by Michael R. Lane
(Also see: Sonic Restoration of Historical Recordings--Part 1 (Jun. 1991))
Unnatural resonances are found on many early electric 78s, but they are especially bad for acoustics, where the conical recording horns (true exponential horns were seldom used) not only resonated on the harmonics of the basic frequency but also generated subharmonics. Occurring only under certain, very specific conditions, subharmonics are not a part of what we consider "natural sound" and are usually perceived as distortion. A half- or third-octave graphic equalizer can reduce the grosser" effects, but only the parametric--with control over gain frequency, and bandwidth--can eliminate the subtler distortions or unnatural horn resonance without damaging the natural sound, With early electric recordings, two or three; bands of parametric equalization usually, suffice, but with acoustics, seven, eight, or even more bands may be needed. Again, if the user, is not prepared to work for weeks, even reports, to develop skills he will do much better to stay with a straight third-octave, multi-graphic equalizer.
We'll give just one example of the possibilities with parametric equalization. A typical Victor acoustic record generally has one of its strongest resonant peaks in the area of 3,100 Hz; the precise frequency will vary from one record to another (Fig. 5A). By adjusting a parametric band for a narrow positive spike and sweeping the frequency back and forth while listening for the worst sound, you can easily find the exact center frequency of the resonant peak (Fig. 5B). It is always easier to find a resonant peak by exaggerating it rather than notching it out. Having found the exact center of the resonant peak, lower the gain of the parametric band to create a deep notch.
The result will sound much better and be something similar to Fig. 5C. Next, reduce the notch (increase the gain), which will bring back some of the ugly sound; then adjust the bandwidth wider or narrower for best sound. Work back and forth between the notch depth and the bandwidth until no further improvement is possible (Fig. 5D). If you are good at it, you will produce an exact but opposite EQ curve to the original resonance, and the resulting signal will be a smooth response without any loss of natural sound qualities.
This technique takes practice, but the results are well worth the work and much better than just making the notch deeper and deeper until the offending resonance is gone along with some of the music. In the cited example you may need to repeat the process on a lesser scale at the second harmonic, 6,200 Hz, and even at the subharmonic, 1,550 Hz. If an electric 78 had this 3,100-Hz resonance, it might also require additional work at the third harmonic, 9,300 Hz. In addition to eliminating unnatural resonances, parametric equalizers like the Orban 642B can serve as excellent notch filters and are very important as phase-modifying devices. Also, they can help reduce surface noise in conjunction with the Packburn 323A noise suppressor.
Anything that other equalizers can do, parametrics can do as well or better, but the amount of work is formidable.
NONLINEARITY AND GAIN RIDING
Another serious problem on acoustics, but almost never encountered on electrics, is that of nonlinearity in the treble frequencies. The acoustic recording process suffered from mechanical hysteresis (slop or play) in the linkage between the horn and the cutting stylus. Hence, the higher frequencies, insofar as they could be recorded, reproduce well on loud passages but poorly on soft ones. Among collectors of acoustic records, this is thought of as a "blasting" effect; that is, if the treble sound is relatively bright on normal passages, then the loud peaks usually come blasting through in an irritating way. The sound is similar to overmodulation on electric recordings.
This nonlinear or blasting phenomenon on acoustics is aggravated by unnatural resonances and internal phase-shift problems, but even when these are removed with parametrics, the original problem remains.
The best way of handling nonlinearity or blasting, at present, is downward gain riding of selected treble EQ on the loud passages while keeping the full treble EQ for normal and softer passages. Sometimes it's helpful to use upward gain riding of treble EQ in exceptionally quiet passages. After selecting the appropriate EQ, the procedure is to play the record a number of times, practicing the EQ gain riding until it can be done smoothly and imperceptibly, before making a final tape. Records with a relatively narrow dynamic range are not likely to have the nonlinearity problem. It is most noticeable with singers, particularly singers with a powerful, dramatic voice, such as Marcella Sembrich or Ernestine Schumann-Heink. Most recordings of such singers can't be successfully restored without this frequency-selective gain-riding technique.
In a similar way, although not for the same reasons, selected bass EQ gain riding can be of great help on many acoustics and some electric recordings. Sometimes, rumble filters, even when combined with low-frequency gating, as with the old Phase Linear 1000, can't totally eliminate rumble without cutting into the sound. In these cases, the bass should be set for the best sound at loud and medium levels and downward gain riding used on the softer passages where the masking effect of the music is absent. Again, careful discrimination is needed so that the results are inaudible to the listener. The most difficult of all are those recordings where selective gain riding is necessary on both high and low frequencies. You may have to practice extensively and then tape the record a number of times before the results are just right.
Fig. 5--A record resonance (A) and steps in its correction; see text. Sweeping
with a narrow parametric spike (B) locates the resonance, which can then be
notched out (C). Notch depth and bandwidth are then readjusted alternately
(D) for smoothest and most natural results.
