GB2140256A - A procedure and apparatus for corrected reproduction of acoustic recordings - Google Patents

Abstract

The transfer function of the horn/soundbox combination used in an early acoustico-mechanical sound recording process (cylinders and discs) is determined by measuring the spectral content of the noise present in non-modulated grooves. The procedure is two-stage: a measuring phase which adjusts the replay equipment and a replay phase. Slow movement of the record during the measuring phase is suggested, permitting many time-slices over which to integrate the transfer function. Replay may also be slow, with the replayed signal being recorded on an intermediate record carrier and then reproduced at the correct speed which may be adjusted according to other criteria. One embodiment described uses a parallel frequency analyzer in a feed-back connection, in order that the noise of a sound recording appears as white noise at the output of the parallel frequency analyzer filter bank. Several embodiments describe the use of digital signal processing means. <IMAGE>

Description

SPECIFICATION A Procedure and Apparatus for Corrected Reproduction of Acoustic Recordings The invention relates to a procedure for the reproduction of sound recordings obtained by means of the acoustico-mechanical recording process in order to re-create the sound signal present at the mouth of the recording horn at the time of recording. The invention furthermore relates to apparatus for performing such reproduction.

It was already realized in the early years of the wax process for disc recording that the recording horn emphasized certain frequencies. The wax process had been in use for cylinder (phonograph) recording for many years but only in 1900 was engraving of the sound signal on a flat wax blank used commercially. In severe cases such emphasized frequencies gave rise to amplitudes and velocities of a magnitude quite beyond the capabilities of the reproduction equipment. This gave rise to unacceptable wear, but the only remedies known were careful orchestration, a suitable groove profile, and the possibility that some resonances of the recording horn might be counteracted by anti-resonances of the recording soundbox.By suggesting to customers a speed of reproduction which transposed the recorded resonances, placing them between the resonance peaks of the reproduction equipment, the record companies eliminated one consequence of the mechanical resonances: the wear. However, this improvement was obtained at the cost of distorted voice timbre and wrong tempo of execution.

In connection with the development of the "Phonodeik" (from 1912), Dayton C. Miller realized that the recording horn and soundbox gave rise to grave resonances. The "Phonodeik" is an instrument for tracing acoustic wave-shapes with a view to perform computational spectrum analysis. By systematic experiment Miller determined the exact transfer function of his recording equipment and devised a method for graphically correcting calculated spectra by means of the inverse transfer function.

The first serious attempt to correct the resonances of recordings made by the acousticomechanical recording process was performed by Thomas G. Stockham, Jr. et al. (Proc. IEEE Vol.

63, No. 4, pp. 678-692, April 1975). Two methods were described, blind deconvolution and power spectra (when weighted differently equivalent to Long Term Average Spectra). In the latter use is made of the fact that for statistical reasons, the power spectrum of an entire orchestra differs little from that of a different orchestra, provided approximately the same number of each instrument is used. However, this is not the way that the above mentioned methods were used. Rather, the assumption was made that two good tenors (male singers) have the same power spectrum and that horn resonances may be determined by subtracting a power spectrum obtained from the acoustico-mechanical recording process from that obtained from a recording made by the electro-mechanical recording process, provided the selection were the same.The horn transfer function thus obtained was inverted and used to filter the acoustico-mechanical recording. However, all that was obtained seems to be the voice character of the tenor represented by the higher-fidelity recording, controlled as to time function by the older, deficient recording. It is obvious that an unfortunate choice of reproduction speed will remove even this remainder of the older recording.

The procedure according to the invention provides for the separation of the speed problem from the horn resonance problem. It becomes possible to remove the horn resonances by inverse filtering without any requirement to know their absolute value, i.e. the speed of reproduction. Once removed it is a simple matter to increase or decrease the speed of a secondary sound carrier to transpose the corrected recording to any key determined by musical research or using other musical acoustic phenomena. On the other hand, determining the apparent horn resonances at an optional standard speed for a number of recordings belonging to one and the same recording set-up will permit relative placement of the speeds of any individual recording (expressed as percentages), in order that determination of one correct speed will calibrate all the other speeds of the same recording session.

