This chapter introduces sound recordings made without the help of any electronic amplification. Such recordings were dominant before 1925, and the technology was still being used in remote locations without electricity until at least the late 1940s.

To oversimplify somewhat, the performers projected their sounds into a conical horn, which concentrated the sound waves onto a small area, and vibrated a tool which cut a groove in wax. In a few cases we know that professional “experts” utilised a few decibels of pneumatic amplification, and I shall take a brief look at this in section 12.25. But the vast majority used simpler apparatus which is shown diagrammatically in section 12.3. Virtually everything I am talking about will be the result of an “unamplified” performance.

We are currently at the leading edge of technology in recovering “the original sound” from acoustic recordings, and significant amounts of experimental work are taking place as I write. In the mid-1990s this author wrote a series of five articles on the subject, which (since it forms the first reference for nearly everything which follows) I shall cite now to get it out of the way! (Peter Copeland: “Acoustic Recordings” (series of articles), Sheffield: “The Historic Record” (magazine), nos. 32 (July 1994) to 36 (July 1995)).

I am very grateful to the Editor of “Historic Record”, Jack Wrigley, for permission to make use of substantial sections of those articles; and the resulting correspondence continues to shed much light. In that series, I attempted to outline how one might achieve the correct equalisation of acoustic recordings, using much the same principles as I described for electrical records in Chapter 6. Since then, I have had the privilege of meeting one or two other operators who accept that faithful reproduction of acoustic recordings might be possible, and new ideas and projects have been started (and sometimes abandoned).

With the current state of the art, it is my duty to report that it is impossible exactly to prescribe the correct treatment for an acoustic recording. But I can foresee that intelligent computer software will eventually replace the human hearing process with its implied subjective judgements. So this chapter outlines the current state of the art in anticipation of further developments; and, to save space, it does not describe the numerous experiments to prove or disprove the theories. (Many such descriptions were included in the above Reference, and will continue to be made available to serious enquirers).

Before 1925, recording engineers were called “experts”. Discographical studies show that in a surprising number of cases (perhaps one in four), we know the identity of the expert concerned. It has frequently been said that experts were individuals with techniques all their own, and therefore we cannot expect to reverse-engineer their products today. Frankly, I think this is rubbish - contemporary image-building, in fact. We can and we must, otherwise we will never learn the full story.

But I agree it might not be ethical to compensate for acoustic recording machinery. I am trespassing upon the subject of the next chapter, but contemporary experts were well aware of the deficiencies of their equipment, and managed their recording-sessions to give the optimum musical result. When we reverse-engineer the recordings, more often than not we get a horribly un-musical result. But I also hope this chapter will demonstrate how and why defects occurred, so we may at least respond intelligently and sympathetically to those recording artists whose legacy seems disappointing today.

There is also a great deal of evidence that professional artists modified their performances before the recording-horn. To deal with one consideration: I remember I was almost overwhelmed by the amount of sound which came back at me when I first spoke into such a horn. A vocalist might notice his A-flats were reflected back strongly, for example; so he would deliver his A-flats with reduced intensity (and possibly modified tone-colour).

The experts were forced to do everything they could to get a loud enough signal above the surface-noise. Yet complete symphony orchestras, concertos, and choirs were attempted - or, to use a better word, simulated. For there is no doubt that the less-than-perfect reproducing equipment of the time enabled artists and experts to get away with faking sounds. This therefore raises the moral dilemma of whether we should try to reproduce the original sounds.

The earliest acoustic recordings were made with machinery so insensitive that there was hardly ever any intended sound audible on playback. This remains generally true today, even with modern restoration technology. Pioneer experts concentrated on making the performers operate as loudly as possible; so early recording locations were chosen more for their advantages in not annoying the neighbours than any other reason. Secondary considerations might be warmth (to help keep the wax soft enough to cut), and height (to allow a reasonable fall for the weight-powered motor). Thus early recording studios tended to be at the top of buildings. They were often lit by skylights. They happened to provide lighting similar to the studios used for photography, painting, and sculpture, which made performers slightly more at ease; and it was easy to close the skylights during a “take”, and then reopen them to let the perspiration dry out.

From a very early date, the advantages of sound reflecting surfaces were appreciated. A reflective surface behind the performer could help drive more sound into the horn to operate the recalcitrant mechanism. (The technique of the “hard screen” is still used in professional recording studios, of course). As acoustic recording technology became better at dealing with groups of artists, it was found that the average stuffed Victorian or Edwardian room was a hindrance, so the nature of the surrounding furnishings radically changed. Hard bare walls and floors and ceilings were introduced. I have no doubt that experts tried different sizes of rooms; but anything greater than about five meters square and three meters high was found to be actually disadvantageous.

A bigger room might have given musicians and music-hall artists a more familiar environment, but the acoustics would not have been picked up by the horn. A much smaller room with reflective walls would be like a jazz band in a telephone box - extremely loud, but quite impossible for satisfactory performances. A slightly larger room would suffer, because the bodies in it would mop up most of the sound reflections; whereas the room I have just described hit an acceptable balance between reflected sound being powerful enough to augment the signal, while the bodies in the room were insufficient to mop up this advantage. Someone once described it as “making music in a sardine tin,” and when we make an objective copy of almost any acoustic recording today, I am struck by how accurate this description is.

I can think of only one acoustic recording where the surrounding environment played a significant part - the recording of the Lord Mayor of London, Sir Charles Wakefield, in 1916. (Ref. 2) In this recording it is just possible to hear a long reverberation-time surrounding the speech, which must have been overwhelming on location. Apparently the Gramophone Company’s experts thought this was so abnormal that they took the precaution of insisting that “Recorded in Mansion House” was printed on the label. As far as I can ascertain, this was the first case where the location of a commercial recording was publicly given, although we now know there were many previous records made away from the usual studio premises.

There is yet another reason for writing this chapter, which we haven’t come across before. A large number of acoustically recorded discs and cylinders survive which document the dialects and music of extinct tribes and peoples. Nearly all experienced sound operators are familiar with the special sound of acoustic recordings, and can mentally compensate for the distortions caused by the recording horn. But less experienced listeners, who are used to the sound of microphones, sometimes find it difficult to adapt. To avoid misleading these people, it is advisable to give them processed service copies with the worst distortions reduced, even though this may not be to the usual degree of precision. It is particularly important for the non-professional performers on such records, who did not modify their performances to suit the machinery. The relationships between different vowel sounds for example, and between vowels and consonants, may be seriously misunderstood.

Most acoustic recordings were made by apparatus shown diagrammatically below. Sounds would be emitted by a performer at X; this represents the position of the principal vocalist or soloist. There might be several performers at different positions around him, behind him, and beneath him; but the recording equipment was so insensitive that anyone not in the optimum position X would be relegated to a distant balance in the recording.

The sounds would be collected by means of a horn H, usually in the shape known to students of Euclid as “the frustrum of a right cone with a semi-included angle between eight and eleven degrees.” (I shall call this “conical” for short !). The purpose was to concentrate sound from a large area at the mouth of the horn upon a much smaller area. Quite often several horns were provided for several principal performers. Sometimes a parallel-sided tube T followed, and sometimes a cavity C serving the function of an acoustic impedance-matching transformer, and sometimes both.

The sound waves then passed to a unit known as the “soundbox” or “recorder box”, where they vibrated a flat circular diaphragm D mounted between compliant surrounds. Both sides of the diaphragm were exposed to the atmosphere, and the difference in pressure between the two sides caused it to move. (A physical screen between the artists and the expert would modify this situation, of course). The diaphragm’s centre, where the maximum movement normally occurred, was usually connected to a lever system L which carried the vibrations to a cutting stylus S. These vibrations modulated a spiral groove cut into a solid disc or cylinder made of wax. Some sort of lever system seems always to have been used, even on hill-and-dale recordings. I mention this because in theory it would be possible to couple a stylus directly to the diaphragm in hill-and-dale work; but a linkage permitted the stylus to be changed more easily, and allowed for variations in the level of the wax.

There were many variations on the above scheme, but it accounts for more than half all acoustic recordings, and the variations do not greatly affect the results. I shall consider the components of the “standard” system first, leaving the variations until section 12.25 when they will make more sense to you.

As we consider each component in turn, we shall learn where our understanding has reached today. Virtually all this understanding will be in the frequency response of the recording machine. I don’t want to befuddle you with mathematics, so I shall describe the effects of each component in words, and give references to scientific papers for further study.

There is some evidence that harmonic distortion occurred, but most of this “evidence” seems to come from writers who did not distinguish between the “blasting” which occurred during recording and that due to reproduction. With the exception of hill-and-dale recordings which were not transferred from a “master” (and which therefore did not have their even-harmonic distortion cancelled), I have rarely heard significant distortion recorded into the groove (see Section 4.15). Although such distortions always happened to a limited extent, they nearly always seem to be outweighed by reproduction distortions, and I shall not consider them further.

This is the point to remind you that if the signal-to-noise ratio of the reproduced recording is good, it is more likely to result in hi-fi sound recovery. But it is particularly important in this chapter, because the insensitivity of the recording-machine resulted in large amounts of noise at both ends of the frequency range. We must therefore apply most of the noise reduction principles discussed in Chapter 4. Also the frequency response had “steep slopes.” When we compensate for the frequency response, we also “colour” the background noise; and this may even corrupt the process of determining the right equalisation in the first place, as we shall see.

Drawn from various written sources, here are details of some of the compromises made during commercial recording-sessions. The piano always caused difficulties, because the sounds came from a sounding-board with a large area. In early days it was always an upright, which was also easier to get close to the horn. The back was taken off so there would be nothing between the sounding-board and the horn, and the pianist was instructed to play “double forte” throughout.

Sometimes the piano’s hammers were replaced with something harder than felt, because of the diaphragm’s insensitivity to transient sounds. There was little improvement in the diaphragm’s sensitivity after about 1906. But better records (with less surface noise) and better gramophones (with greater acoustic efficiency) meant that the apparent sensitivity increased, and by 1910 classical pianists could be recorded playing relatively unmodified grand pianos.