DYNAMIC EXPANSION
Among many listeners to old recordings, the processing methods of dynamic expansion, reverberation, synthetic stereo, compression, limiting, and others have very bad names and often deservedly so. It is not because these things are inherently bad but rather because of their excessive use or, in some cases, because the equipment used is improper for the specific job. We cannot stress too strongly that processing must not be heard when it is applied. One should only notice when it is turned off; its absence should be noticeable, but never its presence.
On most electric 78s, dynamic expansion can greatly aid the sense of life and realism. Even acoustic records may need some expansion, as the recording horn and system of mechanical recording compressed the sound in unnatural ways. If the nonlinearity problems are corrected, then 3 to 6 dB of expansion will be of great help.
Noise pumping must be avoided at all costs, but proper noise-reduction techniques usually reduce noise pumping to inaudible levels except on records in really bad condition. Downward expansion of quiet passages may also improve realism and additionally can further reduce noise. Again, subtlety is the watchword. For normal dynamic expansion we favor our old, modified Phase Linear 1000, but for subtle downward expansion the Dynafex DX 1 is the unit of choice.
PHASE EFFECTS!
We have observed that, with some acoustic records undergoing restoration, the normal bass and treble controls (or the gain of adjacent parametric bands) seem to operate in reverse.
Increasing the bass level seems to produce less bass; decreasing the treble level seems to increase treble, etc.
We have observed this strange phenomena in varying degrees quite a number of times, always on acoustics and never on electrics. Assuming proper operation and use of the electronic equipment, there seems to be some form of phase cancellation or augmentation between the effect of the equalizer control and acoustic recording-horn resonances. There may also be interaction between the phase shifts inherent in analog equalizers and the resonances that are unique to acoustic records.
Digital equalizers, without the inherent analog phase shifts, may solve this problem. While some are available now, their sound quality does not yet seem comparable to that of the analog units. The physics of the complex phase and resonance relationships on acoustic records seems to defy analysis. For now, the problems can only be treated empirically-by trial and error.
Without doubt this is the most complex area in sonic restoration. It is the reason that work on acoustics is so much more difficult than on electrics. If you can do a good job on acoustics, then electrics will be relatively easy.
SPECTRUM ANALYZERS AND HIGH-FREQUENCY ENHANCEMENT
A third-octave (or narrower), real-time spectrum analyzer can be a very useful tool. It should have storage capability, a fast sampling rate, and peak and averaging display modes covering up to 4 or 5 seconds--perhaps longer. It's useful in locating resonant peaks, checking inherent record EQ, rebalancing weak treble or bass, and many other applications. We use the Gold Line 30, which, among its many other features, can store the waveform of the sound at any six moments in time, as selected by the user. These can then be recalled from memory at a later time and compared with each other as desired. For example, if you are working on the 1928 recording of Richard Strauss conducting the Berlin Philharmonic in Beethoven's Fifth Symphony, you may wish to store in memory the spectral pattern of the very first opening chord. Then by similarly storing that same chord from a modern recording or a number of modern recordings, you can leisurely compare the 1928 spectral pattern with modern ones having much better sound, and thereby get an idea of where the sonic problems lie and what to do about them. This type of comparative technique is useful and a time saver, provided you don't become so enamored with the technology that you forget to listen. More expensive and elaborate spectrum analyzers can show all kinds of helpful comparisons and information, but at times the spectrum analyzer can appear to give misleading information, so always use your ears.
Fig. 6--The BBE Model 422A Sonic Maximizer.
There are several devices on the market that can be used to enhance the weak higher frequencies on old records. The BBE Sound Model 422A Sonic Maximizer (Fig. 6) is an extraordinary device. It breaks the sound into three frequency bands divided at 150 Hz and 1.2 kHz, introduces small time delays to the lower bands, and uses a voltage-controlled amplifier to alter the apparent level of the highest band based on the content of the middle band. It is effective and unobtrusive on any sound source, giving better apparent highs and improved clarity. Its low band level control enables the user to correspondingly increase the bass frequencies. We use this device in all our restorations, and its outstanding characteristic is naturalness of sound. The Sonic Maximizer is very easy to operate and has superb sound qualities, but it can be overused.
REVERBERATION, TIME DELAY, AND AMBIENCE
Even when all the techniques discussed so far are employed, there is still something missing from the sound of early electric 78s and even more so with acoustics. Sound exists in some kind of environment--an acoustic space. The sonic reflections of the recording hall or acoustic recording studio are weak or missing on old records. A quality reverb, time-delay, and ambience system can be of great aid in restoring the sense of acoustic space. Acoustic records were very close-miked, or perhaps we should say "close-horned." It was the only way to get enough energy to drive the mechanical cutters. The medium and long room reflections associated with a good music environment are almost entirely missing, with only a few of the shorter room reflections being present.