The procedure according to the invention is characteristic in that in a measuring phase the record reproducing equipment is used for playing non-modulated grooves only containing noise signals, preferably preceding or following the sound recording proper, that the noise signal is spectrum analyzed in order to obtain resonance peaks present in the perceived noise, that a bank of filters is adjusted according to the result of the spectrum analysis in order to obtain a transfer function which is the inverse of the spectrum obtained, and finally in a replay phase to use the adjusted filter bank to correct the sound recording.

An embodiment of the invention is characteristic in that the spectrum analysis is performed after the noise signal has been filtered in the bank of filters, and with automatic means for adjusting the Q and amplification at each frequency corresponding to a filter in the filter bank, the automatic means being controlled by the output of the spectrum analyzer in order to obtain a combined output from the bank filters which displays essentially the characteristics of white noise, i.e. the same (arbitrarily chosen) energy per Hz.

Another embodiment uses a slow revolution of the sound carrier in the measuring phase and rerecording of the corrected recording onto an intermediate sound carrier, which is then speeded up in order to obtain a corrected sound recording at the correct speed.

Still another embodiment uses a slow revolution of the sound carrier in the measuring phase, an analogue-to-digital converter, frequency analysis by digital means, adjustment of coefficients in a digital filter representing a bank of filters, recalculation of said coefficients to correspond to the correct replay speed in the replay phase, and converting the filtered, digital representation of the sound recording by means of a digital-to-analogue converter.

Still another embodiment for use with a number of sound recordings performed with identical set-up of recording equipment furthermore contains digital storage means for storing spectral information in the form of coefficients for the digital filter, means for multiplying sets of coefficients by numbers representing the percentage deviation of recording speeds from a reference recording speed in order to obtain identical sets of coefficients, and means for indicating the deviations.

The invention is based on the realization that there is an intimate acoustical coupling between the recording horn, through the soundbox which not only acts as a transformer but which introduces resonances of its own, to the cutting stylus tip. It has been observed that the subjective character of the background noise of in particular very early acoustico-mechanical recordings was formant-like (the word "formant" taken in the meaning used in phonetics). The concentration of energy at certain frequencies is due to the resonances of horn and soundbox.There are two reasons for the influence of horn and soundbox, viz. collecting the background noise in the recording studio and filtering it by means of the horn and soundbox as it would occur with any other signal to be recorded, and, the cutting wax having a somewhat grainy character, the attempt to drive the cutting stylus point by these hardness variations being controlled by the loading of the cutting soundbox and horn. It will be readily apparent to the acoustician that these two mechanisms lead to the same result, whereas the latter contributes by far the most. Also, the separation of the swarf from the wax surface gives rise to impulsive action on the cutting stylus, even when cutting a silent groove.

The idea of the invention is to use spectral analysis to extract horn and soundbox resonance information from silent grooves. In practice, the signal-to-noise level is very low which is further aggravated by the fact that most early sound recordings available have been pressed in a material which was by itself of a grainy character and so superimposed a further, non-correlated noise signal on the desired wax noise signal. The various embodiments described hereinafter facilitate the extraction of the desired information.

The invention is to be further described in detail in the following with reference to the drawings, in which, Fig. 1 shows the chain of influences on the transfer function, from the recording set-up through the apparatus according to the invention, Fig. 2 shows a reproduction apparatus according to claim 3 Fig. 2a shows a detail of Fig. 2, Fig. 3 shows a fully digital version of the reproduction apparatus according to the invention, and Fig. 4 shows a detail of the fully digital apparatus according to claim 5 for performing a relative determination of reproduction speeds according to claim 6.

In Fig. lit is seen that the sound of the performance to be recorded is collected by the recording horn 1 which amplifies sound which is converted to a pressure at the recording soundbox diaphragm 2. This in turn drives the recording stylus so that an undulation is inscribed on the recording blank 3. However, because the recording blank is made of a mixture of waxes and metal soaps, and because the separation of the swarf from the surface occurs by cutting rather than by embossing, there is a reaction on the recording stylus which attempts to drive it by impulses. These impulses are stochastically distributed, however the movement of the recording stylus is controlled by the acoustic coupling to the recording equipment. Hence the recording equipment has a filtering effect on the noise spectrum which is recorded even in nonmodulated grooves.

