...
Transferred by Karsten Lehl from Deutsche Grammophon 043370, matrix 341 av, (rec. 1921) issued on Nippon Polydor 60010, Polydor 76418, and Polydor 85297. See also Barbara F. Stone, ‘Discography of Mme. Frieda Hempel’, The Record Collector 10 (1955), 65-71. Back to context...
More will be said about this below. See also Roger Beardsley’s more detailed explanations at http://www.charm.kcl.ac.uk/history/p20_4_4.html. Back to context...
Transferred by Karsten Lehl from Odeon 36034, matrix xP 1317 (rec. April 1905), acc. Mr. Pickaert (Mustel harmonium). Back to context...
For another curious keyboard sound see John Hunt’s 1933 recording of Chopin on a ‘Neo-Bechstein’, which turns out to be an early electric piano. Matrix 2B4545-1 (rec. 6 April 1933), issued on HMV C 2567. There’s a transfer available for download either via the CHARM discography or transfer project. Back to context...
For the following summary see esp. George Brock-Nannestad, ‘What are the sources of the noises we remove?’ Paper delivered to the 20th Annual conference of the Audio Engineering Society, 2001 (AES Paper 1960). Back to context...
See also Brock-Nannestad (1997), 21. Back to context...
Though reinforcement of the bass line continued on occasion for the sake of listeners still using acoustic equipment. Back to context...
Surface noise was reduced, especially by HMV (Victor and US Columbia issues tended to be noisier), but it reappears easily on discs that have been stored in damp conditions, so surviving copies can be misleading. (My thanks to Roger Beardsley for the information in this and the previous note.) Back to context...
Transferred by Karsten Lehl from Columbia LFX 250, matrix CLX 1872-1. Back to context...
Students interested in working on this performance may find it useful to compare it with others of ‘Die junge Nonne’ available via Daniel Leech-Wilkinson, ‘Sound and meaning in recordings of Schubert's “Die junge Nonne”’, Musicae Scientiae, 11 (2007), 209-36. Martinelli’s irregular vibrato is discussed in chapter 8 below (using Data File 9 (sv file)). Back to context...
Brock-Nannestad, (1997), 13 n. 2, 16 n. 23. Back to context...
Renee Timmers used this approach in testing the influence of surface noise on people’s perception of performacne quality. ‘Perception of music performance on historical and modern commercial recordings’, Journal of the Acoustical Society of America 122 (2007), 2872-80. Back to context...
[Unsigned], ‘Gramophone music: encouragement of close listening’, The Times, 42963, 23 (February 1922), p.10 col.B. I am most grateful to Nick Morgan for sending me the text of this article. Back to context...
For practical advice on using gramophones see Douglas Lorimer, ‘Reproducing 78s: the old-fashioned way’, International Classical Record Collector no. 6 (Autumn 1996), 62-71. Back to context...
Day (2000), 7, calculates that the 1913 set of discs containing Beethoven's 5th conducted by Nikisch cost 50% more than the average weekly wage. I have used Alan Stanier’s tables of equivalences at http://privatewww.essex.ac.uk/~alan/family/N-Money.html. Some discs cost even more: Ruth Edge and Leonard Petts, ‘Keeping tabs on Tamagno’, Classic Record Collector no. 50 (2007), 40-5, esp. 42. Back to context...
For a useful general introduction by a well-known transfer engineer see Roger Beardsley’s http://www.charm.kcl.ac.uk/history/p20_4_4.html. Back to context...
For one expert’s approach see David Patmore and Tully Potter’s interview with Mark Obert-Thorn, who has done many of the best transfers on Naxos, in Classic Record Collector No. 47, Winter 2006, 64-7. For his views ten years earlier see Arthur S. Pfeffer, ‘Mark Obert-Thorn: the moderate interventionist’, International Classical Record Collector no. 6 (Autumn 1996) 48-56. On Ward Marston, another giant among transfer specialists, see Arthur S. Pfeffer, ‘Ward Marston: the compleat transfer engineer’, International Classical Record Collector no. 3 (November 1995), 63-8. For the view of Andrew Walter, who has done many of EMI’s reissues, see Andrew Walter, ‘Back to the future: a personal insight into the remastering of EMI’s back catalogue’, International Classical Record Collector no. 10 (Summer 1997), 46-51. Back to context...
For competing solutions see http://www.keithmonks-rcm.co.uk/, http://audio-restoration.com/monks5.php, http://discdoc.com/. For concerns see the debate with subject heading ‘Highly unorthodox cleaning methods for LP's... ’ in the Association of Recorded Sound Collections list archive for May 2004: http://palimpsest.stanford.edu/byform/mailing-lists/arsclist/. Back to context...
http://www.elpj.com/about/ Back to context...