Orchestras were also difficult. Stringed instruments always gave problems, because they had less power than the wind section, and their operators needed more “elbow-room.” The Stroh Violin was commonly employed; this was a violin fitted with a soundbox and horn so the upper partials of violin tone could be directed at the recording machine. Yet even with this artificial clarity, contemporary string players were encouraged to exaggerate the portamento and vibrato features of their instruments to help distinguish them from the wind. As bass notes were recorded very weakly, the bass line was often given to the tuba, which had more partials; these could be recognised on playback as constituting a “bass line.”

So brass or wind bands would be preferred for vocal accompaniments. Stage-management was needed during a “take” to give soloists uninterrupted access. Vocalists had to bob down out of the way during instrumental breaks. Because of the bass-cut difficulty, those unfamiliar with recording technique had to be man-handled by the recording director to bring them close on low notes and push them away on high notes.

A healthy amount of reverberation will help musicians, because it helps them to stabilise their sense of pitch. It is a vital part of every trained singer’s singing-lessons that they should be able to judge the pitch of their singing by a combination of three techniques: voice-production method, which ensures the pitch is as perfect as possible before the note starts; hearing the sounds through their own heads to provide a “tight feedback loop”; and hearing sounds reverberating back from around them to provide longer-term stability and match what other musicians are doing. The sense of pitch is not directly related to frequency. It varies with volume in a complex way, and the trained singer must balance these three pieces of evidence. An untrained one will rely almost completely upon the third piece; hence the phenomenon of “singing in the bath.” (I suppose I had better explain this remark, now that bathrooms usually have thick carpets and double-glazing. In the first half of the twentieth century, bathrooms were bare places, with reverberation-times longer than any other room in a private house; so that is where aspiring singers would try their voices).

The most notorious example of someone upset by wrong acoustics was Amelita Galli-Curci, who found the relatively small studio and the loud accompaniments of the Victor Company’s acoustic recording studio very difficult. She was frequently sharp in pitch. History doesn’t record the “takes” which never reached the processing-bath, nor the trial-recordings which must have been made so she could adapt to the alien conditions; but the surviving evidence sometimes shows twenty or thirty processed takes of each song. Galli-Curci was a celebrity for whom this was worthwhile; but this is one explanation why books like Brian Rust’s “Discography of Historic Records on Cylinders and 78s” and “The Complete Entertainment Discography” are filled with important artists who made test-recordings and nothing more.

Although I have little written evidence to support my next remarks, I strongly believe that when the medium of sound recording began to dominate over sales of sheet-music, techniques in instrumentation and song-writing changed. The popular song became a “three-minute sound-byte” until liberated by the 45rpm “disco single” in 1977, for example. But, more fundamentally, the acoustic recording process was unable to handle both low frequency and high frequency sounds, so the “bass and drums section” of bands before the 1930s would have been impossible to record effectively.. So the rhythm was also forced to be in the middle of the range, and it had to be a simple rhythm. Thus it was given to the banjo and/or piano, and fox-trots and waltzes dominated. I have even seen it written that, in acoustic days, “the bass drum was banned from the recording studio”; but I can think of one or two exceptions. Modern reproduction methods are beginning to reveal such broken rules.

If our aim is simply to “reproduce the original sound,” it may reasonably be asked why we do not simply measure the performance of surviving acoustic recording equipment to quantify it. There are several reasons.

• Firstly, contemporary amateurs frequently modified phonographs (or accidentally damaged them) in such a way that the performance would be very significantly affected. In the professional field, only one complete machine survives at EMI Hayes (Peter Ford, “History of Sound Recording,” Recorded Sound No. 22, p. 228). I owe a deep debt of gratitude to the EMI archivist, Mrs. Ruth Edge, for allowing me to make a very close visual inspection of the equipment, and my dependence will be clear in later sections where its description overshadows everything else.
• Secondly, the performance of all such equipment depended critically upon the properties of perishable materials such as string and rubber, which have altered with time.
• Thirdly, some experts used their own personal soundboxes and recording horns which were their own trade-secrets. Indeed, we do not always know which expert made which recordings, let alone which equipment.
• Fourthly, the way the equipment was used - the placing of artists relative to the horn - also makes rigorous measurement largely irrelevant.
• Finally, some measurements have actually been done, but the results are so complex that full understanding cannot be gained.

Because we generally know little about the equipment, most successful work has proceeded on the basis of getting what we can from the record itself. Yet we must not ignore the few exceptions to this idea. These fall into two classes.

One is where we know that the same equipment was used for two or more recordings (or sessions). This enables us to take the common features of both and use them as evidence of the machine which made them. For example, Brock-Nannestad used a particular horn resonance to distinguish between Patti singing a song transposed into another key, and the speed of the turntable being altered.

The other is extremely important. The Gramophone Company of Great Britain used standardised equipment for all its recording experts from 1921. Works of music longer than the four-minute limit were featured about this time, and standardised equipment permitted any one side of a multi-side set to be retaken at a later date with a quality which matched preceding and succeeding sides. George Brock-Nannestad has explained how the US Victor Company (who had a majority shareholding in Gramophone) were unhappy with their matrixes imported from Europe, so they sent one of their recording experts (Raymond Sooy) to reorganise the recording operation (Ref. 1). From that date, the Gramophone Company's “Artists’ Sheets” documented the equipment actually used for each take in coded form.

The “Artists’ Sheets” show both published and unpublished recordings made by the Gramophone Company between 15th March 1921 and approximately February 1931, but they exclude recordings for the Zonophone label. They may be consulted at the British Library Sound Archive. British and Red-label International recordings are on microfilms 360 to 362; Vienna and points east, 385-6; Italy, 386-7; Spain, 387-8; France 388-9; other countries, 390-1.

The acoustic sessions are distinguished by not having a stamped triangle after the matrix number (which would mean a Western Electric recording). Three columns at the right-hand end of the sheets document the equipment used. The order of the columns and their contents are not absolutely consistent, but due to various redundancies (as we would say nowadays), there are no ambiguities.

We lack a “Rosetta Stone” to enable us to decipher the coded information. Ideally my deciphering should be regarded as a preliminary suggestion, which needs testing by other workers on other sound recordings; but I have checked the results for self-consistency in a number of ways.

The first column generally contains an integer between 1 and 31, which appears to be the “serial number” of the soundbox. It is applied consistently throughout England and on the continent of Europe (we never have the same soundbox turning up at different places at the same time, for example). For the first few months letters occasionally appear; these might be ex-Victor soundboxes, remaining with The Gramophone Company until “standardisation” was completed.

The EMI Archive has ten surviving recording horns, of which I have taken the dimensions. The second column logs the horn(s) used. In this case, the numbers aren’t “serial numbers,” but “type numbers.” Thus, a Type 100 horn would be used for orchestras, or orchestral accompaniments to soloists. Types 11, 11½, and 17 would be for speech, solo vocalists, or small groups of vocalists. Type 11½A was used for solo violin or viola; the surviving example is not a “right cone,” but one with its mouth at an angle, and the suspension hook suggests it “looked down” onto the bridge of such an instrument.

One Type 01 horn would be used for piano accompaniments, or a pair of Type 01 horns for solo classical piano; here, a surviving Type 01 has an angle in its axis, and was evidently suspended so it “looked down” onto the sounding-board of a grand piano. Horns would be combined for more complex sessions up to a maximum of four. Two of the surviving horns lack type numbers, and there are four types in the Artists’ Sheets which do not match numbered ones in the collection. Standard horns do not seem to have reached all HMV’s continental studios until the end of 1921.

The third column appears to describe the parallel-sided tube; in section 12.16 we shall see this was where several horns might be combined. Its effect upon the recorded sound is less easy to decide, because the whole purpose of the sheets was to ensure consistency from session to session. I found it impossible to find recordings of the same music recorded with the same soundbox and horn(s) in the same recording-room on the same date with only this factor changed. I found eight examples of different music which fitted the other requirements, so I prepared a CD with the different music side-by-side on the same track for comparison by expert listeners.

Unfortunately, it proved impossible to get consistent responses; but I wish to thank the listeners for the time they gave. A further test of self-consistency appears at this point. The code number for this artefact always changed when the horns increased from two in number to three (for example, when an extra soloist was featured). We shall see why in section 12.16; but in the meantime the purpose of the extra horn can be deduced from the enumerated artists, and correlates correctly with the type-number of the third horn.

The Artists’ Sheets provided me with a major breakthrough, because we appear to know the equipment used to make many recordings, and we can check if its performance can be compensated by our equalisation theories. We can also check the inverse procedure, equalising first and seeing if we can tell what equipment was used. If this proves to be successful, we might then extrapolate our techniques to records made by unknown equipment.

The next sections will describe four parts of the recording machine which affected the sound that was recorded - the mouth of the horn, the air in the horn itself, the parallel-sided tube, and the recording soundbox. To a first approximation, it will be satisfactory to assume that the pattern of the sound wave in the groove was the linear sum of these four effects. If it wasn’t, various forms of harmonic or intermodulation distortion would probably have become apparent; in practice such distortions were at an insignificant level. The most significant error to this assumption will be where mismatches occurred at the interconnection of the individual parts. Additional deviations in the overall frequency response would arise if the parts mismatched at certain frequencies.

The “megaphone” quality of acoustic recordings is due to the use of a horn. Horns were used because they have the property of matching the impedance of a relatively heavy vibrating diaphragm to that of a lighter medium such as air (see section 12.14 for an explanation of acoustic impedance). They can do this in both directions - whether used for recording sound or reproducing it - but most of the studies have been done for the latter purpose. A theorem known as the “reciprocity theorem” (Ref. 2) suggests this doesn’t matter too much. For most purposes, it is possible to take the work for sound reproduction, and use it to help us understand obsolete sound recording machinery - and much of my work begins from this assumption.

The evidence provided by photographs of commercial acoustic recording sessions suggests that about ninety percent featured a single conical horn between six and eighteen inches in diameter at the larger end (150 to 450mm), with lengths between three and six feet (one and two metres). The horns used on “domestic” recording phonographs for amateur use tended to be about half these sizes. Other evidence suggests that multiple horns, in particular, were commoner than the photographic evidence suggests; but this is a complication I shall leave until section 12.15. In an experiment carried out by the author, the sensitivity of a diaphragm to male speech one foot away (300mm) was improved by about 22 decibels when a single conical horn was used.

