US2860187A - Artificial reconstruction of speech - Google Patents

Description

Nov. 11, 1958 E. E. DAVID, JR., ETAL 2,860,187

ARTIFICIAL REcoNsTRUcTIoN -oF SPEECH 5 Sheets-Sheet 1A Filed Dec. 8, 1955 E. E. DAV/QUE; S. Mc DOA/,4m

A TTORNEV Nov. l1, 1958 E. E. DAVID, JR., ETAL ARTIFICIAL RECONSTRUCTION OF SPEECH Filed Deo. 8, 1955 5 Sheets-Sheet 2 l/s 4/ Y 2/3 4E 27 2711 E7 l f TA PPEo TA PPE@ TAPPED TAPPEO OELAY L/NE DELAY L/NE DELAY L/NE DELAY L/NE CONTACTS OE CONTACTS of' cONTAGTS OP CONTACTS OE RELAY TREE RELAY TREE RELAY TREE RELAY TREE L l .J r, //r* cO/LS OF RELAY TREES ANO SEL Ec TOR SW/TcH 2\6 /23 NAR/ER ANALOG PULSE LP E. To O/G/TAL TT 50 GEN. CONVERTER MARKER l L PULSES FROM TP GEN. 2G +1 2*- PHANTASTRON V] |'1 NONOSTAELE D Y N. P/TcH ELA GE VOLTAGE \5/ RESET f 55 PROM P/L TER 23 5T? S/4b 54@ 52 PULSE GEN j?) SGALE SCALE 1# OF OF /00 KC- TWO TWO \SS f S627 /Nl/E/VTORSE 5 DAV/0. JR

H. S. MCOONALO Nov. 11, 1958 E. E. DAVID, JR., ET AL 2,860,187 ARTIFICIAL REcoNsIRucTIoN oF SPEECH 5. E. DAV/0, 'J/a /NVENTORSH s. Mc 00A/ALD 554% CMMI',

A TTOR/VEV i Nov. 11, 1958 E; E. DAVID, JR., ET AL 2,860,187

ARTIFICIAL REcoNsTRucTIoN oF SPEECH Filed Deo. 8, 1955 5 sheets-sheet 4 0 OUTPUT 5 E DAV/0, JR.

BVHMM'C.

A NORA/5y E. E. DAVID, JR., ETAL ARTIFICIAL RECONSTRUCTION OF SPEECH Nov. 11, 1958 5 Sheets-Sheet 5 Filed DSG. 8, 1955 .5. E. DAV/0J@ /NVENTORS/f. 5. McDo/VAL@ ATTORNEY` United States Patent O ARTIFICIAL RECONSTRUCTION OF SPEECH Edward E. David, Jr., Berkeley Heights, and Henry S. McDonald, Summit, N. J., assignors to Bell Telephone Laboratories, Incorporated, New York, N. Y., a corporation of New York Application December 8, 1955, Serial No. 551,940

14 Claims. (Cl. 17915.55)

This invention relates to economy of frequency bandwidth in telephony, and particularly to the artificial reconstruction of speech from narrow band transmitted signals. The principal object of the invention is to `improve the quality of such artificial speech.

It has long been known that a speech wave is almost periodic in character, the difference between the wave of each period and the wave of the following period being small compared with the variations of the wave within any period. Many proposals have undertaken to turn this situation to account in the reduction of the band of frequencies required for the transmission of the information content of the speech wave. According to one such proposal, described by H. W. Dudley in his Patent 2,115,803, May 3, 1938, an electrical speech wave originating, for example, in a microphone, is chopped in synchronism and in phase with its fundamental period and certain full periods, for example every other one, are discarded. The retained periods are transmitted to a receiver station, preferably after stretching each one in the time dimension to fill the gap left in the wave train by elimination of the discarded period. Such stretching carries with it a corresponding reduction in the frequency band required for transmission. At the receiver station the received wave is resto-red to its original dimensions in time and frequency, thus leaving blank intervals corresponding to the eliminated periods. The teaching of the Dudley patent is to fill each of `these gaps or blank intervals with one or more repetitions of the last received period. As a result the reconstructed wave comprises a sequence of identical periods, followed by another sequence of mutually identical periods, and so on. The price paid for the bandwidth economy in transmission is therefore an increase in the abruptness with which the waveform of each such group of periods changes to that of the next such group. Thus, for example, when two out of each group of three successive periods of a changing speech wave are eliminated at the transmitter, i. e., with a chopping factor N=3, the transmission band can in principle be reduced by a factor of three; but the abruptness of transition, at the receiver, from the last of each group of three like periods to the first of the following group of three like periods, is likewise increased by a factor of three.