See http://perso.wanadoo.fr/socket/warcheophone.htm. Back to context...
See especially the procedures used by Nimbus, described at http://www.wyastone.co.uk/nrl/pv_intro.html. Back to context...
Optical scanning of the groove removes the need to select a stylus, preserving all the information there is, but turning that information into sound is another matter, and one that has yet to be mastered to everyone’s satisfaction. Back to context...
Equalisation can also be done later in the digital domain, and there are strong feelings among specialists about the merits of the two approaches. Back to context...
See, for example, James R. Powell, Jr., and Randall G. Stehle, Playback Equalizer Settings for 78rpm Recordings (2nd rev. ed., Portage, Mich., Gramophone Adventures, 2002) which, precisely because it attempts to standardise solutions to a wildly non-standardised practice, is not to be preferred to the judgement of a musical ear. Back to context...
Though electronic filters are not always as perfect as we might wish. Back to context...
For a more detailed description of the types of noise coming off 78s see Roger Beardsley’s commentary at http://www.charm.kcl.ac.uk/history/p20_4_4.html, under ‘Record Noise’. Back to context...
This research on non-contact surface scanning of cylinders is in progress at the University of Southampton: http://www.soton.ac.uk/ses/research/projects/project016.html, http://www.sesnet.soton.ac.uk/archivesound/publications/. Back to context...
HMV DB 7648/51. These extracts are from DB 7648, matrix 2B6821-2 (end); DB 7649, mat. 2B6822-1A (start & end); DB 7650, mat. 2B6823-1 (start & end); DB 7651, mat. 2B 6824-2 (start). Back to context...
See also Philip 2004, 35-6. Back to context...
An easy comparison to make, since the discs are always available, is between different transfers of the Menuhin/Elgar recording of Elgar’s violin concerto, recorded on HMV in 1932. Andrew Walter’s transfer for EMI, in particular, sounds significantly different from Mark Obert-Thorn’s for Naxos, brighter but also more muddled and more dominated by the brass. (EMI Classics 7243 5 66979 2, Naxos 8.110902.) Back to context...
The argument is well-made in Plack (2008), ch. 1 p. 4, chiding an earlier version of this chapter. Back to context...
The same colours were used for the UK 12-inch Series C, as well as the 10-inch B and equivalent series issued in other territories such as EG, EH, AM, AN etc. All were collectively known as ‘the plum label’. Back to context...
For the story see Gelatt (1977), 107-9. Back to context...
The purchase tax stamp came in during the Second War. The codes are ‘NT’ – no tax (small runs and tests), ‘DT’ – double tax, ‘TT’ – triple tax. The information is often in the pressing instead, right in the centre where the clamp was fitted. Back to context...
Alan Kelly, HMV Matrix Series: MAT 201 (CDR privately published by Alan Kelly.) Back to context...
Different companies had different procedures and different means of identifying takes. Some didn’t do so at all. I use a Gramophone Co. disc as my example because they were so ubiquitous. The Gramophone Co. also went to unusual lengths to use well-ordered sequences of prefixes and suffixes surrounding their matrix numbers, and these generally indicate the local office responsible for the recording as well, in some series, as the identity of the recording engineer. Consequently they can be extremely informative. For details of all the HMV matrix series see Alan Kelly. ‘General Introduction: The structure of the Gramophone Company and its output, HMV and Zonophone: analysis of papers and computer files’, and the introductions to the individual matrix series in Kelly’s ongoing publication of the Gramophone Co. discography on CDR (available from Alan Kelly). Summaries will be available through the CHARM online discography via www.charm.kcl.ac.uk. Back to context...
Metals, let alone unissued metals, rarely survive today, though there are some fine examples still in the EMI Archive. An interesting collection, including previously unknown takes, was issued in 2007 by Historic Masters, pressed in vinyl from the metals preserved at Deutsche Grammophon (originally a branch of the Gramophone Co.) in Hamburg: ‘Francesco Tamagno 12” recordings complete’, HM FT1/7. Back to context...
Frank Andrews and Ernie Bayly, Catalogue of HMV Series “B” Records (Wells-next-the-Sea: City of London Phonograph and Gramophone Society, 2000). Back to context...
Roger Beardsley, who has contributed most of these paragraphs on early transfers, comments: ‘All these were made by the Gramophone Co. affiliate Victor in the USA, but pressed by HMV as part of the matrix exchange agreement. When Victor could no longer supply stampers for popular titles, a mechanical transfer was made. Interestingly, the mechanical dubbing equipment used by HMV was made by Victor since HMV’s own machine was not good enough. However, it was acknowledged that the HMV engineers got the best results from the system.’ Back to context...