The acoustic properties of conical horns have not been analysed to a very advanced degree, because it was shown in the mid-1920s that the “exponential” shape gave better results for playback, and research tended to concentrate on that type of horn. Conical horns suffer two significant effects which overlap in a way which makes it difficult to separate them.

All horns transmit low frequencies with reduced efficiency unless they are made impracticably large, and this has been known from time immemorial. A good mathematical summary for the conical shape may be found in Crandall (Ref. 3). This analysis makes a number of simplifying assumptions, but shows the bass cut-off is twice as powerful as you’d get with the average “tone control.” To express the matter in engineering language, the effect is of two slopes of 6dB/octave, the two turnover frequencies being separated by exactly one octave.

Because this shape does not occur elsewhere in analogue sound recording, an approach is to quantify the effect of the higher-frequency turnover. When this is done, the “-3dB point” (see fig. 2 in section 6.7) conversationally becomes a “-4dB point”. (It’s actually 4.515dB). Both theory and practice suggest that the exact frequency of the “-4dB point” depend on the wavelengths concerned. Although he does not calculate an example, a graph drawn by Crandall suggests a horn with a mouth diameter of 600mm will have a “-4dB point” at around 730Hz. Some second-order effects (such as the proportion of cone which has become a “frustrum” under Euclid’s terminology) will affect this turnover frequency slightly, but not the shape of the bass-cut.

Exact measurements of bass-cuts of practical conical horns are difficult, because another effect happens at the same time, which I shall now describe.

The length of a conical horn is taken into consideration in an analysis by Olson (Ref. 4). The effect is to superimpose peaks and troughs at harmonic intervals on the overall bass-cutting shape. To state the matter in words, it is because sounds entering the mouth of the horn are reflected back from the narrow end, and again where the mouth meets the open air, so the air in the horn resonates, exactly as it would in an orchestral brass instrument. (Except, of course, that there is no embouchure to sustain the resonances). I shall call this effect “harmonic resonances” for short.

The pitches of these resonances depend on the length of the horn, so we may need to know how long the horn was before we can compensate it. I have no historical basis for what I am about to say, but it seems recording horns were made with a semi-included angle between eight and eleven degrees, because empirical trials gave harmonic resonances compensating for the bass-cut due to the mouth. But to restore the original sound electronically, we need to separate the two processes, because they have quite different causes and therefore cures.

Unfortunately, it is not possible to measure the first effect, because of the second effect. There is no way of dampening the harmonic resonances of a conical horn, except by “putting a sock in it,” and then we cannot measure the exact bass-cut, because the treble is now muffled! This suggests several approaches, and currently there is no consensus which is best.

• (1) Quantify the bass-cut by listening to acoustic recordings reproduced to a constant-velocity characteristic (section 6.4), and set a bass-compensator empirically. Conducted by a large number of practised listeners, on subject matter known to have been made with a single horn of known mouth-size, should give results consistent within a couple of decibels.
• (2) Rig an acoustic recording-horn (which must be terminated by an acoustic recording soundbox, or the harmonic resonances will have inappropriate magnitudes), perform into it and simultaneously into a nearby high-fidelity microphone, and adjust both the bass-cut and the anti-resonant circuitry until the two sound the same. This has the practical disadvantage that sound is reflected back out of the horn, and the microphone must ignore this sound.
• (3) Observe the impulse response of a real horn (for example its response to an electric spark), neutralise the resonances as displayed on an oscilloscope, and then measure the remaining bass-cut.

Brock-Nannestad was the first to carry out measurements of the axial response of one of the surviving acoustic recording horns at EMI Hayes (Ref. 5). These measurements included all the effects of the horn piled on top of each other, but both the bass-cut and the harmonic resonances are visible in Brock-Nannestad’s graphs.

At the University of Cambridge, Paul Spencer undertook seminal research for his thesis “System Identification with application to the restoration of archived gramophone recordings” (Ref. 6). This included theoretical studies of the harmonic resonances of a horn, and he tested his theory with measurements upon a small phonograph horn so a computer might know what to look for (to oversimplify drastically). He then wrote a computer program to analyse a digital transfer of an acoustic recording. He confirmed his process was accurate when the computer successfully found the resonances after a recording of a BBC announcer had been made through the same horn.

The main part of the paper outlined two mathematical techniques for determining the spectral distribution of the resonances in an ancient recording, which in turn tells us the length of the horn - the “System Identification” part of the problem. Then, the computer can be made to equalise the resonances - the “Restoration” part of the problem.

Unfortunately the program was never marketed; but the first point of significance is that the mathematical process gives results of an objective rather than a subjective character. So it has more validity to the archivist, and can be repeated and (if necessary) reversed at a later date if the need arises.

Another significance is that Spencer’s work gave the first unchallenged evidence of when it is inappropriate to neutralise the resonances of a horn. Spencer specifically tried his process on a 1911 Billy Williams record, which comprised speech, instrumental music, and vocal music recorded during the same “take,” recorded through a horn with conspicuous metallic vibrations. This was supposed to check the process didn’t give different answers on different subject matter when the horn was known to have been the same. The process indeed gave consistent results; but Spencer found resonances suggesting there had been two horns of different dimensions - one for the speaker/vocalist and one for the band. While it might be possible to crossfade from one to the other for certain passages, it would be difficult to achieve correct equalisation when the singer and the band were performing together.

This particular recording was also chosen because it was “noisy” – audio historians may appreciate the sentiment when I say it was a First World War “Regal” laminate! Spencer’s thesis makes it clear that the identification of the resonances was considerably less precise, but the presence of two horns was quite unambiguous.

Using the theories of Crandall, Olson, and Spencer as a basis, I designed an analogue equaliser which would do the same job at a fraction of the cost. I am very grateful to Hugh Mash for constructing the first prototype, and I am grateful to Adrian Tuddenham for two subsequent versions (based on different acoustic models). At present, subjective judgements show that all three have promise. If an ideal circuit emerges before this goes to press, I shall supply an appropriate circuit-diagram. However, when used on actual historic recordings, we have learnt that analogue restoration operators must have continuously adjustable controls to “tune” and neutralise the “honk” of a practical horn.

In experimental work, test-tones, live speech and suitable music were played into a horn. The theoretical works by Crandall, Olson and Spencer were confirmed in broad outline; but work is still in progress to measure the exact strength of the bass-cut and of the harmonic resonances. There are considerable experimental difficulties, some of which are not fully covered by the theories.

This is a section for which I shall not be providing recipes for compensation purposes - for two reasons. Firstly, we cannot envisage any electronic or mechanical way of solving the difficulty, and secondly there are ethical reasons why we shouldn’t anyway.

The mechanical resonances of the metal of the horn might theoretically be neutralised with analogue electrical networks as well; but it would require circuitry with unprecedented precision. It would also require the invention of a new technique to determine the values of hundreds of separate resonant elements whose features overlap. And all this cannot be done unless we know which horn was actually used on each recording, and its precise mechanical characteristics are known in great detail and with unparalleled accuracy.

Most horns used by recording experts seem to have been made of tinplate. With one exception, the ones surviving at Hayes have a thickness of 25 thou (0.635mm). (The exception is made from ceramic material an inch thick). We think the experts selected such horns partly so they would reinforce the music with resonances of their own (a technique re-invented thirty years later, when the steel “reverberation plate” provided artificial reverberation instead of a soundproof “echo-chamber.”). It is also known that the decay-time of a horn’s resonances could be controlled by wrapping damping tape around it, and several horns would be provided with differing amounts of tape so that rapid comparisons could be made by means of trial recordings. (Ref. 7).

In effect, the horn functioned somewhat like a bell, having transverse waves moving circumferentially around the metal; but unlike a bell, a conical horn resonated at a wide spectrum of frequencies, not a few fixed ones. Unhappily none of the damping tapes have survived. They are only depicted in photographs. It is not even certain what they were made of, nor how they were fixed to the horn.

Most of the surviving metal horns at Hayes have about a dozen small holes about one-sixteenth of an inch in diameter. It is possible they were for anchoring the tapes, but there are at least two other explanations. Railway engineers who fitted carriages with the first steel wheels found they made a terrible shrieking noise (the wheels, I mean). The cure was to drill small holes through the face of each wheel to break up the tendency to resonate. And I am very grateful to Sean Davies for drawing my attention to a memorandum about Victor’s recording processes, written by the Gramophone Company’s expert Fred Gaisberg after his visit to the USA in 1907. He reported “The Horns are made of Block Tin with very few perforations for ventilating.” My experiments have been done with horns with no holes; but I found it easy to “pop” the diaphragm when speaking into the horn and uttering a “p” sound. (Much the same can occur with microphones today). It is possible small holes allowed the blast to escape without too much sound energy leaking away.

So we have two major difficulties in dealing with the metal of the recording horn. First, ethical considerations which arise from neutralising conscious choices made by the original experts, and second the difficulties of analysing such behaviour using conventional calculus or equalising it using conventional equalisers.

But digital processing was found to offer a route towards a solution of the latter problem at a surprisingly early date. In 1974 Stockham (Ref. 8) introduced his process for the Victor records made by Caruso, which had remained best-sellers more than half a century after the tenor’s death in 1921.

It should perhaps be explained that the term “convolution” means subjecting a digital data-stream to a numerical process in which the digits are handled in the time-domain - that is, not necessarily in chronological order. Because both the wanted signal and the nature of the distortion are unknown in acoustic recordings, Stockham called his process “Blind Deconvolution,” and this phrase has now entered the vocabulary of signal-processing researchers.

To oversimplify greatly, Stockham took a modern recording of the same piece of music (assumed to have been made with a flat frequency response), and used it as a “prototype” so an older recording could be matched to it. At the time his paper was written the bandwidth was limited to 5kHz (possibly to reduce the costs of the computer processing, which took two hours to process a four-minute recording using a Digital Equipment Corporation PDP-10 mainframe). Judging from the graphs in his paper, he divided this frequency range into 512 bands.