This abruptness of transition naturally makes Ifor noise and unnaturalness in the reconstructed speech. It is a specific object of the present invention to reduce the noise and distortion which arise in a chopped speech transmission system for these reasons. This object is attained, in accordance with the invention, by inserting in each blank interval of the received wave, not merely a repetition of the last received period, but a modication of it. One such modification is a weighted average of the two waves which respectively precede and follow the blank interval. Thus, if the blank interval is only a single period in length, (i. e., if the chopping factor N has the value 2) the preceding and following waves are weighted alike `and the modified wave is obtained by 2,860,187 Patented Nov. 11, 1958 equal mixture, while the third is filled with a mixture` of one part of the preceding wave and three parts of the following wave. The apparatus `thus averages the waveforms of earlier and later transmitted waves, and from such averages constructs artificial waves and inserts them in the blank intervals.

One of the principal differences between each period of a speech wave and the next is a difference in ampli-` tude. Indeed, some experimental evidence indicates that this is the most important single difference. Accordingly,

the invention also provides a simplified system. in which' amplitude differences are taken account of in the intercalation of artificial wave periods between successive transmitted wave periods, while disregarding other differences. In this simplified system an auxiliary amplitude control signal is derived at the transmitter station and is transmitted, along with the chopped speech waves, to the receiver station where it controls the amplitudes of repetitions of each transmitted speech Wave for insertion in the blank interval which follows it.

To provide that the auxiliary gain control signals shall affect only the repetitions of the received Wave and not the received period on its first occurrence requires switch p ing apparatus of some complexity, operating in syntion terminals of a switchmorgate 5 to whosepontrol.`

chronism with the fundamental pitch frequency or a subharmonic thereof. In accordance with a further feature of the invention these complications are avoided by use of compandor apparatus including an amplitude compressor at the transmitter station and a corresponding amplitude expandor at the receiver station. The amplitude control signal, which modifies the amplitudes of the repetitions of the received wave at the receiver station, may conveniently be derived from the compressor at the transmitter station. With this arrangement the amplitude control signal may be applied to the final output signal at the receiver station including each originally received period as well as its various repetitions.

The invention will be fully apprehended from the following detailed description of preferred embodiments thereof taken in connection with the appended drawings in which:

Fig. l is a block schematic diagram showing a transmission system embodying the invention;

Fig. 2 is a Vschematic circuit diagram showing a weighting network for use in the apparatus of Fig. 1;

Figs. 3 and 4 are block schematic diagrams showing electronic alternatives to the interpolator of Fig. l;

Fig. 5 is a block schematic diagram showing an electronic version of the time scale stretcher of Fig. 1;

Fig. 6 is a group of waveform diagrams of assistance in explaining the operation of Fig. 5;

Fig. 7 is a block schematic diagram showing an elec-` passed into atransmission path 2 and two control paths 3, d. The transmission path 2 includes the two conducd terminal a control signal derived in the first control path 3 is applied.