This results in an extremely dead sound. What is needed is to introduce a fair amount of the longer reflections and an even larger amount of the medium delays. We use the a/d/s/ Model 10 Acoustic Dimension Synthesizer.
No longer available, it was a home entertainment system, not terribly flexible in its use with tape recording, so a few interface modifications were needed. However, it was more natural sounding than some professional units costing three or four times as much.
There are also some currently available reverb, time-delay, and ambience systems by Yamaha, Lexicon, JVC, and other manufacturers which are well worth investigating.
Remember, with any audio equipment, judge primarily by using your ears, not necessarily the published specifications, which don't always reflect what the sound is like. Reverberation, time delay, and ambience can be very useful in reducing the dry, dead sound of many historical recordings, but the result should never sound like some of the over-reverberant modern recordings.
COMPRESSION, LIMITING, AND DE-ESSING
The more these three processes can be avoided, the better, but sometimes they are necessary. In the acoustic era a form of mechanical compression was sometimes used with singers who had very powerful voices. It was really quite simple: The singer moved up close to the recording horn before singing very soft passages and backed away rapidly before singing very loud ones. It worked to some extent but could be pretty hard on the singers! Compression is the reduction in the level of loud passages in a graduated manner, and it is the opposite of dynamic expansion. With recordings having wide dynamic range, a small amount of compression may be needed, especially if the results are going onto cassettes.
Limiting, the abrupt stop of level increases beyond a preset point, should generally be avoided in restoration work. In a few cases, limiting helps to minimize the nonlinearity problem with acoustics, discussed in the earlier section. It should be used in addition to, but not to replace, selective-E0 gain riding.
Fig. 7--Playback attenuation of high frequencies in 15-ips tape recordings
by application of yellow china-marker or short lengths of 0.001-inch-thick
splicing tape to the oxide side of the recording tape.
De-essing, which minimizes the sibilants associated with the "S" sound on vocals, isn't needed on 78s because of their limited frequency range, but there is one exception. Worn records that sound "edgy" on highs may be helped by this, since distortion, barring actual sound breakup, is worse on the transient or attack where de-essers function. We have Orban's Model 424A Gated Compressor Limiter/De-esser and find it to be a useful, subtle, and natural-sounding device when applied with discretion.
EDITING
The standard method most of us have to rely on is the old mechanical method of tape cutting and splicing.
Most 78s are less than four minutes per side, and with long musical works, the unavoidable interruptions to change record sides are awkward and disconcerting. Where direct splicing, instead of cross-fading, is used to join sides, it is best done on the start of a note or possibly on the end of a note from the previous side; it should never be done in the middle of a pause as the shift in background hiss would be very disconcerting. A splicing block with a very gradual 2-inch cutting groove can be helpful in making splices less noticeable. When editing sides together, a gradual decrease in treble at the end of a record is sometimes audible on early records. This is inherent in the nature of disc recording; it was compensated for, in later recordings, by diameter equalization (a gradual increase in treble added on the inner grooves). When this treble loss problem is encountered, the restorer can add diameter equalization with a gradual increase of 3 to 5 dB of treble on the inner grooves, followed by a return to normal before the start of the next side.
An interesting editing approach to clicks, pops, and swish has been developed by Richard C. Burns. It involves using tiny bits of splicing tape on the oxide surface of the tape to create artificial dropouts! The advantages of this approach are that there is no alteration of the timing and the degree of dropout can be controlled by using one or more layers of splicing tape. Of course, exact location of the click, pop, thump, or swish is necessary, but this is no more difficult than with the traditional tape cutting and splicing method. A further advantage is that, if you make a mistake, correction only involves careful removal of the splicing tape, rather than splicing back in the portion of tape removed, as is the case in the cut-and-splice method.
Swish is usually too lengthy to be removed by tape cutting and splicing, but can be handled by this technique, using a compromise between degree of dropout and loss of high frequency.
In some cases of swish, where even a single layer of splicing tape causes too great a dropout, Burns recommends using a china-marker on the oxide surface and lightly coating this with baby powder to prevent the tape sticking to the recorder heads. (See Fig. 7 for the dropout characteristics.) The new digital, stereo-editing work stations offer greater flexibility, accuracy, and speed. They enable the restorer to try the blending of 78 sides by direct connections or cross-fades in any place and with any blending slope.
The work can be tried time after time until the results are perfect, and all this can be done without any mechanical tape cutting. Digital editing of any residual clicks or pops, which the Pack burn might leave, becomes relatively simple. In mechanical editing, there are many occasions where audible clicks and pops are masked by the sound which is heard at the very low tape speeds encountered while "rocking reels" on a recorder. It can be a maddening experience to try to locate a defect's precise location on the tape.
With digital editing it becomes relatively easy. There is only one problem--the expense of such equipment is far too high for most of us who live in the real world.