Stockham’s pioneering work involved comparing Caruso’s 1907 recording of “Vesti la giubba” with Björling’s modern version of the same aria. An experienced sound operator would immediately see several defects in this idea. Firstly, there is no such thing as a “flat” recording of such a piece of music - it could only be approached, and never achieved, using a sound calibration microphone in an anechoic chamber. Even the best recording studios offer chances for many multipath routes between sound source and microphone, giving narrow frequency-bands with complete cancellation of sound. When convolved from prototype to processed recording, these would become peaks with amplitudes far beyond those discovered by Brock-Nannestad. Secondly, the voices of Björling and Caruso differ because of the different dimensions of their nasal, mouth, and throat passages, which cause a specific distribution of resonances to be superimposed on the sounds of the vocal cords. Humans use these resonances to tell one person from another. There is a risk that this process would make Caruso sound like Björling. Thirdly, Caruso had a wind band accompanying him, Björling a conventional orchestra. The effect of this is unimaginable. Stockham reduced these difficulties by “smoothing” the characteristics of his prototype, so the shorter resonances were ignored while the longer resonances were treated.

By comparing the smoothed prototype with the 1907 version, Stockham obtained a very spiky response curve which was supposed to indicate the response of the acoustic recording machine. When this was inverted and applied to the acoustic recording, many effects of the horn certainly disappeared. The result was issued on a published L.P disc (Ref. 9), and created a great deal of interest among collectors of vocal records, without any consensus about the fidelity of the technique emerging. One defect which everybody noticed was that the raw material supplied to Stockham frequently suffered from clicks and pops, which the process turned into pings and clangs of distracting pitches.

Stockham continued to process RCA’s Caruso recordings for many years, presumably making improvements as he went along, although details have not been published. For their part, RCA supplied transfers from their surviving metalwork. Eventually all the band-accompanied and orchestrally-accompanied published recordings were processed. (This immediately rings alarm-bells to a restoration operator. Does the process fail on recordings with piano accompaniment, or is it simply that a piano-accompanied prototype must be available?). The final corpus was issued in 1990. (Ref. 10).

So far, Stockham’s process has not been applied to recordings made with horns with known properties, so we cannot judge how faithful his process is. But the technique is very powerful, and although my ethical reservations remain, it is the only possible way to cure resonances in the metal of the horn.

Obviously, the closer an artist was to the recording horn, the louder he would seem to be. One writer, who was admittedly extolling the advantages of electrical recording, stated that a tolerance of one inch was aimed for. Frankly, I doubt this; if so, recording experts would have tied their singers to a stake! And Gaisberg makes it clear the Victor Orchestra had more freedom than that. However, the balance between two or more artists, and the balance between soloist and accompaniment, would only have been settled by distances from the horn(s). Small movements by each party might accumulate, and imbalances would then double.

It has widely been reported that the quality of the sound also varied with distance. Some collectors have puzzled why the famous Melba “Distance Test” recording was made in London in May 1910, especially since the matrix number suggests it was the last recording of the session! (Ref. 11). Brock-Nannestad did spectral analyses of the four vocal passages from this recording (Ref. 12), without publishing any concrete conclusion; but in Reference 1 he showed the recording was an attempt by the Gramophone Company of London to mollify the US Victor company, which wished to publish these recordings in America.

My own speech tests with a replica horn suggested little variation with distance, except that when the source of sound was very close I had bigger pressure differences between the inside and the outside of the horn, so the metal of the horn itself vibrated more. There is also the consideration that the closer I was, the more my face reflected sound into the horn than the open air did. But any differences are small, and typical of the clues listeners use to establish the “perspective” of artists (clues such as the ratio of direct sound to reverberant sound from the studio walls). I see no need to delete these clues, since I consider they are part of the original sound.

One major effect which remains unresearched, and which I suspect will be impossible to solve, is that the mouth of the horn was so large that diffraction effects occurred when the artist was off-axis. The theoretical work has assumed axial pickup so far; but it is well known that high frequencies will be attenuated if they are emitted off-axis. Indeed, Reference 7 shows a photograph of a session with the brass placed to one side specifically to muffle them somewhat.

Thus it should be remembered that electronic equalisation can never be made exact for higher frequencies (above about 500Hz) performed off-axis. The only cure would have been two or more horns pointing in different directions and connected at their necks, and this may explain why there is an imbalance between the photographic and other evidence: It was regarded as a trade secret. Other horns would be unplugged when the photograph was taken; yet there is ample evidence that multiple horns were fairly common. Which brings us to the next piece of hardware.

Before I continue, I’d better explain a scientific concept which isn’t at all easy - the concept of “acoustic impedance.”

When an electrical voltage is applied to a wire, the quantity of the resulting electric current depends upon resistance of the wire. When alternating voltages are applied to electronic components, the resulting current may not only depend on the resistance, but may also vary with the frequency of the applied voltage. To make the distinction clearer, engineers call the latter phenomenon “impedance” rather than “resistance.” It is precisely because electronic components exhibit different impedances at different frequencies that we are able to build electronic filters to emulate acoustical and mechanical phenomena. When sound energy flows down a tube, it too has to overcome the impedance of the tube - its “acoustic impedance.”

We may lose energy if we do not take impedances into account. Ideally, impedances should be “matched” whenever energy flows from one component to the next. This applies to electrical energy, acoustic energy, or the energy of vibrating mechanisms; and it even applies when we are converting one form of energy into another. For early sound recording experts without amplifiers, this was essential. The performers emitted only a few watts of acoustic energy, and the mechanism collected only a small proportion of this; yet it had to be sufficient to vibrate a cutting tool. A horn was used to match the acoustic impedance of air - quite low - to that of something comparatively high - the recording diaphragm.

But to collect sound from a number of performers, a number of horns were needed. I have no doubt that early experimenters tried connecting two horns to a recording machine by means of a Y-shaped piece of rubber tubing. Yet this did not double the loudness of the recordings; on the contrary, it made matters worse. In plain English, half the sound picked up by one horn would then escape through the other horn.

In practice things were even worse than this, because the recording diaphragm could not be matched to the first horn equally well at all frequencies. The acoustic energy had to go somewhere; if it wasn’t vibrating the cutter, it could only either be absorbed - turned into heat - or reflected back; so considerably more than half the sound energy might escape through the second horn. An exception occurred when the sound entering the two horns was very similar (the technical term is “highly coherent.”) In this situation, the sound pressures and rarefactions at the necks of the two horns were near to or at synchronism, so little or no sound escaped. Something like this occurred on HMV solo pianoforte records made with two Type 01 horns. Although the horns were situated over different parts of the piano, much of the sound from the piano’s sounding-board would be coherent, so leakages would have less effect.

When there was no coherence, there might be a third reason for the faintness of the actual recording. Each horn had its own impedance characteristic. We saw earlier that horns had longitudinal acoustic resonances depending on their lengths, and their acoustic impedances were highest at resonance. (A trumpet player uses this effect to match his embouchure to get a certain harmonic from his instrument). If the two horns were the same length, the two acoustic impedances would have their maxima at the same frequencies, and the sound energy would indeed be divided by two. But if the horns were different, their acoustic impedance characteristics would be different. While one horn was resonating and strengthening the sound, the other would have a lower acoustic impedance, so considerably more than half the energy from the first horn would escape through the other.

Earlier I described some of the horns surviving at the EMI Archives at Hayes. There are many shapes and sizes, and when we calculate their characteristics, we find they never exactly coincide. So how did acoustic recording experts manage as well as they did?

Electrical effects are often excellent analogues of mechanical or acoustic phenomena. Many readers will remember similar problems when connecting two microphones in the early days of tape recording, when you couldn’t get an electronic sound-mixer for love or money. How did you stop one microphone from short-circuiting the other?

A few years ago, Mr. Eliot Levin (of Symposium Records) donated to the British Library Sound Archive a small collection of acoustic recording artefacts, such as diaphragms and soundboxes. There is one Y-shaped connector fashioned from three pieces of laboratory rubber tubing joined together with rubber solution. It is very brittle, being some seventy years old; but by shining light down it, you can just see that one arm of the Y is partially closed off. I conjecture this arm would have been connected to a horn for a soloist, while the other went to a horn for picking up the accompaniment. Thus the accompaniment would be conveyed to the cutter with reduced leakage, at the expense of the soloist being quietened by a few decibels.

This exactly mirrors the early days of amateur tape recording. If you were balancing a soloist in front of (say) an orchestra, you would have one microphone for the soloist, and another picking up the overall orchestral sound. Because the soloist was close to his/her microphone, that volume control did not have to be fully up, and the resistance of the control itself stopped the orchestral mike from being short-circuited.

In section 12.12, I mentioned how Fred Gaisberg (of the London-based Gramophone Company) visited the US Victor Company in 1907 to study their methods. His memorandum contained this hitherto-puzzling sentence: “Piano accompaniments are made by using the ordinary Y with a small hole such as we now use for Orchestra accompaniments.” I think that sentence is no longer puzzling.

So far as the record-buying public was concerned, the big breakthrough happened at Victor in 1906. In his book The Fabulous Phonograph, Roland Gelatt observes that Caruso’s recordings made that February were not only his first to have an orchestra, but they were also slightly louder and more forward-sounding than anything achieved previously. To my mind, the significant phrase is “more forward-sounding.”

Although the acoustic recording process modified the original sound very considerably, the distance of the artist from the apparatus is usually abundantly clear! The only practical method of getting a “more forward sound” was to use several horns with artists placed close to each of them. Fred Gaisberg’s memorandum depicts three horns in use at once, plus the secret of how the horns were connected. (Actually, I must be honest and say it is only one of several possible technical solutions, and that I quite expect a supporter of Edison technology to tell me a solution was found at Edison’s studio first).