The first control path 3 comprises a period marker pulse generator which may advantageously be of the type which forms the subject matter of E. Peterson Patent 2,593,694, issued April 22, 1952, and which is further described by O. O. Gruenz, Ir. and L. O. Schott in an article published in the Journal of Acoustical Society of America for September l949 (volume 29) page 487. The principal feature ofV this generator is a detector followed in tandem by a shapnig network which accentuates the amplitudes of low frequency components at the expense of higher harmonic component amplitudes. Preferably, each of these steps is carried out two or more times in succession, and all of themV may be preceded by an auxiliary shaping step as by the first unit shown. With this arrangement, as is more fully described in the patent and publication referred to above, the output of the generator 6 comprises single sharp spike of current which occurs at the instant of each major peak of the speech signal wave, o-r between the principal zero and the major peak. For present purposes, the principal zero is the last zero value or axial crossing of the speech wave preceding each of its major peaks.

When the speech is voiced, its waveform is periodic, and this marker pulse thus indicates the instant of inception of cach full period of the speech wave, and the frequency at which such pulses are repeated is the fundamental pitch frequency of the speech wave.

For voiced sounds of speech, the train of marker pulses is applied through the front fixed contacts of a relay 7 to a frequency divider 5, for example a multivibrator, proportioned to deliver one output pulse for every N input pulse. The input terminal of the divider S, bearing a train of marker pulses which recur at the recurrence rate of the major peaks of the speech wave, is connected to one input terminal 12 of a bistable multivibrator il) while the output terminal of the divider 8, bearing a train of pulses whose repetition rate is l/N times the fundamental frequency of the speech wave, is applied to a second input terminal 11 of the bistable multivibrator it). This multivibrator its and the arrangement of its input terminals and its output terminals with respect to its internal structure are such that application of a pulse to its inpult terminal 1l produces a signal on this output terminals, While application of a pulse to its input terminal 12 terminates this signal. As a result, the output signal of the bistable multivibrator l@ operates to close the switch at the inception of a period of the speech wave while cessation of this output signal, which takes place at the conclusion of this speech wave period, opens the switch 5 and so prevents conduction of the ensuing speech wave periods until the next output pulse of the divider S appears. Thus, of every N successive period of the speech wave, N-l

. of them are blocked.

' The switch 5 thus operates to chop any voiced speech wave synchronously with its successive fundamental periods, passing one out of every N such period and leaving blank intervals in the signal on the conductor 13 corresponding to the remaining N -l of such periods. The signal as thus chopped may be transmitted to a receiver station without further change. it is advantageous, however, to turn the blank intervals thus produced to account, either by reducing its vulnerability to noise, by reducing the bandwidth required to transmit it, or in some other fashion. By way of example a time scale stretcher 14 is included in tandem with the outgoing conductor 13. Its construction may be of any desired sort, for example as described in the aforementioned patent of H. W. Dudley. lt is controlled in its stretching action by the output of the marker pulse generator 6. If preferred, various electronic counterparts may be employed instead of Dudley-s apparatus. Whatever the structural details of the time scale stretcher 14 may be,

4 its operation is to stretch each full period which passes through the gate 5 over the ensuing N-l blank intervals.

The second control path 4 extends from the output terminal of the microphone 1 to a discriminator 16 as between voiced sounds and unvoiced sounds. This unit is well known and may have any desired construction. For example, it may comprise a band-pass filter, a rectifier and a lowjpass filter connected in tandem in the order named. The band-pass filter being proportioned to pass only the principal frequency components of voiced sounds, e. g., those in the range 604,000 cycles per second, and the low-pass filter being proportioned to pass only frequencies of syllabic rates, e. g., those in the range 0-25 cycles per second, the unit gives an output f of substantial magnitude when the sound waves impinging on the microphone 1 are of the voiced variety, and gives no such output signal when the sound is unvoiced.