On replay, the signal is picked off by means of pick-up 4 and it is linearly amplified in the pre ampiifier 5. This pre-amplifier must not contribute any equalization or de-emphasis, since it is the transfer function of the recording system and not the possible pre-emphasis that occurred at the time of recording which is eliminated by the method according to the invention. In the measuring phase the signal path indicated by "A" is used. The measuring phase occurs at a slow speed of revolution of the sound carrier; in this case a speed of 1/10 of the correct replay speed is chosen. This means that all frequencies are transformed to 1/1 O of their real-time values. An anti-aliasing filter 6 is used as is the practice in digital signal processing. This is followed by a Analogue-to-Digital converter 7, from which the sampled signal is entered into a digital memory 8.

A power spectral density calculation is performed by means of an FFT algorithm, the result being the mean value of a number of samples. This is performed at 9, and the result is stored as a 512 line spectrum in the memory 10. This representation is not directly suited for changing the transfer function of the replay equipment in that it is proportional to the square of the intensity at each frequency. Therefore the square root is taken by the unit at 11. The need for linear amplification in amplifier 5 is again stressed, since any modification of the unity transfer function from the pick-up 4 to the anti-aliasing filter 6 will show up as a modification of the transfer function just determined.Indeed, in order to eliminate any such influence it has been shown to be expedient to perform an analysis of the signal from a grainy but totally unmodulated groove obtained from a shellack pressing of an electrical recording, and to normalize the power spectra with respect to said analysis before taking the square root. Since the result of the analysis is to work on a bank of 1/1 0 octave filters, and since the range of interest is from 100 Hz to 3.2 kHz, it is necessary to average the 512 lines to obtain just 50 values logarithmically distributed. This averaging and inversion of the transfer function is performed in unit 12.

The filters 14 are 1/10 octave filters which have their mid-band gain determined digitally.

The centre frequencies are proportional to each of the 50 values decided on in the analysis, however the whole range is for use at actual replay frequencies. This means that each of the binary values on the 50 control lines c from the filter controller 13 take into account that the replay speed is 10 times the measuring speed.

The signal path "B" is used in the replay phase at correct speed and is no different from other equalized replay. The 50 filters 14 a through n each act on 1/10 of an octave in the range 100 Hz to 3.2 kHz which has been found to carry the most of the resonance information. However, the manner of working the invention is by no means limited to this choice of parameters.

In Fig. 2 a different approach has been used in that the bank of filters 14 a through n is used in a feed-back configuration and so could be interpreted as performing a parallel frequency analyzer function. The signal from pre-amplifier 5 is recorded onto an infinite loop tape machine 17 which has been demonstrated to work in this configuration. However, any other cyclic memory with comparable resolution would work equally well. The repeated noise signal is parallel analyzed by the voltage controlled parallel filters 14 a through n, the mid-band gain of which is controllable by a voltage at the control terminal c.

Each filter is followed by a detector-comparator unit 19 a through n which are only active in the measuring phase when the switches 18 a through n are in the horizontal position. The rectified output from each filter is compared to a reference voltage which is weighted by means of a voltage divider 20, the supply for which is the stable source 21. The taps on the voltage divider are distributed logarithmically in order to simulate the fact that the wider absolute bandwidth of fractional octave filters with increasing frequency also represents a greater noise power when the input to the filter is white noise. The voltage controlled filters do not form part of the invention and are readily designed by persons skilled in the art.Referring now to Fig. 2a it is seen that the amplifier 25 is coupled in a feed-back arrangement so that the voltage on the capacitor 26 sets the gain of the appropriate filter so that the amplified and rectified (and integrated, by means of a capacitor in the rectifier unit 24, not shown) voltage becomes equal to the reference voltage at the other input terminal of the amplifier. The storage capacitor 26 has a value which is sufficiently large to keep the gain setting of the filter for an appreciable time, even when the detector-comparator units 19 are switched out (which they are in the replay phase).