It seems that all Victor horns of the time (and of the Gramophone Company in Europe) ended with a short section of tube 11/16ths inch (17.5mm) inner diameter and 3/4 inch (19mm) outer diameter. Connections were made by fitting such sections together with short pieces of rubber tube over the joins. Gaisberg’s notes show a gadget for combining three horns to record “The Victor Orchestra.” Basically it consisted of a straight piece of tube of the same diameter as the end of the horn. (The exact length wasn’t given; it was apparently four or five inches). It is clear that the main orchestral horn would be connected to one end, and the recording soundbox to the other.

But this straight tube was modified by two more tubes entering it from two different directions, with bores tapered like a sharpened pencil. The conical shapes formed natural extensions to the narrow ends of the two extra horns, with the diameter at the narrow end of the taper being 3/16ths of an inch (4.75 mm). Sound waves from the main horn were unobstructed, but individual soloists could be brought close to the additional horns and their sounds injected into the main tube. The leakage from the main horn would have been less than one decibel (hardly noticeable), and because the hollow conical extensions accurately matched the additional horns, losses due to poor matching were smaller.

Gaisberg’s memo also shows how the musicians were located to give the correct musical balance, to counteract the horns not being equally sensitive. The memo also deals with the layout for sessions featuring vocal soloists when they were accompanied by the Victor Orchestra. Here a different gadget was used, because the soloist (who needed the unobstructed horn) was to one side. There is no detailed drawing of that gadget.

In section 12.6 I mentioned how HMV documented the coupling-gadgets used for each recording. We now have an explanation - not a proof - for why the identification always changed when the horns increased from two to three. For two horns a Y-tube might be used, for three a gadget. I also thought the identification was noted because the gadgets influenced the sound quality. My listening-tests showed that the sound quality was indeed influenced, but only by a small amount. I now think the gadgets’ numbers were entered because they provided an extremely concise way of logging the layout of the musicians.

There is another way to use two horns whilst minimising leakages. This is to connect one horn to one side of the diaphragm, and the other horn to the other. Paul Whiteman, the dance-band leader who recorded for Victor from 1921 onwards, later gave his recollections of acoustic recording days (Ref. 13). Instead of the recording-machine being placed behind a wall or vertical partition at one end of the studio, he describes it hidden inside a four-sided box, with what looked like ladders on each side, erected in the middle of the studio. The four walls of the box each had a recording-horn protruding some five feet above the floor (“in the form of a four-leaved clover”), and the recording-expert was encased with his machine so no-one could see what he was up to.

In this context, it seems the four horns fed the recording-machine by the shortest possible routes, namely through a pair of Y-tubes, each to a different side of the diaphragm. The only apparent alternative would have been a complex array of pipes all terminating on one side of the diaphragm. Not only would this have been less pure acoustically, but why entomb the recording-expert?

The electrical equivalent of a double-sided diaphragm would be two unidirectional microphones on one stand pointing in opposite directions, or one bi-directional microphone (which comes to the same acoustically). Here it would be necessary to move either the microphone-stand or the soloist to get the right musical balance, so this idea wasn’t used by early tape enthusiasts very often; but again, no electrical energy is lost, and I myself used a similar technique on several commercial recordings before I got my first stereo mixer.

As a result of my writings, I am very grateful to George Brock-Nannestad for effectively confirming my theory. He wrote (Ref. 14) that Victor had invented an improvement which they called “the DR System”, in which two recording soundboxes were coupled together at their centres by a steel wire under tension. The effect of this would be very similar to one soundbox addressed from both sides, while permitting as many as eight horns.

How may these researches be applied to the task of reversing the effects of the recording equipment?

The air in a parallel-sided tube has harmonic resonances rather like those in a horn, and matching at each end is similarly important. But what happens when a single conical horn is joined to a parallel-sided tube? Do the resonances of the air in the horn and in the tube continue to exist separately? Or does the air in the combination behave like one body of air? Or are there elements of both of these? I am not aware this problem has been handled mathematically; most published studies were done with exponential horns after acoustic recordings ceased to be made. Would it depend on the semi-included angle of the horn - in other words, the taper by which the horn differed from the tube?

The question can be answered experimentally. Listening trials with a full-sized tube and a replica conical horn are quite unambiguous. The air in the horn and the parallel-sided tube each have their own harmonic resonances in approximately equal proportions. Thus any quality side-effects of the parallel-sided tube may be compensated - so long as two conditions are met. First, the resonances are audible or measurable; but as I indicated in section 12.3, I did not get consistent results in A/B listening-tests. Second, it is assumed the parallel-sided tubes were not imposed for purely subjective reasons, in which case it would be unethical to reverse the experts’ work.

The HMV Artists’ Sheets show one case (the Catterall String Quartet) where the coupling artefacts were changed from “9” to “33” as the music changed from Beethoven to Brahms. That was on 18th June 1923; and the following day the exact reverse occurred as they completed the Brahms and started some more Beethoven. The horns remained constant throughout all this (four “Type 60”s). The two Beethoven quartets were from his early “classical” period (Opus 18, Nos. 1 and 2), and since the Brahms was from the height of the “romantic” tradition (Opus 51 No. 1), a definite musical effect may well have been sought. My subjective judgement is that the Brahms has more “warmth” in the octave 125Hz - 250Hz.

Although we know which tubes were used for HMV recordings in the years 1921-1925, we have no knowledge of their dimensions. For all other acoustic recordings, we have absolutely no historical knowledge at all! Thus my current view is that we should put the parallel-sided tubes (if any) “on the back burner” until future methods of analysing recordings enable us to quantify their effects.

In the meantime, the following observation is offered. When I made the tube for the above-mentioned experiment, I deliberately chose a length in which the tube’s harmonic resonances would be low-pitched enough to be heard by a 1920’s recording expert, yet quite different from those of the horn so they would be easier to hear. Thus one set of resonances filled in the frequencies between the other set. The overall sound was noticeably truer to the original; but there was a drop in overall volume (about six or eight decibels).

The next point is to decide whether a particular recording was actually made with two or more different horns. (If they were all the same, the horn-quality cure can fix them all at once). For certain types of subject matter (mainly solo speech and amateur phonograph cylinders), only one horn would be used. For HMV records 1921-1925 we can look in the Artists’ Sheets. But in the majority of cases, we can only “try it and see.” Once the cure has dealt with the principal sound, does other “hornyness” remain?

If so, we must first decide whether to take the matter further. All three methods of coupling (the Y-shaped connector, the parallel-sided tube, and the double-sided diaphragm) effectively separate the additional horn(s) from the main horn, so the main horn continued to operate as if the additional horn(s) weren’t there. Thus it is always perfectly valid to have the cure for the main horn in circuit. If it cures the sound of the principal performer, that may be sufficient for the intended application. (The accompaniment may be considered a convention of the recording studio, rather than representing real life). But this may not be true of other music. A band or orchestra with many parts of equal importance, a concerto, or a “concerted” session (this is Gramophone Company terminology for several vocalists), may tempt one to correct the additional horn(s) as well.

My first technique was to copy the unfiltered recording to one track of a multitrack tape-recorder. I then “track-bounced” this track through my so-called “hornyness-cure” to several other tracks, each with appropriate settings. This kept the several tracks in phase, and they could then be mixed together depending on which horn(s) were in use at which time. The current practice is to have several hornyness-cures with their inputs connected in parallel, and their outputs routed to different faders on an electronic mixer.

The hornyness-cure often “colours” the background noise, so sometimes you can hear the surface-noise change distractingly whenever you concentrate on getting the foreground sounds right. Fortunately, most horn resonance frequencies have good separation from surface-noise frequencies, and most horn resonances can be removed without side-effects. But one sometimes has to make a subjective compromise, either by leaving the tracks at a fixed setting, or by changing the balance very gradually.

This may give an artistically preferable result; but unfortunately I cannot think of a rigorous archival way of neutralising the effects of different horns used simultaneously. Even without the noise problem, we have to assess the contributions of the different horns by ear. Subjective judgement always seems to be needed, and we shall need another paradigm shift before we can be sure of restoring all the original sounds in an objective manner.

In the next few sections, I hope to show how the recording soundbox and its diaphragm affected the sound we hear from acoustically-recorded grooves today. I use the phrase “recording soundbox” to mean the artefact which changed acoustic energy into the mechanical vibration of a cutting-stylus. It may not be a very appropriate word, because it wasn’t always like a box; but I borrow the term from its equivalent for reproduction purposes. I shall assume readers are familiar with the latter when I discuss the differences between them.

But I must start with another short lecture on a further aspect of physical science, so let’s get that over with first.

Electrical recording was a quantum leap forward in the mid-1920s thanks to an idea by Maxfield and Harrison of the Bell Telephone Laboratories. Methods had been developed for electrically filtering speech sounds and transmitting them over long telephone cables. The two scientists realised equations for designing electrical filters could be applied to mechanical and acoustical problems by taking advantage of some analogous properties of the different bits and pieces. An electrical capacitor, for example, could be regarded as analogous to a mechanical spring, or to the springiness of air in a confined space. Using such equations to design mechanical and acoustical devices on paper saved much trial-and-error with physical prototypes. The analogies soon became very familiar to scientists, who switched rapidly between electrical and acoustic concepts in their everyday conversations, to the utter bewilderment of bystanders.

The principles were used for the design of an electrical cutting-head and an acoustic gramophone, which were marketed in America in 1925. They were fully explained in Percy Wilson and G. W. Webb’s book “Modern Gramophones and Electrical Reproducers” (Ref. 15), which naturally concentrated on the issues affecting reproduction. But the same principles continued to help recording until at least 1931, when Western Electric designed a new type of microphone (section 6.17 and Ref. 16). This was the first electromagnetic microphone to have a reasonably wide flat frequency response with low background noise - in fact, the first of a long list of such mikes which appeared over the next half-century. It was surprisingly complicated to design, and in my view it was the most successful application of the principles of analogies.

However, these analogies all depended upon a hidden assumption - that a pure electrical capacitance would be exactly equivalent to a mechanical spring or a trapped volume of air. This was an oversimplification of course. You cannot have a mechanical spring or a trapped volume of air without also having mass and friction (amongst other things). So the theory, which started from the behaviour of idealised electronic parts, could only be applied to “lumped components” - springs which had no mass, for example, or whose mass could be assumed to be concentrated at just one point.