STEREO
Artificially synthesized stereo from a mono source doesn't work out very well. The Orban 245F creates quasi stereo by the use of phase filters and shifters, but even when the control settings are very low, the results are audible and do not sound very natural.
Their broadcast version with more bands is a bit more subtle, but it still sounds artificial. We have never heard any synthesized stereo that didn't sound better when the channels were paralleled to produce mono.
A very subtle and natural stereo spread can be produced by using slightly different settings on parametrics to create two channels. A somewhat fuller and richer sound can be achieved this way without any artificial quality. Mixing in some of the stereo output of a reverberation, time-delay, and ambience system can also be useful in helping produce a subtle and natural stereo effect. Perhaps better stereo synthesizers will appear in the future.
PROCESSING SEQUENCE
We have found that our best results are obtained when the restoration sequence is as follows: Clean the record and make repairs.
Play the record, adjusting for optimum stylus size, geometry, tracking force, and turntable speed to ensure correct pitch.
Adjust inherent EQ for best sound, using our record EQ chart (Table II, June issue) as a starting point.
Adjust for best noise-reduction.
Do preliminary sound rebalancing with a graphic equalizer.
Carefully study the results for unnatural resonances, blasting, and other problems.
Work with parametrics for best sound.
Work with high-frequency enhancement and selective EQ gain riding as needed.
Work with dynamic expansion and/ or compression if needed.
Work with reverb, time delay, and ambience as needed.
Work with quasi-stereo if deemed appropriate.
Carefully listen to how all the processing interrelates, and adjust to eliminate poor interactions. Also check to see if it is possible to eliminate the use of any processors! Theoretically, the fewer the better.
Repeat the second through twelfth steps until no further improvement is obtainable.
Document the audio chain sequence and all control settings.
Put the resulting work aside for a day or two, then come back and listen again to see if it still sounds the same; if not, adjust as needed. In some tough cases, it may be necessary to start over again.
When you are satisfied, tape the results at 15 ips to create a working or edit master tape. It may be necessary to make this tape several times until any selective EQ gain riding is imperceptible.
Edit the edit master tape carefully as needed.
Using the edit master tape as a source, repeat any of the preceding steps that yield further improvement.
Check and adjust for best absolute polarity.
Tape the final result, which becomes the copy master tape and is the source of all user records, cassettes, CDs, etc.
OTHER CONSIDERATIONS
For restoration work we use and recommend Otari reel-to-reel and Nakamichi cassette tape recorders. They are of superb sonic quality, but as with all analog recorders, there are problems with noise and high-frequency losses that increase with each generation. Tape print-through can also be a problem. Digital recording, which has overcome its own early problems, solves the difficulties inherent in analog tape recorders and is now the medium of choice. Unfortunately, as is the case with digital editing, the expense is still prohibitive for most of us. Digital audio tape recorders are beginning to make their appearance and offer the possibility of digital recording at a more affordable price.
Reliable long-term storage of recorded information is vital in the field of sonic preservation. It involves not only the stability of the media on which the information is stored but, equally important, the ability to play back that information without damaging either the storage medium or the information itself. Analog tape recording, barring catastrophic tape failure, is proven by over 30 years of experience to work well, but there are some losses. Digital recording, on the other hand, potentially offers permanent storage with near 100% accuracy for an indefinite time, and with no increase in hiss when a tape is copied from one machine to another.
Digital recordings using Reed-Solomon codes can be largely self-correcting in playback from information errors.
It is theoretically possible to lose a high percentage of digital information and still obtain 100% recovery if high-order Reed-Solomon codes are employed in the recording and playback process.
(Editor's Note: Dr. Toshi T. Doi's April 1984 article, "Error Correction in the Compact Disc System," went into this in great detail. -E.P.)
Physical deterioration of the storage medium-be it wax, shellac, wire, magnetic tape, vinyl records, or even optical discs-will eventually occur from time alone if nothing else. This may be rapid, as in the case of wax or acetate recordings, or very slow, as with the newly developing optical storage disk. But even if the physical storage medium deteriorates very slowly, information loss can be rapid in the recovery or playback process. Tin-foil recordings deformed disastrously with two or three plays, acetate and wax recordings deform seriously with a few dozen plays. Shellac and vinyl records suffer some wear with each playing.
The same is true of magnetic tape but to a much lesser degree. Only electronically or optically read information results in no wear on the storage media, but even here reading and tracking errors are common and appear to increase with the passage of time. Digital storage using high-order Reed-Solomon codes can potentially solve many reading and tracking problems.
Likewise, it can potentially solve the long-term storage-medium problem by re-recording every 50 years or so, before medium deterioration has progressed to the point where high-order Reed-Solomon codes can no longer correct errors. Of course, this storage technology hasn't yet been proven by experience, but archival storage for hundreds or even thousands of years is more than just a dream.