For a fine picture of the wear test in action (though from the LP era) see Martland (1997), 140. Back to context...
For example CWR 1234-II would become CWR 1234-II T1. Back to context...
Björling’s ‘Di quella pira’ (Il Trovatore) recorded in Stockholm was originally a 0SB prefix recording. When transferred for issue, it was allocated a 0EA (English Head Office) series number. Back to context...
On the business aspects of the early duplication process: Raymond R. Wile, ‘Duplicates in the nineties and the National Phonograph Company’s bloc numbered series’, ARSC Journal 32 (2001), 175-216. Back to context...
Failing a good discography, two useful starting points for dates are Eddie Shaw, Dates About all Those English Seventy-eights, Part 1 Commercial, (London: Shaw, 1994; 2nd rev ed 1995); and Steven C. Barr, The Almost Complete 78 rpm Record Dating Guide (II) (Huntington Beach, CA.: Yesterday Once Again, 1992). Back to context...
Trezise (forthcoming 2009). Back to context...
Rolls are nowadays increasingly distributed as MIDI files. See http://www.pianola.co.nz/ for one useful source. Back to context...
The fullest online source of information at the moment (November 2007) is http://www.pianola.org. Back to context...
Information supplied by Denis Hall in November 2007, updating his essay, ‘The Reproducing Piano and its Recordings’, booklet with Ignace Jan Paderewski in Recital, CD catalogue no. Aeolia 2002. I am extremely grateful to Mr Hall for his very detailed comments on an earlier draft. Numerical catalogues of classical music issues can be found, unfortunately without dates of issue, at http://www.rprf.org/Resources.htm. Back to context...
On the Ampico patents, beginning in 1908, see Rex Lawson, ‘Cleaning the windows of time: an examination of recording methods for the Ampico re-enacting piano’, The Pianola Journal 12 (1999), 13-44. Back to context...
For a slightly more sympathetic account of piano roll reproduction, based on period comments, see Day (2000), 12-6. Back to context...
For a stronger statement of this problem see the very useful discussion of reproducing pianos in Philip (2004), 30-4. Back to context...
Grieg, Welte roll 1276 (1906), G&T 35517 (1903); Pugno, Welte 543 (1905). G&T 35504/5 (1903). The roll and disc transfers are by Denis Hall, to whom I am extremely grateful for sharing them with me and for allowing me to copy them here. Back to context...
Eliot Levin, ‘Beyond the Groove’, International Piano 8 (2004), issue 32, 42-59, and issue 33, 38-53, this passage at 39. (My thanks to Nick Morgan for this reference.) The recordings were issued simultaneously as ‘Piano Rolls & Discs—Selected Comparisons’, Symposium 1211. The Grieg is track 13, transferred from Odeon ‘Historic series’ 38266. Back to context...
Compare especially 1903 32”-53”, 1930 31”-50”, modern 34”-55”. Note that timings are from the start of the music, not the start of the track. Back to context...
For a published sample of Hall’s work see his ‘Ignace Jan Paderewski’, Aeolia [CD no.] 2002. Back to context...
‘Ignaz Jan Paderewski, Xaver Scharwenka: Chopin’, Nimbus NI 8816, track 11 (music lasts 4’ 36.5”). ‘Ignace Jan Paderewski’, Aeolia 2002, track 2 (music lasts 4’28”). Back to context...
Denis Hall informs me that the Duo-Art push-up used by Nimbus was fitted with a late version of the take-up spool, larger than previously, which would account for the slowing. Back to context...
For details and discussion see Roy Howat, ‘The New Debussy Edition: approaches and techniques’, Studies in Music (1985), 94-113; ‘Debussy and Welte’, The Pianola Journal 7 (1994), 3-18; and Howat (1997). For a slightly different view see José A. Bowen, ‘Finding the music in musicology: performance history and musical works’, in ed. Nicholas Cook and Mark Everist, Rethinking Music (Oxford University Press, 1999), 424-51 at 438-41. Back to context...
Ed. Roy Howat, Oeuvres complètes de Debussy, Série I, volume 5: Préludes (Paris: Durand, 1985), 42-7 & 167. Back to context...
See, for example, Walter Gieseking’s recording on HMV Treasury RLS 752, disc 1, side 2, band 1 (rec. 1953). It may be worth adding, as a further caution, that I have heard four recordings of this roll, all at different speeds and different pitches: ML4291; Telefunken GMA 65, side 1, band 6; ‘Debussy: early recordings by the composer’, Condon 690-07-011 (issued 1992); and ‘Claude Debussy: the composer as pianist’, Pierian 0001 (issued 2000), track 2. Back to context...