Maxfield and Harrison’s playback soundbox differed from pre-1925 soundboxes in two ways, the second somewhat counter-intuitive. Firstly, they made the diaphragm of light stiff aluminium, which was domed and corrugated to make its centre substantially rigid. (This then behaved like a “lumped component.”) Secondly, they put a spring between the stylus-lever and the diaphragm - the “spider.” Starting from page 69, Wilson and Webb’s book took many pages explaining “lumped component” principles to show why a spring between the stylus-lever and the diaphragm gave better high-frequency reproduction. The idea seemed insane at the time.

By 1932 the lumped component analogy was at its height, and when the engineers at RCA Victor decided to reissue the acoustic recordings of Caruso with new electrically-recorded accompaniment, they assumed the resonance of the acoustic recording soundbox could be neutralised electronically. Such a resonance would have emphasised notes of a narrow range of frequencies between 2 and 3kHz, which might have accounted for the “tinnyness” of the sound. But they were wrong; parts of the soundbox behaved like “distributed components.” The springiness had been distributed throughout the material of the diaphragm and its surrounding rubber, so “lumped component” analysis simply didn’t work, and it wasn’t possible to recover Caruso’s voice to suit the new electrically-recorded accompaniment.

On the contrary, listening often suggested notes in this region didn’t get recorded at all. Acoustic recording-experts had avoided these situations whenever they could, of course; but I can think of several examples from my own collection, and I’m sure you can as well (Ref. 17). By working out the frequencies of the missing notes, it can be shown that (in Britain, anyway) the missing frequencies gradually became higher and higher over the years, varying from 2343Hz in 1911 to 2953Hz in 1924.

More recently, at least three different workers - I will spare you their names - have proposed identifying the acoustic recording diaphragm resonance by analysing the “coloured” background-noise, either because the diaphragm was stimulated into its own vibration by the wax blank, or because it was picking up the hiss of the vacuum pipe nearby. These results have always proved inconclusive. I am afraid there has never been a conspicuous diaphragm resonance we can latch onto - although this doesn’t mean the background-noise is useless, as we shall see in section 12.27.

So how did earlier soundboxes work - whether for reproduction or recording? I have been searching for a complete account without success. Lord Rayleigh’s book “The Theory of Sound” (Ref. 18) goes into the theory of circular diaphragms with distributed properties, but develops a model which is oversimplified for our purposes - a uniform circular plate like those in early telephone earpieces. Soundboxes behaved in a much more complex manner. The diaphragms were circular, it is true; but they had mass stiffness and resistance distributed throughout them, they were surrounded with rubber which provided both shear and longitudinal springiness and resistive damping, the stylus-lever had springiness at its pivot, and it formed a lumped mass where it was attached in the middle.

In general, recording-soundboxes were smaller and more delicate than reproducing soundboxes, so surviving examples always seem to be “in distressed state” (as auctioneers say). Rubber has always perished, diaphragms are often broken, and cutting styli have usually vanished. So we cannot simply measure their behaviour today. But it is clear that recording diaphragms had diameters from 1 3/16 of an inch (30mm) to 2 inches (51mm), these extremes being represented by examples from His Master’s Voice and Columbia. Towards the end of the acoustic recording era, they all seem to have been made of flat glass with thicknesses from 7 thou (0.17mm) to 10 thou (0.254mm). The edges were clamped in rubber, of various configurations but essentially coming to the same thing - rectangular cross-section pressing against both sides of the glass. None of them have much springiness at the pivot of the stylus-lever, which therefore feels surprisingly “floppy.” (Refs. 19 and 20).

I have been experimenting with a lateral-cutting soundbox of unknown origin intermediate between these two diameters (1 3/4 of an inch, 44mm) to find how it worked. I replaced the perished rubber (whose dimensions were still apparent), but could not be certain the new rubber had exactly the right physical properties. Unlike the HMV and Columbia patterns, this soundbox also had a mechanism for “tuning” the diaphragm, comprising an annular ring in a screw-thread which bore upon the rubber rings. Tightening it would raise the frequency of the diaphragm’s “piston-like” mode of resonance.

It is clear that there were two effects at once, one counteracting the other. The classical piston-like resonance was certainly there at the edge of the diaphragm, and it could be tuned by the screw mechanism. But the stylus-lever added mass at the diaphragm’s centre. This made it comparatively difficult to move, rather like a small island in the centre of a circular fishpond. Low frequencies caused the whole diaphragm to vibrate normally, as if the whole fishpond was being shaken up and down slowly by an earthquake. At higher frequencies waves were set up in the diaphragm, and a mode could develop in which waves were present at the edge, but with a “node” (no vibration) at the centre, so the stylus was not vibrated. This phenomenon had been described by Lord Rayleigh, and is known as “the first radial mode of breakup.”

Thus acoustic recording experts used soundboxes made from certain materials with certain dimensions so the “piston-like” mode of resonance would be counteracted by the first radial mode of breakup. I am certain they did not use “lumped analysis,” let alone “distributed analysis,” to design their soundboxes! So how did they do it?

Obviously, they used their ears. Probably the best way would have been to say “she sells sea-shells beside the seashore” into the recording machine, trying several soundboxes one after another. Then they would pick the soundbox which recorded the sibilants as clearly as possible without turning them into whistles. To judge this, a playback soundbox might also be needed whose performance was beyond reproach; but this could be tested by listening to the surface-noise, not the recording. A playback soundbox which made the surface-noise sound like a steady hiss, rather than a roar or a whistle, would have been a good candidate. In any case, the human ear can easily judge the differences between two soundboxes, eliminating the characteristics of the one which is making the continuous noises.

Numerous written accounts show that the diaphragm was considered the most critical part of the acoustic recording process. Not only did it have to be assembled so the first radial mode of breakup and the piston-like resonance counteracted each other as closely as possible, but different diaphragms were used for different subject matter. The standard stories say that thick glass was needed for something loud with high frequencies (such as a brass band), whereas thin glass was used for something faint and mellow (such as solo violin). It was the height of the romantic era in music, and mellowness was, if anything, considered an advantage. Thus experts intuitively called upon another scientific principle, the “power-bandwidth product” (section 2.3), which makes it possible to trade high-frequency response for sensitivity.

The balance between the piston-like resonance and the first radial mode of breakup cannot be perfect, because they have different causes, and therefore give different shapes to the frequency response. The piston-like resonance causes a symmetrical peak about 10 or 15 decibels high, while the first radial mode of breakup takes out a comparatively narrow slice to a great depth, and the two sides of the slice aren’t the same. Even with laboratory test-gear, I have not been able to decide precisely how to tune my experimental soundbox, because you cannot get a perfect frequency response whatever you do.

Thus each individual recording soundbox still had a sound quality of its own, and we can sometimes hear this on multi-sided sets of orchestral music. (Ref. 21). Straightforward listening suggests two different soundboxes were used during these musical works. Electrical equalisation has so far proved powerless to permit these pieces of music to be joined up seamlessly.

Record companies evidently learnt from examples like these, because consistency became important for multi-sided sets. According to a 1928 article in The Gramophone magazine (Ref. 20), English Columbia used only two soundboxes for the remainder of the acoustic era.

According to the Artists’ Sheets for the years 1921-1925, HMV had many more soundboxes (they had studios in many more locations), but the serial number of the diaphragm used on each take was always logged. Analysis also shows that experts travelling to a new location always took two diaphragms from their “base” studio with them. Whenever possible, they recorded each title twice, once with each diaphragm, in case one had changed its sound.

But analysis also shows that the same diaphragm(s) were nearly always used for different subject matter. Thus it seems the idea was to have a large number of diaphragms sounding identical, so Chaliapine might sound the same whether he was recorded in London or Paris. Volume adjustments seem to have been achieved by positioning the artists and selecting horns and coupling gadgets, as we saw earlier, rather than changing the diaphragm. The only times different diaphragms were used was on certain solo piano records. Evidently “romanticism” was considered preferable to a fainter recording!

You will notice that I suddenly switched from saying “soundbox” to “diaphragm” during that last paragraph. This is because HMV didn’t use a soundbox with a matching cavity, as described in Wilson & Webb’s book. Instead, the end of the horn terminated in a parallel-sided tube cut precisely at right-angles and clamped firmly to the recording machine. A diaphragm was pivoted close to the tube less than a hundredth of an inch away, so the diaphragm was free to move up and down with the level of the wax, without the horn having to move. There was therefore no matching cavity, and it was much easier to change diaphragms.

We now study the remainder of the soundbox, namely the pivot, the stylus-lever, and the stylus. To minimise the mass, a tiny cutting stylus was attached directly to the stylus-lever with sealing-wax. It therefore seems logical to consider the latter two together.

Brock-Nannestad’s paper (Ref. 5) shows two patterns of soundbox used by the Victor/Gramophone companies, with quite different pivot mechanisms. However, I consider it is too early to worry about the performance of these. “Distributed-component” analysis clearly suggests that both types would have vibrated exactly like “lumped components” at frequencies below 5kHz, and we shall have quite enough difficulties restoring these frequencies anyway. Perhaps studying the two types may be critical for recovering the original sound with fidelity one day, but frankly I am doubtful. Even with the best modern technology, such high frequencies are usually drowned in background noise. (This doesn’t mean high frequencies aren’t audible, only that we must rely on psychoacoustic knowledge to get them out). And lumped-component analysis suggests even-higher frequencies would be attenuated rather than boosted, so matters can only get worse the higher we reach.

It is also clear the wax would have imposed a load on the stylus-lever, but it seems the load would be “resistive” in character (that is, the same at all frequencies). Its effect would have been to dampen the piston-like mode of resonance. It would have varied inversely with the recorded diameter on a disc record. Since the piston-like mode and the first radial mode of breakup were designed to offset each other, there is little audible difference between the outside edge of a disc and the inside. Thus we may consider the load of the wax to be almost insignificant - compared with the other effects, anyway.