INTERCONNECTIONS
In professional audio systems, interconnection of equipment is made with balanced connections to keep hum and noise to a minimum. With our sonic restoration audio system, some of the most useful equipment is designed for home music systems and has only unbalanced (i.e., RCA) inputs and outputs. Coaxial-type interconnect cable, where the shield carries one side of the signal, is satisfactory in small systems, since interconnects are relatively few and short. On the other hand, very high-quality and complex audio systems, such as used in sonic restoration, can involve many electronic units and long lengths of interconnecting cable with resulting hum caused by r.f.i. and e.m.i. One good solution is to use quad mike cable for interconnections. We like Mogami Neglex 2534; the two twisted white wires are soldered together at both ends, as are the two twisted blue wires, with the shield grounded at one end only. This method of interconnection (bifilar configuration) eliminates electromagnetic hum induction almost completely.
One of the most frustrating problems for those working with complex audio systems is to have a patch panel, jack, or plug become intermittent during an important tape recording. Intermittent patch panels, jacks, plugs, switches, and connectors, along with noisy pots, are basically due to dirt or oxidation.
There are many products for this, but the outstanding one is Cramolin, which comes in either spray or liquid. It effectively cleans, preserves, and lubricates. It reduces contact resistance and prevents contact intermittents.
DISTORTION AND SAMPLERS
We have previously mentioned a number of things that may help reduce distortion including stylus size selection to track above or below the damaged area in a record groove, and de-essing to reduce distortion on transients, along with parametric settings.
In the usual situation where the noisier side of a mono groove wall is also the more distorted, the Packburn's Switcher can be a big help, but occasionally the noisier groove wall is less distorted than the quieter groove wall. If this is encountered, the Switcher shouldn't be used. A two-track recording should be made in which the treble (with bass filtered out) is recorded from the wall with noisiest but cleanest treble, and the bass (with treble filtered out) from its cleanest wall. The two tracks are then mixed together, yielding an overall reduction in distortion even though the noise will be increased. Where distortion is severe, with actual sound breakup or shatter, downward gain riding on the treble during the bad passages is about the only even partial remedy. Perhaps in the future the rapid explosion in computer and digital technology may offer some help.
In a few cases, electronic samplers may be of use. For example, there are cases of groove damage where a sizable portion of a musical note is missing. The sampler can take what remains, even if only a few milliseconds, and "loop it" to the degree necessary to extend the note to its original length.
It is possible to do this by the old method of repeated recording and splicing, but the results are usually not satisfactory. Every note consists of three parts-attack, sustain, and decay. With a good sampler, coupled to other equalizers if needed, it is possible to mimic all three note parts based on any fragment of the note that remains. Good samplers are expensive, but if you're doing a lot of work with badly damaged records, the cost is justified. Additionally, there is sometimes no choice but to copy a note or even a musical phrase from an undamaged portion of a record to fill in a totally destroyed area. The sampler is a great advantage in doing this, and it is possible to alter the intensity or timing of the note or phrase where the reinsertion occurs, to be more consistent with the artist's interpretation at that point in the performance. Whenever this type of alteration is employed, we believe that it is mandatory, in the interest of authenticity, to clearly state what was done in the documentation that accompanies the restoration.
ABSOLUTE POLARITY AND POWER-LINE CONDITIONING
Some maintain that the absolute polarity between musical source and listener should be maintained for best sound. With live musical performance this is automatic, but with electronic processing of any type, 180° polarity reversal may occur. There is no industry-wide standard in this matter. Our own simple tests convince us that this effect is real and probably inherent in the recordings. On many recordings, reversing the polarity of the audio system output makes a definite change in clarity, especially in the lower frequencies. Therefore we suggest that restoration systems-and all quality listening systems, for that matter-employ a circuit like that shown in Fig. 8. Restoration systems should be checked to maintain consistent polarity after the polarity switch so that the final restoration tapes will have the optimum polarity.
Fig. 8--A polarity-inversion circuit for line-level use. The IC is a TL072,
and the switch is a two-pole, two –position make-before-break type.
Many in the restoration field are unaware of the importance of having quality a.c. power, stable in voltage and free from spikes and garbage.
Standard power-line conditioners are inadequate for high-quality audio work.
The Tice Power Block is expensive, but it solves most a.c. power-line problems, other than outright loss of power, and we find a definite audible improvement in sound quality. Historical records have enough problems without further degradation from poor power.
ARTISTIC ELEMENTS
Except for live performances, records lack the interplay between performer and audience. Those who knew the great actress Sarah Bernhardt said that her recordings do not give the slightest idea of the impact she made live on the theater stage. All great artists of the past had that elusive quality called projection; without microphones, they could be heard clearly in the back row. But projection isn't so important now that small voices can be electronically amplified and enhanced to have tremendous impact. The charisma of many artists can't be expressed by sound alone-and in this regard, motion pictures, and now video recordings, have expanded the possibilities. Some artists, like Glenn Gould, preferred recordings and didn't concertize very much. With others, as diverse as Ignace Jan Paderewski and Elvis Presley, audience interaction and the visual impact of the artist were a vital aspect of their art.