In the replay phase the switches 18 are thrown into the vertical position, thereby connecting the parallelled outputs of the filters 14 to a summing amplifier 22. The replay also occurs at a low speed, in that the signal from pre-amplifier 5 bypasses the cyclic memory 1 7 and is filtered in the filter bank 14. The summed output from amplifier 22 is recorded on an intermediate sound carrier 23. This is finally reproduced at the correct higher speed, and the corrected replay may be studied by means of amplifier 15 and loudspeaker 16. In this way all linear distortions of the recording system transfer function have been eliminated, and the exact and correct speed of the acoustico-mechanical sound recording may be determined by other means.

In Fig. 3 a schematic representation of a fully digital signal processing is given. Again the two signal paths "A" and "B" may be recognized. The analysis is identical, however as compared with Fig. 1 there is no need to limit the number of filters to 50, provided the computing power is available, it is intended that each line in the line spectrum shall contribute. The unit 2 7 is actually a a program algorithm which assigns coefficients to the digital filters. Such a program algorithm is well known, e.g. from the textbook "Digital Signal Processing" by A. Peled and B. Liu, John Wiley s Sons, New York, NY, 1976, Chapter 2 or "A Computer Program for Designing Optimum FIR Linear Phase Digital Filters" by J. H. McClellan, et al., IEEE Trans. Vol. AU-2 1, No. 8, December 1973.The only change in the procedures described is the re-calculation of the coefficients in order to make the digital filters work at the replay speed. This inly means the introduction of a constant factor. It is obvious that one may use the complex FFT spectrum representation directly in the frequency domain instead of going by the power spectrum as in the present description, without departing from the invention.

Replay occurs at a different speed of the original record, and therefore the signal path "B" uses a different setting of the anti-aliasing filter 29. The filtering then occurs by Analogue-to Digital conversion at 30, filtering in what essentially simulates a parallel bank of filters at 28, re-conversion to analogue form at 31, amplification and conversion to sound waves at 15 and 16 respectively.

In case the same recording set-up has been used for a series of recordings the speed of recording may still be different. The peaks and troughs of the transfer function of the recording system as determined by means of the apparatus as described above form a distinctive "fingerprint" which is independent on the recording speed. This statement is true for all known variations, since the recording speed only influences the damping of the recording stylus and since the variation for flat disc records is at least 1:2 from outside to inside of the record anyway. Obtaining complete overlapping of peaks and troughs by varying the speed of the individual recordings in a series of recordings will give a relative calibration of the recording speeds, and other means may then give an absolute calibration by determining the absolute speed of just one record in the series.This is shown schematically in Fig. 4 which is actually an extension of the signal path "A" of Fig. 3. For each recording the coefficients are stored in a particular memory register32 with labels attached, in order that multiplication and division may be performed in a known manner, using any recording in the series as a reference. The output is either in the form of a listing of quotients against lable or may be used in controlling the speed of replay (signal path "B") directly.

Wear of the run-in grooves may have distorted the spectral content of the actual record as compared to that present in the original recording wax, and therefore it may be desirable to make use of the silent periods between notes or utterances. To ensure that the spectral content is not distorted by reverberation of the recording room with the note just sounded, it is necessary to distinguish recording set-up noise from individual notes. This is obtained by playing the recording backwards in order that the reverberation comes after the silent period and by taking time slices which are analysed in order to obtain a mean power spectral density averaged over a number of time slices.The end of a silent period (rather the beginning of the decaying reverberation played backwards) is indicated by the sudden rise in a time-slice of certain frequency components which were not present in the previous time-slices. In accordance with all the procedures and embodiments described above this may be performed either digitally or by partly analogue means. An obvious extension is to let a digital apparatus search for silent periods, finally obtaining a large number of permissible time-slices on which to perform the analysis upon which to base the corrected replay.

The corrected replay as hitherto described is strictly only correct when working on a recording set-up using one recording horn only. In the case of multiple horns the results are not unified.

However the use of the peaks and troughs of the transfer function as a means of identifying a specific recording set-up is still valid and an important improvement over the state of the art concerning corrected replay of acousticomechanical sound recordings.