The output signal of this discriminator 16 is applied to the winding of the relay 7 and serves to actuate it and to hold its moving contact against its front fixed contacts when the sound is voiced. This permits the output pulses of the marker pulse generator 6 to follow the path hereinabove described. When, to the contrary, the sound is unvoiced the moving contact of the relay 7 falls against the back xed contact. ln this condition of the relay 7 the output pulse train of the marker pulse generator is converted to a slowly varying signal by a low-pass filter 17 and this signal is applied to a relaxation oscillator 1S which delivers pulses at a slowly varying but nearly regular rate through the back iixedcontact to the moving Contact of the relay 7. pulses of the relaxation oscillator 18 thereupon follow the paths hereinabove described for the output pulses of the marker pulse generator 6.

The marker pulse generator 6 delivers a pulse for every major peak of the speech wave. When the speech wave is of the voiced character these marker pulses recur periodically at the fundamental pitch rate. When, to the contrary, the sound is unvoiced, the marker pulses recur at randomly spaced instants on the time scale, and at an average rate lying within the normal pitch range. For effective chopping of the wave of an unvoiced sound the chopping should occur in a regular fashion. It is the function of the combination of the relaxation oscillator 18 and the low-pass lter 17 to substitute, at the control terminals of the gate 5 and of the time scale stretcher, 14, a comparatively regular train lof pulses which recur at the average recurrence rate of the randomly distributed pulses of the marker pulse train itself.

lf preferred, the apparatus shown may be modified in such a fashion that the relaxation oscillator 18 delivers a train of pulses which recur, not at the average repetition rate of the marker pulses during an unvoiced sound, but rather at a rate corresponding to the fundamental frequency -of the most recent voiced sound; i. e., `at the rate which obtained at the instant of the change of the relay 7 from its voiced position to its unvoiced position. Apparatus of this character is disclosed in ran application of H. L. Barney, Serial No. 459,333, filed September 29, 1954, now matured into Patent 2,819,341, granted January 7, 1958.

The resulting signal, stretched in time or otherwise modified as desired, may now be transmitted to a receiver station over a channel 19 which is of much lower quality than would be required for transmission of the unmodiiied speech wave. Thus, if stretching in time be employed, the bandwidth of the channel 19 is reduced by a factor N as compared with the bandwidth of a channel required to transmit the original speech wave.

Upon arrival at the input terminal 21 the receiver station the received signal is first restored to its original dimensions in the time scale as by a restorer 22 which again may be as described in the aforementioned Dudley patent or an electronic counterpart or refinement thereof.V

The output Inasr'nuch as any such apparatus must operate in synehronism with the received wave, a period marker pulse generator 26 is shown connected to the input terminal 21 and delivering its output period marker pulses to the control terminal of the restorer 22. Its construction may be identical with that of the period marker pulse generator 6 at the transmitter station. In addition, the train of period marker pulses, one of which occurs for each period acually received, is converted into a slowly varying voltage as by a low-pass filter 23. As in the case of the transmitter, these marker pulses are always of the same polarity so that no rectifier is required, and the low-pass filter 23 operates to convert the pulse train into a voltage of varying magnitude proportional to the fundamental pitch of the received signal and of unvarying polarity.

After completion of the time scale restoration, the signal is applied to an interpolator which may comprise a plurality of delay devices 27 connected in tandem. The first of these delay devices 27a is provided with a number of lateral taps 28a, any one of which may be engaged by a conductive wiper arm 29a, and the other delay devices 27b, 27C, 27d are similar in every respect. The several wiper arms 29 are mounted on a common shaft 30 arranged to be driven by a direct-current motor 31, preferably through a reduction gear box 32. The full length of each delay device 27 is such that the propagation time of a Wave through it from its input terminal to its most distant tap is equal to the longest fundamenal period to be expected; i. e., to the reciprocal of the pitch of the deepest bass voice.