Attempts to fully re-create the artistry of performers are problematical and incomplete at best. Our standard for sonic restoration is limited to what is possible. What we can say, and this is important, is that a given sound recording represents the sonic portion of a performance under one set of conditions, at one specific time and place.
Sonic restoration means to re-create as fully as we can the specific interpretation of the artist when the record was made. We cannot know the impact Sarah Bernhardt made when acting on the stage. The most we can hope to know is what she sounded like and looked like when she made her acoustic records and silent films.
ENGINEERING ALTERATIONS
Historical recordings made before 1960 have relatively few intentional alterations by recording engineers.
There are a few gross ones such as level shifts or changes in mike position, but these are very obvious and relatively easy to undo in our restoration work, although in the case of drastic level shifts the required corrective gain riding may be a bit tricky. For a shift in mike position, a gradual or abrupt, as may be required, change in compensating EQ will usually do the job. With modern recordings the situation is entirely different. Many of today's recordings are an artist and engineering blend not entirely authentic to the artists, being an intentional alteration of performance for effect. The current standard, especially in the pop field, is 24- or 36-track recording with innumerable alterations and mixings.
This is not an appropriate model for historical recordings. The earlier recording engineers did not have the means to make such modifications, and this is fortunate for the restorer.
Had historical recordings been able to employ a lot of artistic and engineering synthesis, we could never recover the artist's interpretation, as it would be impossible to distinguish between artist and engineer.
The sonic restorer must compare the past with the present to become aware of stylistic changes. Attempts in movies, TV, or new recordings to re-create the style of the past are frequently unsuccessful. For example, modern performances of Glenn Miller's arrangements are mostly syncopated dance music; they don't have Miller's quality of swing. With current technology it is quite possible to remove the voice from an historical vocal recording and add a new accompaniment. It is true that many of the earliest vocal acoustics had mediocre piano or poor in-house orchestras for accompaniment, but that's the way it was! While 78s of electric orchestras dubbed over the acoustic Caruso voice sounded fuller, the effect was grossly out of character.
It is dishonest to represent such inferior results as accurate. They can't be an integrated performance, and they lack the true style of the time. In the sonic restoration of historical recordings, the key word is historical, and discrimination is the essence of the art.
Gross tinkering is easily spotted, but the line between restoration and enhancement may be very, very thin.
COMPARATIVE TECHNIQUES
Through comparative techniques, we can infer, with considerable accuracy, what an historical recording sounded like at the original performance. The basic reference must be live sound heard with no electronic intervention. Next best is "flat" recorded sound such as can be heard on the old Mercury Living Presence and Westminster Lab LPs, Sheffield Direct Discs, most Telarc CDs, and some others. Timbre (sound quality) is all-important. If, through the use of filters, all overtones are eliminated, then different instruments or voices all sound alike except for pitch. It is the overtones, their presence or absence and their relative intensities, that produce timbre and the articulation of transients.
From personal experience and good recordings, we know what a symphony orchestra sounds like. Of course, there are subtleties of timbre distinguishing, for example, an English horn in one orchestra from another English horn in a second orchestra, but all these timbre differences tend to cancel out when you are dealing with many instruments. The main differences from orchestra to orchestra are created by the conductor. Tempo, accenting, liberties with the printed score, and emphasis of one instrument or group of instruments over another are inherent in the recording and depend very little on the quality of the sound. The sonic restorer will have to make allowances for some possible differences, such as lower pitch and smaller or different orchestral forces. Sometimes conductors in the acoustic record era used special musical arrangements and/or special seating arrangements for the members of the orchestra, to accommodate the acoustic recording process. A special violin, known as a Stroh violin, with a horn attached to direct weak violin sound at the acoustic recording horn, was used by some conductors. Bass string instruments were sometimes replaced with wind instruments. But keeping these differences in mind, if the restorer modifies the overall sound qualities of a primitive orchestral recording to sound much more like a modern one, he has done a good job.
INSTRUMENTAL
With instrumental soloists, the problems are more complex, because with very small groups or soloists, timbre differences are more obvious. While all violins have a similar overtone structure, the unique overtone structure of, say, Mischa Elman's Guarneri sets it apart from all other violins and is an important element in his art. A comparative approach is the way to solve the problem of inaccurately recorded overtone structure. The sound on Elman's acoustic records is compared with that on his later electric recordings, which have much better sound; then we can get a good idea of what modifications are necessary to make Elman's acoustic records sound more authentic While Elman undoubtedly always used his Guarneri violin, other artists using physically large instruments, such as a piano, could not be so selective and had to perform on the instruments that were available at their recording sessions. Nevertheless, it is reasonable to suppose that they would choose instruments of similar sonic qualities to the instrument they normally used. It is unreasonable to suppose that a pianist of the stature of Alfred Cortot, if he could not have his own magnificent Pleyel piano with him when he made acoustic recordings for Victor at Camden, would consent to play on any piano that was incompatible with the tone of his own. Unfortunately, available discographical information has little to say on what artist used what instrument for what recording. Still, much of instrumental performance was, as in the case of the orchestra, conveyed by acoustic records. If, a: the end of our work, a piano sounds Ike a piano and a cello sounds like a cello, we won't have gone far wrong.