Claims (7)

1. A procedure for the reproduction of sound recordings obtained by means of the acousticomechanical recording process, the reproduction having the objective of re-establishing the sound signal present at the mouth of the recording horn at the time of recording, comprising the steps of rotating said sound recording, obtaining an electric signal from a pick-up engaging the groove of said sound recording, dividing the reproduction into a measuring phase and a replay phase, during the measuring phase performing a spectral analysis of the noise signal obtained from nonmodulated grooves before or after the recording proper, the result of the spectral analysis being used for adjusting a bank of filters in order to obtain essentially the characteristics of white noise at the combined output of said bank of filters when acting on said noise signal, during the replay phase utilizing the adjustment of said filter bank obtained during the measuring phase for acting on the sound recording proper.

2. An apparatus for the reproduction of sound recordings obtained by means of the acousticomechanical recording process, the reproduction having the objective of re-establishing the sound signal present at the mouth of the recording horn at the time of recording, comprising a gramophone/phonograph, a pick-up, amplifying means and filtering means, characteristic in ~that in a measuring phase the gramophone/phonograph is used for playing non-modulated grooves only containing noise signals, preferably preceding or following the actual sound recording, ~that spectral analysis on the noise signal is performed by means of a narrow-band spectrum analyzer, ~that the result of the spectrum analysis is used for adjusting a bank of filters in order to obtain a transfer function which is the inverse of the spectrum obtained, ~that in a replay phase the gramophone/phonograph is used for playing the actual sound recording, correcting the signal by means of the filter bank as adjusted in the measuring phase.

3. An apparatus for the reproduction of sound recordings obtained by means of the acousticomechanical recording process, the reproduction having the objective of re-establishing the sound signal present at the mouth of the recording horn at the time of recording, comprising a gramophone/phonograph, a pick-up, amplifying means and filtering means, characteristic in ~that the noise signal obtained from non modulated grooves is passed to a bank of parallel filters, ~that--thatthe amplification of each filter is automatically adjusted in order to obtain an output of the filter which is of the same magnitude as that of a similar white noise signal, ~that in a replay phase the actual sound recording is played back through the bank of parallel filters adjusted in the measuring phase, the outputs of the filters being summed.

4. An apparatus for the reproduction of sound recordings according to claim 2, characteristic in that a slow revolution of the sound carrier is used in the measuring phase and for re-recording the corrected recording onto an intermediate sound carrier, and that said intermediate sound carrier is speeded up to natural reproduction speed in the replay phase.

5. An apparatus for the reproduction of sound recordings according to claim 2 by essentially digital means, characteristic in ~that in the measuring phase a slow revolution of the sound carrier is used, ~that the result of frequency analysis by digital means is used for the obtaining of coefficients in a digital filter representing a bank of filters, ~that said coefficients are recalculated according to the ratio of measuring speed of revolution to replay speed of revolution, ~that in the replay phase the actual sound recording is converted to digital representation, digitally filtered, and reconverted to analogue form.

6. An apparatus for determining the relative correct replay speeds for a number of sound recordings obtained by means of the acousticomechanical recording process, utilizing an apparatus according to claim 5, characteristic in ~that spectral information on the recording set up is stored as the digital coefficients for each of the number of sound recordings, ~that a label distinguishing each sound recording is put against each set of coefficients, ~that division of one specific set of coefficients is performed into each of the number of sets of coefficients corresponding to said number of sound recordings, ~that a ranking is performed, listing said labels against the quotients obtained by said division.

7. A procedure for the reproduction of sound recordings obtained by means of the acousticomechanical recording process, the reproduction having the objective of re-establishing the sound signal present at the mouth of the recording horn at the time of recording, comprising the steps of rotating said sound recording, obtaining an electric signal from a pick-up engaging the grooves of said sound recording, dividing the reproduction into a measuring phase and a replay phase, during the measuring phase performing a spectral analysis of the noise signal obtained from apparently non-modulated grooves between notes or utterances, or from grooves before or after the recording proper, the result of the spectral analysis being used for adjusting a bank of filters in order to obtain essentially the characteristics of white noise at the combined output of said bank of filters when acting on said noise signal, the cessation of non-modulation being determined by the sudden rise in spectral content at certain frequencies, during the replay phase utilizing the adjustment of said filter bank obtained during the measuring phase for acting on the sound recording proper.