As the speech directed into the microphone 1 progresses, its fundamental frequency changes and the output voltage from the low-pass filter 23 changes proportionally. This voltage is added to a fraction, determined by the position of a wiper arm 33 of the voltage which appears across a resistor 34 which shunts a battery 35. The sum of these voltages is converted by an amplifier 36 to a driving current for the motor 31, which drives the wiper arm 33 through shaft 37. In accordance with the well known behavior of servo systems, the motor 31 then rotates, driving the shaft 37 and the wiper arm 33 in a direction and through an angle such as to bring the net voltage input to the amplifier 36 to the value zero. In doing so, it drives each of the several wiper arms 29 to similarly numbered taps on the several delay lines 27, and the tap at which each wiper arm comes to rest is that at which the propagation time from the input terminal of any delay line to that tap is equal to the instantaneous value of the fundamental period of the speech wave.

As remarked above the output signal of the low-pass filter 23 is, proportional to pitch in contrast to period. Hence the wiper arms 29 of the delay devices 27 are located, in response to the movement of the shaft of the motor 31, at positions proportional to pitch. Because each wiper arm is required to engage a particular tap of its delay line in a fashion to match individual periods of the wave, an inversion operaton must somehow be effected as between the controlling signal and resulting tap locations along the delay line. This is conveniently achieved simply by numbering the several taps of each delay line starting with the tail end of that line, instead of with its input end, and by an appropriate nonuniform spacing among the taps in accordance with the required reciprocal relation.

The input terminal of the first delay device 27a is connected to one input terminal of a weighting network 40 and each of the several wiper arms 29 is connected to another such input terminal. In the example shown in which the value of the chopping factor N is 3, four delay lines 27 are employed and the weighting network 40 has five input points. More generally, for any value of the chopping factor N the number of delay lines 27 required is' 2N-2 and the number of input points to the weighting network 40 is 2N -l. Again in the example shown the Weighting factors of the several input points to the weight- N-1 PFT Page.

A weighting network 40 of any desired` sort may be employed, a convenient one being shown in Fig. 2 for the illustrative casein which N=3. It operates in a manner well known in the art to give the proper weights to the signals applied to its various input points and then to add these weighted signals together in the fashion shown in the figure.

The output of thisnetwork 40 thus constitutes a correctly weighted sum of the various inputs. It is applied through an amplier 41 to a reproducer 42 which delivers acoustic waves of artificial speech.

For a full understanding of the operation of the network 40, consider an instant at which one of the received wave periods of the speech has passed through the first and the second delay devices 27a, 27b and has reached the third input terminal of the weighting network 40. It is then reproduced without change, inasmuch as the weighting coefficient for this input point is unity. One fundamental` period later the same wave has passed through the third delay device 27c and appears at the fourth input terminal of the weighting network 40. There it is multiplied by the factor 273. At the same' instant the next transmitted wave period arrives at the input point to the first delay device 27a and is supplied by way of a conductor 43 tothe first input point of the weighting network 40 Where it is multiplied by a factor 1/3. Hence, the reproduced wave is a mixture of 7a of the earlier wave and 1/3 of the later one. It is reproduced as audible sound by the reproducer 42 in the first blank interval following the earlier transmitted wave. Still later by one fundamental period of the speech wave, the earlier transmitted wave has passed through the fourth delay device 27d and it is applied to the fifth input point of the weighting network 40 where it is multiplied by a factor 1A. At the same instant the following transmitted wave has passedthrough the first delay device 27a and has arrived at the second input point of the weighting network 40 where it is multiplied by the factor 2A. Thus, the reproduced wave is a mixture of Va of the earlier transmitted wave and 2/3 of the later one, and it occurs at a time such that it is inserted in the second blank interval. The next event which occurs is that the following transmitted wave has passed through the first two delay devices 27a, 27b, and is applied to the third input point of the weighting network 40. From here on the cycle of operations is repeated.