With vocalists, especially acoustically recorded vocalists, the difficulties are much greater. While tempo, inflection, intensity, phrasing, and other characteristics are well conveyed even by acoustic records, the overtone structure is not. This overtone structure contains the qualities of tone and resonance that make each singer unique.
But what did the tone qualities of acoustic vocalists really sound like? We believe that the majority of vocal electric recordings made between 1925 and 1960, when played back properly, do a fairly good job of capturing the qualities of the artists they represent. Of course, the highest overtones are missing or weak, especially on the earliest electrics. In addition, the early electrics were sometimes still made in poor-sounding acoustic recording studios. Yet by 1930 or 1935 the worst problems were overcome, and the lower harmonics, which are most important, were being recorded in their correct proportions. Hence we can, with some obvious exceptions, employ many of the recordings made from 1925 to 1960 for comparison standards.
CROSSOVER TECHNIQUE
The great coloratura soprano, Amelita Galli-Curci, spanned the 1925 crossover from acoustic to electric recordings. If we play a number of her early electrics, we can get her unique sound quality firmly in mind. Then in restoring her acoustics, we use our equipment to increase the natural resonances and overtone structure which the electrics reveal should be present in her voice. Additionally, we decrease or eliminate the unnatural resonances of the recording horn. The procedure isn't perfect, since voices age and change with time, but such changes are usually a darkening of the voice and perhaps a lack of steadiness, not a basic change in tonal quality. There are hundreds of such comparisons that can give us a true picture of what the timbre of these 1925 crossover artists actually sounded like when they made their acoustics.
COMPANION VOICE
With Enrico Caruso the problem is more complicated, for he didn't live long enough to make any electric recordings. However, he did make many records with what we call companion voices-duets, trios, quartets, etc. So we adjust a Caruso and Galli-Curci collaboration for what we know is the Galli-Curci electric sound quality and obtain a "set" on Caruso's sound quality.
We do the same with the Caruso and Emilio de Gogorza duet and obtain another set on Caruso. We can also obtain sets on Caruso via his acoustic recordings with Frances Aida, Giuseppe De Luca, Ernestine Schumann Heink, and others whose careers and recordings extended into the electric era. Always, we adjust the companion voice for those qualities present on its electric recordings. Soon we have many sets on Caruso, and if he sounds the same on all these sets, then we have with considerable accuracy the true voice quality of Caruso as he sounded at the recording horn. We may speculate that he sounded the same on the stage of the Metropolitan Opera House, but we can't know for certain because there are too many variables between studio recordings and live performance. The memories of those still living, who heard him live, are colored by many years of nostalgia and diminished hearing, and the written commentaries of his contemporaries give only a word picture. We must settle for what Caruso and his fellow artists sounded like when they stood before the recording horn.
The companion-voice technique is a bit less reliable than that for singers who spanned the 1925 acoustic-to electric crossover where direct comparisons between acoustics and electrics can be made of the same artist.
When Caruso and Galli-Curci recorded together, he may not have stood the same distance from the recording horn as she did (he probably didn't), and the horn resonance characteristics for the two voices would be different.
Companion voices may have been considerably older when they made their electrics. Further, early electrics have their own host of resonance problems and defects even though they are vastly superior to the even earlier acoustics. However, as the number of companion-voice sets increases, so can the accuracy and authenticity of our results. To help establish the validity of using early electric recordings as sound references for acoustic recordings, we can make comparisons with artists who lived to make acoustics, electrics, and early LPs. By comparing their LPs with their late electrics, early electrics, and finally their acoustics, we can gather useful information as to the reliability of using later records as references for work on earlier records.
While, for example, Beniamino Gigli's voice changed in quality between his acoustics and his LPs, the change between his early and late electric 78s is not so great that useful comparisons can't be made.
Once we have established Caruso's true tone qualities, we can then use him in turn as a companion voice to obtain sets on earlier singers he recorded with where acoustic/electric crossover recordings don't exist. There are many singers that can be referred backward from more recent and, in some cases, relatively modern recordings. Certainly as we go backward in time, the comparative process to determine vocal quality becomes less reliable, but it does offer a technique based on the evidence in the records themselves.
COMPANY-AGE TECHNIQUE
The singers most difficult to pin down in timbre or vocal quality are those for whom there are few or, in some cases, no companion voices that can be traced into the electric era.
Even here, the situation isn't hopeless.