See also Leech-Wilkinson in Cook et al. (2009). Back to context...
Carl Reinecke, Mozart Piano Concerto K537, 2nd movt. (extract). Transferred from Welte 237 (1905) by Denis Hall and used by kind permission. Back to context...
Jennifer Doctor, The BBC and Ultra-Modern Music, 1922-1936 (Cambridge University Press, 1999); Humphrey Carpenter, The Envy of the World: fifty years of the BBC Third Programme and Radio 3 (London: Weidenfeld & Nicolson, 1996). Back to context...
The surviving multiple copies are now in the King’s Sound Archive at King’s College London. Back to context...
My main source for this paragraph is Copeland (1991), 21-30, where further details will be found. Back to context...
Bruckner, 8th Symphony: Preussischer Staatskapelle cond. Herbert von Karajan, Koch Swann 3-1448-2 (Berlin, 28 June (mts. 2-3), 29 September (mt. 4) 1944). Back to context...
There is a photo of the first model of EMI tape machine, the BTR1, in Martland (1997), 152. Back to context...
Walter Legge, continuing from the passage quoted earlier about Gaisberg’s view of records as sound photographs, says, ‘My ideas were different. It was my aim to make records that would set standards by which public performances and the artists of the future would be judged―to leave behind a large series of examples of the best performances of my epoch.’ (Schwarzkopf (2002), 16) And in that last clause we can see how in Legge’s mind the difference between performances and recordings was already blurred. For a good introduction to Legge as a record producer see Day (2000), 40-3. Back to context...
On the notion of liveness modified by the studio production of popular music recordings, see Philip Auslander, Liveness: performance in a mediatized culture (Abingdon: Routledge, 1999). Back to context...
John Culshaw, Ring Resounding: the recording of Der Ring des Nibelungen (London: Secker & Warburg, 1967). Back to context...
For an example see Robert Philip’s interesting discussion of a recording session with producer Andrew Keener, Philip (2004), 50-62, especially the example on p. 55 reproducing a page from Keener’s annotated score indicating the takes to be used. Back to context...
The following table has already appeared in Daniel Leech-Wilkinson, ‘Expressive gestures in Schubert singing on record’, Nordisk Estetisk Tidskrift 33/34 (2006), 50-70. Back to context...
Godlovitch (1998), 14 and 128, see also 26-7. Back to context...
On the Musical Performance on Record discussion list: mus-perf-rec@jiscmail.ac.uk, 9 December 2002. Back to context...
Cf. Cook (2003). Back to context...
Albin Zak in Cook et al. (forthcoming 2009). Back to context...
Davies (2001), 304-5. For discussion of the disadvantages of repeatability see Chanan (1995), 118-9. Back to context...
Davies (2001), 309. Back to context...
Davies (2001), 296-7. Back to context...
Philip (2004), 50-62. Incidentally, many further studies of producers and recording sessions are needed. Back to context...
Jonathan Freeman-Attwood told me how perplexed he was when producing the late Strauss wind Sonatina no.1 to be faced with a piece in which no span could easily be realised in the recording: the work seemed not to offer, compositionally, any obvious incremental solutions or musical 'sweep'. It seemed to him that his role as a successful producer required him to derive a new kind of structure from the raw material rather than delivering a series of takes recognisable by the artist from a live performance. (I am grateful to him for allowing me to use this anecdote.) Back to context...
DG 419 237-2, track 9, 0’ 57”B1’ 10” Back to context...
Leech-Wilkinson (2006a), 58. The following discussion derives from that article, 58-9. Back to context...
Glenn Gould: ‘good splices build good lines, and it shouldn’t much matter if one uses a splice every two seconds or none for an hour so long as the result appears to be a coherent whole.’ Gould rather spoils it, however, in his continuation, which only emphasises his determination to seem mechanistic: ‘After all, if one buys a new car, it doesn’t really matter how many assembly-line hands are involved in its production. The more the better, really, insofar as they can help to ensure the security of its operation.’ (Glenn Gould, ‘Music and technology’, originally published in Piano Quarterly (Winter 1974-5), reprinted in ed. Tim Page, The Glenn Gould Reader (New York: Knopf, 1984), 356.) My thanks to Alex Robinson for this reference. In work on one of my MMus course units at King’s College in 2004, Robinson also came up with an interesting comparison of two performances conducted by Frans Brüggen of Schubert’s 9th Symphony , one live, one studio. He asked listeners which they thought was which; most were wrong. Their judgement depended on the vividness of the recorded sound, and in fact the studio sound was notably livelier than the sound from the concert. This only emphasises, as Robinson pointed out, that our preconceptions about the differences are more important than the differences themselves. Back to context...