While the apparatus of Fig. l illustrates the principles of the invention the shafts 30, 37, the mechanical wipers 29, 33, the follow-up motor 31 and associated apparatus impose limitations on the speed with which it can operate. For these reasons an electronic counterpart of the apparatus of Fig. l may well be preferred. Such electronic apparatus, an example of which is shown in Figs. 3 and 4,

'i is the same as the apparatus of Fig. 1, except ,QI the substitution of a relay tree and its contacts for the wipers of Fig'. 1 and circuits to control the windings of the relay tree .for the servo system of Fig. l. In order always to select a tap that will match the period of the delay lines 27 with the pitch period, the pitch voltage is converted to a number representing the particular tap on each of the several delay lines which introduces a delay approximately equal to the pitch period. To this end the pitch voltage, derived by the filter 23 from the train of marker pulses from the generator 26, is applied to a converter 50, whose details are shown in Fig. 4. The pitch voltage is sampled at the instant of a period marker pulse and applied as the initial charge across a Miller type capacitor in a phantastron delay generator 51, `as described, for example, by Chance,Hughes, et al., in Waveforms (Radiation Laboratories Series, volume 19, McGraw-Hill 1949), page 197. The phantastron 51 is of the monostable variety and its action is initiated by each marker pulse. The output derived fromV the screen grid of the phantastron 51, is a rectangular pulse starting at the instant of a marker pulse, and having a duration directly proportional to the pitch voltage. The duration of the rectangular pulse developed at the screen may be converted into a binary number by applying the pulse tothe control terminal of switch` or gate 52 which, when its conduction path is thus established, admits a sequence of very rapid pulses from a generator 53 into a string of scale-oftwo counters 54. In a typical system the pulses are generated at a 100 kilocycle rate, so that the counting process is completed in a small portion of a pitch period. The binary state of the group of scale-of-two counters, returned to zero after the completion of each count by `a reset circuit 55 under control of each marker pulse, is now directly proportional to that tap of the delay devices 27 whoseV numerical designation is most nearly proportional to the fundamental period of the speech wave. The coils 56 of a relay tree S7 are connected to the scaleof-two counters 54 in a fashion such that the presence of a' l closes the relay contacts and the presence of a 0 leaves them open. Thus the relay tree contacts operate to connect the proper tap of each delay device 27, as indicated by the counters 54, to the input terminal of the next delay device 27. As a refinement, ,the delay devices 27` may be divided into two groups, and the delays for similarly numbered taps in the two groups may be set at two slightly different times; e. g., at times which differ by one pitch period, to avoid switching when signals are present in the delay devices. A second counter and a delayed marker pulse generator may be employed for this purpose.

Returning to Fig. 1, it was remarked above that the time scale stretcher 14 at the transmitter station and the time scale restorer 22 at the receiver station might take the form shown in the aforementioned H. W. Dudley patent. However, especially for cases in which the chopping factor N has the value 3 or more, an electronic apparatus for carrying out these operations may be preferred. Fig. 5 shows one such electronic stretcher which employs a double beam storage tube 6i), for example a tube of the type dsecribed in the RCA Review for March 1949 (volume l) page 59 and is common known as a Graphecho-n. This tube generates separate writing and reading electron beams 61, 62 so that onepitch period of the speech signal can be written on the mosaic of its target anode 63 by the writing beam 61 and the reading beam 62 can scan the target 63 independently. Fig. also shows, in block diagram form, apparatus which acts to chop out an integral number of pitch periods from the speech wave and stretch each retain'ed period to till the gap left by the chopping, while Fig. 6 shows the waveforms which appear at the various similarly labelled points of the circuit of Fig. 5. The speech wave (A) from the microphone 1 Yis applied to the marker Pulse generator 6 which produces e pulse J(1C) et the Ainitial .instant of each pitch period. The resulting period marker pulses arecon-r verted by the low-pass filter 17 into a slowly-varying pitch voltage (D) which is directly proportional to the repetition rate of the period pulses (C). This pitch voltage (D) determines the rate of change of the sawtooth output wave (E) of a writing sweep generator 64 whose sweeps are initiated by the period pulses (C). The result is a sawtooth wave that has a constant amplitude, starts with a period pulse, and has the same period as the fundamental voice period. This is the signal that determines the writing rate of the beam 61 of the tube 60. A divider 65 reduces the recurrence rate of the period pulse train (C) by a factor N, so that at its output (F) a pulse appears for every N pitch pulses (C).