During this chapter we have studied all the parts of the hypothetical acoustic recording-machine I described in section 12.3, and examined how they affected the wanted sound. I am sorry we end in a down-beat fashion. We do not actually know how to equalise the performance of a diaphragm! There are three reasons for this. First, we often do not know the frequency at which the first radial mode of breakup took place. Modern spectral analysis may sometimes help, but as I said earlier, recording-experts often dodged musical notes in this region, and the evidence simply may not exist. With the aid of the HMV Artists’ Sheets, we may sometimes identify the actual diaphragm, and get the information from other recordings made with the same kit on nearby dates; but then comes the second difficulty. We would need to boost the attenuated frequencies, and we would get a highly coloured peak in the surface-noise. (Or we might learn how to synthesise the missing sound one day). Thirdly, we do not know precisely how the breakup was judged to compensate for the piston-like mode of resonance, so we have no idea of the effects of the main resonance on either side of the gap. At present, we can only leave the effects well alone.

In section 12.3 I outlined a “typical” acoustic recording-machine, and promised to deal with some of the “variations” later. So here we go.

The first point is that many early cylinder and disc records were not made with recording horns at all, but with speaking-tubes. To judge from pictures of Berliner’s and Bettini’s machines in Roland Gelatt’s book (Refs. 22 and 23), these comprised a hosepipe about an inch in diameter, with a mouthpiece like a gas-mask which was pressed to the speaker’s face. I use the word “speaker”, because I would imagine such a mouthpiece would inhibit breathing for a singer (it covered the nose as well), and it would eliminate “chest tones.” It is said that Berliner himself was the vocalist on the first disc records.

Acoustically, the effect would be that there was no bass-cut due to the mouth of the horn, and certainly early Berliners have a “warmth of tone” which seems impossible to explain any other way. However the speaking-tube had its own harmonic resonances, very analogous to those in a conical horn. I did a rough-and-ready experiment with a similar tube coupled to a soundbox, and found it just as easy to detect when a resonance was taking place, and to modify my performance accordingly. I note these details because, if someone wishes to recover the sound from these early records with “fidelity,” a different approach may be required.

It is also apparent that some phonographs had a hybrid between a speaking-tube mouthpiece and a proper horn. For example, a small horn only about six inches long might be coupled to a rubber tube. The performance of such an apparatus would depend sharply upon how “soundproof” was the space between the mini-horn and the speaker’s face. There would be much more bass-cut if the seal wasn’t perfect.

Another variation includes horns not of “conical” shape. (I defined this expression in section 12.3). To judge from photographic evidence, there were comparatively few of these. Most had a curve along their axis, instead of straight boundaries to form a cone. This gave them a shape between that of a trombone and that of an exponential horn. I am very grateful to Adrian Tuddenham for lending me an exponential loudspeaker horn of circular cross-section, which I coupled to a recording-soundbox to study how the non-conical shape would have affected the recording process. When used for recording, the air resonances were at similar frequencies to those in a conical horn of the same effective length and mouth-diameter, but the second and higher harmonics were rather more pronounced. I am afraid my technical knowledge is helpless to diagnose the reason for this - exponential horns are always better for reproduction purposes - but one possibility is a failure in “reciprocity” (section 12.7), because the sound waves are curved in the opposite sense. Lord Rayleigh’s definition specifically excludes transitory effects; curvature of sound waves would be significant in this context.

Roland Gelatt’s book shows a photograph of a third recording session with another type of horn whose cross-section comprised two straight-sided sections. A basic conical horn had an added flare which was also conical (Ref. 24). This picture was taken at Columbia’s London studio in 1916, with Alfred Lester, Violet Lorraine, and George Robey posing for the camera. These three artists only made one record together, so we can unambiguously identify the session as the one which produced “Another Little Drink” (Columbia L1034 or 9003). It is rather difficult to estimate the size of the horn (one has to guess the size of George Robey’s head), and the neck of the horn is beyond the left-hand edge of the picture; but I reckon it was about 915mm long (965mm when you allow for the end-correction).

Now I am also very grateful to Alistair Murray for lending me a superb copy of the record actually used in Spencer’s experiments (Ref. 25). It was recorded for Phoenix in 1911 and later reissued on Regal. It had been chosen for Spencer’s research because it was more “tinny-sounding” than most. I decided to compare the two records. Initially they were quite different when played at 80rpm; but when I increased the speed of the Phoenix to raise the pitch by about two-thirds of a semitone (when, as it happened, the two songs came into the same key), the match seemed very good indeed. The fundamental frequency shown by my hornyness-cure was almost the same as that established by Spencer, and matched a column of air 965mm long. Thus I believe we may have a photograph of the actual horn analysed by his computer. The fact that it was effectively two sections may also explain the uneven spacing of the resonances shown in Spencer’s Table 6.2.

This “conical-flared horn” from the “English Columbia” stable features so often on British acoustic records, that we have provided a special switch-setting for it. Because it has a flare, the longitudinal resonances are not only better-distributed, but more damped (because they match the open air better). Similar effects are often audible on continental recordings from the Odeon/Parlophon stable.

A great many horns were conical but with a kink in their central axis to allow the horn to “look at” something. There are four such horns at Hayes, all numbered with the prefix “0”. We have already mentioned the “Type 01” used for grand pianos. The “Type 04” appears to be made for duettists, since it comprises two such horns side-by-side in “Siamese Twin” fashion (Ref. 26). Unfortunately the duettists theory must be discarded, because it simply isn’t possible for two human beings to get their mouths sufficiently close together - the centres of the two open ends are only nine inches apart! I did an extensive search of the HMV Artists’ Sheets to discover what Horn 04 was actually used for, without success. Adrian Tuddenham has suggested that it was designed to look down at a concertina or melodeon, instruments featured on Zonophone rather than HMV.

I am very grateful to John Gomer for pointing out that many hill-and-dale soundboxes (both for recording and reproduction) had a mechanism to allow for variations in the level of the wax. Instead of the pivot of the stylus-bar effectively being fixed, it was free to float up and down. (The same principle may be seen on “fishtail weight” reproducers for cylinders). Vibrations were transferred only because the pivot was on a comparatively massive sub-chassis, forming a “high-pass filter” which cut the bass (namely, the unevenness of the wax). It would be effective below about 100Hz, taking low frequency sounds away from the cutter. So we have a further bass-cut at 12 decibels per octave on the recording in addition to the other bass-cuts. Precisely the same effect would also occur whenever a master-cylinder was copied using a so-called “pantograph”. In the latter case, the cut-off frequency can sometimes be estimated from the spectrum of the rumbling-noises transferred from the master; but the cumulative effect is so great that I cannot believe we will ever neutralise it.

However, the French Pathé company (the principal advocates of hill-and-dale discs in Europe) extended similar techniques to their disc records. Master-recordings were generally made on large wax cylinders some five inches (13cm) in diameter, and then transferred to published moulded cylinders and pressed discs as the demand arose. The discs might be of several sizes, including 20cm, 27.5cm, 35cm and 50cm; but not all recordings were issued in all sizes. However, a new possibility exists for reducing the background rumble, because different-sized discs of the same performance would now have different recorded rumble characteristics. Some experimental work on these lines has shown that a small amount of “mechanical amplification” was also introduced for larger discs (35cm and 50cm). Presumably this was achieved by lengthening the cantilever of the cutting-stylus so it functioned as a magnifying lever; so these discs should also have lower surface-noise (all other things being equal).

My final “variation” consists of “pneumatic amplification.” About one percent of all lateral-cut commercial acoustic recordings apparently comprise discs which have been copied through “an Auxetophone” (or equivalent). This was a reproducing machine whose soundbox was replaced by a valve mechanism fed by compressed air; it is reliably reported that such a machine could be heard six miles away. A wax disc would be processed into a low-noise metal “mother”, and then played on a studio Auxetophone into another acoustic recording machine. With careful choice of the three horns involved, the resulting disc could sound almost identical to the original, except louder; the technique was particularly used on the continent of Europe for string orchestras. Three things might confirm this has happened. First, the normal bass compensation for the mouth of a recording horn becomes impotent. Second, modern methods of surface noise reduction may reveal a second layer of surface noise behind the first. Thirdly, “peak overloading” may be audible, because the acoustic power at the neck of the Auxetophone drove the air into non-linearity, causing intermodulation products at mid-frequencies.

The “harmonic resonances” of the air in a horn occurred at a fundamental frequency and its harmonics; we saw earlier how singers might adapt their techniques. There is no doubt that certain elocutionists in acoustic recording days adapted very successfully - people like Harry E. Humphrey and Russell Hunting, for example. But other speakers obviously didn’t, and it may be ethically justifiable to reverse the effect.

The question then arises, who else might have been affected? We mentioned such performance compromises in section 12.4; but I found that my wife, a viola player, was unable to hear the effect when a Type 11 1/2 horn was picking up her instrument. The matter would have been even more difficult with the correct Type 11 1/2A, designed to “look down on” the bridge of a stringed instrument rather than face the musician square-on. The effect was also inaudible to a pianist when a Type 01 horn was placed anywhere near the hammers of a grand piano, although the recording expert may have positioned it away from critical notes of course.

The acoustic process has always appeared more successful with vocal music than other subject matter. This has usually been attributed to the limited frequency range; but now we have another reason. The lesson is that new playback techniques will probably be more successful on instrumental records than vocal ones.

Most other problems faced by artists in the acoustic era also applied to electrical recordings, so I shall leave those considerations until my final chapter.

Because we need resonant circuits to compensate for the acoustic recording machinery, we must apply de-clicking techniques before the compensation circuitry. The slightest “grittiness” means the cure will be worse than the disease, because the circuitry will “ring”.

At present (as with electrical records), I prefer to convert back to analogue for equalisation purposes, so the relative phases always come right. But I can see that various statistical techniques might be used in the digital domain to quantify resonant frequencies, and give a “figure of confidence” for the results. For multi-horn recordings it might even be possible to do several statistical analyses at once. It remains to be seen how much data is needed to achieve consistent results - it might be as little as one second or as much as several minutes. But if it is the former, we could then quantify not only how the different horns were “mixed”, but when the balance “changed.”