From our experience with thousands of records going back to the late 1890s, we have found that acoustic recordings of a given company and date tend to have very similar resonance and sound characteristics. The great Italian tenor Eduardo Garbin didn't make a great number of records, and his Fonotipia discs were very early, around 1905. We found only one companion voice, giving a single set on his voice quality, and that isn't enough. However, it is extremely likely that his Fonotipias have resonance characteristics similar to other Fonotipias of the same era that we can trace up to the electric era via the companion-voice comparison technique. It is reasonable to suppose that the corrective parametric settings that apply to these other Fonotipias should also apply to Garbin's recordings. It is never certain, but it seems the only approach, and it offers much more than just guesswork. Gar bin represents a style of singing not heard today, and it is important that restorations of his recordings and those of his contemporaries have some degree of authenticity.
THE ART
Sonic preservation must be preservation of whatever sound is on the recorded source and of that physical source itself. It's an objective process with little or no room for opinion. Sonic enhancement is primarily subjective--it can include altering sound quality, pitch, and reverberation, as well as adding quasi-stereo and substituting modern performances-and is done for "effect." Sonic restoration attempts to walk the very thin line between preservation and enhancement. It attempts to be authentic to the original performance. Acoustic recordings and, to a lesser extent, early electrical recordings cast veils of haze between the artists and the listener. The sonic preserver leaves these veils as they are; the sonic enhancer adds and subtracts veils of change for effect. The sonic restorer tries to remove all the veils between artist and listener.
Whether the restorer is successful will be determined by the results, which must speak for themselves.
The current trend in musical culture, especially in the popular field, seems to be for ever more and greater sensory impact. To us this seems to be a substitute of quantity for quality. The new buzzwords are computer, synthesizer, and digital. Just as the piano and violin are the instruments through which the pianist and violinist create music, so too can the computer be the instrument through which the musician programmer creates music. But music is a human experience, and if with technology the human element should disappear, whatever sound remains won't be music as we know it. While new technologies are important in the sonic restoration of historical recordings, the new technological art forms should not be imitated.
Work in the restoration field does not always move forward. The Nimbus Prima Voce CDs are a case in point. Nimbus plays both acoustic and electric historical recordings on an antique acoustic phonograph and records the resulting sound using the latest technology. This 60-year-old re-recording technique lacks transients, has many unnatural resonances, and eliminates the lower frequency characteristics of the voice. On acoustic records it is poor; on electrics it is a disaster. The glorious lower tones of Rosa Ponselle are gutted, and the already reverberant Eva Turner record sounds like it was recoded in a rain barrel! Nimbus does a disservice to the artists.
It is essential not to confuse today's extraordinary technology with its applications. The technology is neutral; the applications rarely are. Modern technology, having overcome many early problems, can produce unequalled realism and naturalness of sound. However, the "a little is good, therefore a lot more is a lot better" approach can easily abuse the sound. Those who argue against use of today's sophisticated processors in sonic restoration confuse the technology with the application. One of the hardest lessons we have learned and are still learning is that if use of a processor produces a good result, then try a wee bit less. It may be even better.
The would-be sonic restorer needs some detachment from the current musical culture plus some historical perspective. He also needs flexibility in looking at both old and new music and technology. Attitude is fundamental and basic to everything else. The art of sonic restoration cannot be taught; it is a perception of, and respect for, the older artists and their achievements. It is the patience to do the job over and over until it is right. Above all, the art is that subjective discrimination, in using all the electronic tools now at our command, to attempt as objective and authentic a restoration as possible. The sonic restoration of historical recordings is both art and science. The science is the knowledge of music-its history and psychology. It is acquiring and maintaining the restoration equipment. It is developing the skills to use that equipment. It is staying abreast of new and rapidly changing technical developments. The art is undefinable.
While patience, discrimination, attitude, cultural outlook, ethics, and love of the work are some of its aspects, it remains an elusive quantity.
When the best of the science is combined with the best of the art, then a clear window of great pleasure and insight opens for us, and blues singer Ma Rainey, King Oliver's Creole Jazz Band, cornetist Bix Beiderbecke, soprano Nellie Melba, conductor Richard Strauss, pianist Josef Hofmann, and a host of other artists and performers all come alive for us.
---------
Without the help of my associate, Mr. Donald H. Holmes, this article would have been impossible. His years of experience with parametrics and many useful suggestions on restoration are a vital part of Lane Audio and this article. Also, I am greatly indebted to Richard C. Burns of Packburn Electronics and Carl J. Malone of Specialty Audio for their review and suggestions.
The records (photos of labels) shown in this article are from the collection of Michael R. Lane and Donald H. Holmes, who run a restoration service, put cut auction lists of historical records, and offer specialty equipment for historical record reproduction. Lane Audio & Records is located at 1782 Manor Dr., Vista, Cal. 92084, USA.
= = = =
Also see: Sonic Restoration of Historical Recordings--Part 1 (Jun. 1991)