Correcting the bass-cut due to the mouth of a horn has sometimes proved difficult, because low-frequency background noise becomes amplified. (to be written) (incl. coloured background noise)


So far I haven’t referred to Stockham’s method for curing even longer-term resonances, such as those within the metal of the horn (Ref. 8). Stockham uses an empirical technique to distinguish between the short-term equalisation and the long term equalisation. Since we can deal with the (comparatively) short-term effects using conventional techniques, we can keep Stockham’s method up our sleeves (it is rather controversial anyway), until we have run out of conventional techniques. Then Stockham’s method will not be used as a cure-all, and will get a flying start on the long term equalisation. Alternatively, his method might be reserved for when conventional processes give ambiguous results.

This chapter has only revealed “the tip of the iceberg.” There is a great deal more work to be done. However, the achievements of Spencer and Stockham (shown in sections 12.10 and 12.12) show that it is possible to harness computers to the task of analysing historic sound recordings, for the purpose of discovering the characteristics of the equipment which made them. It isn’t the fault of either worker that they could see only part of the picture, so they tackled only part of the problem. And I’m sure I haven’t thought of everything either!

Before long, computers programmed to look for the known defects should allow acoustic recordings to sound as if they had been recorded through a microphone. But due to the compromises made by the performers, I am also certain human beings will be needed to override the results when they are artistically inappropriate.

TECHNICAL APPENDIX - now out of date, to be replaced

Prototype equalisers have been built to neutralise the effects of the air in a conical recording horn. A fundamental resonance and seven harmonics were equalised in one circuit, and the bass-cut due to the mouth of the horn in another. When tested against several conical horns under anechoic conditions, correct axial response could be restored within the limits of experimental error (about 1 decibel, and plus or minus one percent in frequency). I was not surprised by this, since Spencer and Olson had separately shown this would be the case (Refs. 4 and 6); but the following extra information came to light.

Spencer’s expression for relating the fundamental acoustic resonance to the length of the horn was found defective in two respects. First, he did not use “end corrections”. I found the usual “8R/3π” end-correction to be appropriate. Secondly, he did not use a suitable value for the velocity of sound in air; I was obliged to measure the ambient temperature of my experiments before I could verify his formula. When this was taken into account, the fundamental frequencies for a number of horns matched the physical measurements within about 0.5%. Assuming acoustic recording studios were quite warm, I used a value of 345 meters/sec in my calculations for historic sessions.

For the fundamental resonances, the prototype equaliser covered the range 150Hz - 300Hz at half-semitone intervals. This range was about right for most HMV records, although the very long horn with the 114Hz fundamental would not have been covered. The intervals were arranged logarithmically; but in retrospect I think it would have been better to have had them linearly-spaced, because the Q-factors were greater at low frequencies, needing more accurate tuning.

Another reason for the prototype was to discover whether I should compensate for even higher harmonics. I expected these would cause variations of only a couple of decibels or so, probably drowned by other sources of error. But a fundamental and only seven harmonics proved to be insufficient for two reasons. Firstly, higher harmonic resonances were within the range of frequencies which were both recorded most efficiently, and which were near the maximum sensitivity of the human ear. Secondly, the presence of uncorrected higher harmonics could easily be mistaken for those of another horn or of the coupling gadget. The fundamental and perhaps eleven harmonics might have been preferable.

However, it proved perfectly acceptable for the bass-cut due to the mouth of the horn to be adjusted on the same control. Its error was never more than one decibel once the resonances had been compensated, and it was much more convenient to have the two processes working together.

REFERENCES

• 1: George Brock-Nannestad, “A comment on the instalments by Peter Copeland on Acoustic Recordings”, Sheffield: Historic Record (magazine), No. 34 (January 1995), pp. 23-24.
• 2: The “Reciprocity Theorem” was first proved mathematically by Helmholtz, and later extended by Lord Rayleigh. Rayleigh’s own verbal exposition is thus: “If in a space filled with air which is partly bounded by finitely extended fixed bodies and is partly unbounded, sound waves be excited at any point A, the resulting velocity-potential at a second point B is the same both in magnitude and phase, as it would have been at A, had B been the source of sound.” (J. W. S. Rayleigh, “Theory of Sound,” (book, second edition (1896), Article 294).
• 3: Irving B. Crandall Ph.D., “Theory of Vibrating Systems and Sound” (book), Van Nostrand & Co. (New York) or Macmillan & Co. (London), July 1926, pp. 154 - 157.
• 4: H. F. Olson, RCA Review, Vol. 1, No. 4, p. 68. (1937). The formula is also quoted in Olson’s book, which is more easily accessible: Harry F. Olson and Frank Massa, “Applied Acoustics” (2nd edition), P. Blakiston’s Son & Co. (Philadelphia) or Constable & Co. (London), 1939, page 221.
• 5: George Brock-Nannestad, “Horn Resonances in the Acoustico-Mechanical Recording Process and the measurement and elimination in the replay situation,” Phonographic Bulletin No. 38, pp. 39-43 (March 1984).
• 6: Paul S. Spencer, “System Identification with application to the restoration of archived gramophone recordings” (doctoral thesis), University of Cambridge Department of Engineering, June 1990.
• 7: G. A. Briggs, “Sound Reproduction” (book, 3rd edition, 1953), Wharfedale Wireless Works, pp. 267-8.
• 8: Thomas G. Stockham, Jr; Thomas M. Cannon; and Robert B. Ingebretsen: “Blind Deconvolution Through Digital Signal Processing,” Proceedings of the I.E.E., Vol. 63 No. 4 (April 1975), pp. 678-692.
• 9: (in Great Britain) RCA Red Seal RL11749 (L.P.), RK11749 (cassette).
• 10: RCA Victor Gold Seal 60495-2-RG. (Boxed set of 12 CDs).
• 11: Gramophone Company matrix number 4195F. Published pressed in vinyl from original matrix on Historic Masters HMB36. Also dubbed to L.P., Opus 84.
• 12: George Brock-Nannestad, Presentation at seminar “Audio Preservation Transfer Technology for the Sound Archivist”, Peabody Conservatory of Music, Baltimore, 26th June 1991.
• 13: Paul Whiteman, “Records for the Millions” (book), New York, Hermitage Press Inc., 1948, page 3.
• 14: George Brock-Nannestad, “The Victor Talking Machine Co. new process of recording” article), Sheffield: “Historic Record” No. 35 (April 1995), page 29.
• 15: Percy Wilson M.A. and G. W. Webb, “Modern Gramophones and Electrical Reproducers” (book), London, Cassell and Company, 1929, pages 29-32.
• 16: Wente and Thuras, Journal of the Acoustic Society of America, Vol. III No. 1, page 44. The following two books may be more accessible: Harry F. Olson and Frank Massa, “Applied Acoustics” (2nd edition, 1939), Constable and company, London, pages 103-105; and A. E. Robertson, “Microphones” (2nd edition, 1963), Iliffe Books, London, pages 80-83. The latter describes the British equivalent microphone, known as the “Standard Telephones & Cables Type 4017.”
• 17: Here are some examples. (1) Beka 40628: “The Vision of Salome Waltz” (recorded 1908). Top D from the piccolo, doubling other instruments, is missing. This corresponds to 2343Hz. (2) Columbia L1067 (matrix 6766), “Till Eulenspiegel” (NQHO/Wood), top E flat on first violin disappears, corresponding to 2488Hz. (3) Two HMV records, DB.788 (matrix Cc3683-7 recorded 23-10-23) and DB.815 (matrix Cc5396-2, recorded 1-12-24) both eliminate top F-sharps from different violinists - Thibaud and Kreisler - corresponding to a frequency of 2953Hz. All these frequencies assume the artists were tuned to modern concert-pitch, which is the biggest potential source of error; but the frequencies I have given are certainly correct within three percent.
• 18: J. W. S. Rayleigh, The Theory of Sound, Volume 1 (second edition of 1894). Chapter X deals entirely with the vibrations of plates, with sections 218-221a specialising in disc-shaped plates. Experimental results are dealt with in section 220.
• 19: For the HMV pattern, see Peter Ford, “History of Sound Recording,” Recorded Sound No. 7, p. 228. (This article says “The recorder box on the Gramophone Company’s machine had a cord and coiled-spring tensioning-device that . . . provided a means of tuning the resonances of the diaphragm assembly to fine limits.” This is not true. In the late 1980s the cord on the surviving machine perished and the recorder-box fell apart, and for the first time it was possible to see how it worked. In fact, the mechanism served merely to allow the stylus-lever and (more important) the cutter to be changed quickly). Brock-Nannestad has also published pictures of the HMV pattern in two places: a photograph in Gramophone magazine, Volume 61 No. 728 (January 1984), page 927; and sketches of two Victor soundboxes (IASA Phonographic Bulletin No. 38, March 1984.)
• 20: For the Columbia pattern, see Percy Wilson M.A. and G. W. Webb, Modern Gramophones and Electrical Reproducers (Cassell, 1929), Plate II. Additional information is given in The Gramophone, Vol. 5 No. 12 (May 1928), page 488.
• 21: Two examples: (1) Liszt’s Hungarian Rhapsody, performed by the London Symphony Orchestra conducted by Arthur Nikisch, and recorded by HMV at Hayes. The clash occurs between sides 2 and 3, recorded on 21st June 1914 and 25th June 1913 (matrixes Ho 564c and Ho 501c). (2) Scriabin’s “Poème d’Extase”, performed by the London Symphony Orchestra conducted by Albert Coates on Columbia L.1380-2. Clashes occur between sides 2 and 3, and between 4 and 5. The work was recorded on 27th April 1920, but sides 3 and 4 were retaken on 7th May 1920. 22: Roland Gelatt, “The Fabulous Phonograph” (2nd edition), Cassell & Co., London, 1977, page 62.
• 23: Ibid., plate of Sarah Bernhardt following page 64.
• 24: Ibid., plate of the stars of “The Bing Boys” following page 64.
• 25: Illustrated London News, 21st December 1907.
• 26: Benet Bergonzi, “Gas Shell Bombardment”, HR 17 (October 1990), pages 18-21; and Peter Adamson, “The Gas Shell Bombardment Record - some further thoughts,” HR 19 (April 1991), pages 13-15.

-