Each output pulse (F) of the divider 65 operates lthe multivibrator 10 and so opens the gate 5 to admit the resulting chopped speech wave (B) to the grid of the tube 60, thus to modulate the writing beam 61. At the same time it initiates the sweep (G) of a readin'g sweep generator 66 which may be a duplicate of the writing sweep generator 64, except for its sweep speed. As in the case of Fig. 1, the gate 5 is opened by each of the pulses (F) for exactly one period of each N periods and is then closed by the next one of the period pulses (C). A low-pass lter 67 controls the rate of decay of the reading sweep wave (G). The various electrodes and the erasing mechanism of the storage tube 60 are to be connected to suitablevoltages and circuits to insure proper operation. The curve (H) shows the retained portions of the speech wave, stretched in time, as they appear at the output terminal of the mosaic target anode 6,3 rof the tube 60. Associated apparatus for chopping an aperiodic speech wave, not shown in Fig. 5, may bey identical with that of Fig. l.

The operation of restoring the original time scale to the chopped and stretched speech can likewise be accomplished with an electronic storage tube 7,0 and associated circuits shown in Fig. 7 that are very similar to those described above in connection with the electronic stretcher of Fig. 5. This system employs the writing beam 71 of the storage tube itl to record the signal incoming at the terminal 21 as a charge pattern on the mosaic of a target 73 at a rate that isV so adjusted that one period of the input signal extends across the target. At the instant that the writing beam 71 starts to write each period, Athe reading beam 72 starts to read the previous period at N times the writing rate. Although these two electronA and recorded, there is a delay of one stretched period i between the reception of a stretched period and its reproduction with a restored time scale.

The controls for this tube operate as follows. Rcferrin'g to the waveform diagram, Fig. 8, the curve (I), shows the stretched wave as received. the curve (H) of Fig. 6. The marker pulse generator 26 produces a pulse (K) at the inception of each Astretched pitch period. These pulses initiate the sweeps of a Vwriting sweep generator 74 and a reading sweep generator 75. A low-pass iilter 23 converts the train of marker pulses (K) into a voltage (L) that is proportional to their repetition rate; i. e., to the pitch of the original speech signal. This pitch voltage (L) from the low-pass filter 23 controls the run down 'rate of the writing sweep generator 74 so that it produces a sawtooth wave (M) that hasV a constant amplitude and is initiated by the pitch pulses (K). The reading sweep of the generator 75 is likewise initiated by the pitch pulses (K), and the sweep rate is .controlled by the Output (L) of the `10W-pass .filter It is the same as i 23. The reading sweep generator 75 produces a sawtooth wave (P) that has N times the slope of the writing sweep generator wave (M). The result is a target current (Q) derived from the discharge of the mosaic of the target 73 of the tube 70 which reproduces the input signal with a new time scale that is N times the original one.

The reading sweep generator 75 and the writing sweep generator 74 may be conventional, as shown in Fig. 9. A capacitor 77 is charged through a cathode follower 73 to a preassigned voltage by each marker pulse (C, F or K) and it is allowed to discharge through the high variational impedance of a pentode 79 that is con'nected as a constant current load. The magnitude of the equivalent constant current Ais approximately proportional to the pitch voltage, (D or L) applied to the control grid of the pentode 79. In this mann'er the capacitor 77 discharges uniformly, and the rate of change of its voltage during discharge is proportional to the pitch of the speech wave. generator and the writing sweep generator is that the `capacitor in' the writing sweep generator 74 is N times as If desired, the above system can be modified to reduce the delay by one period in the unstretched speech by initiating the s weep of the reading' beam 71 when the writing beam '72 has completed a fraction of its excursion across the target 73. In this manner the reading beam 72 completes its reading operation of each period at the same instant that the writing beam 71 cornpletes writing that same period. This can be accomplished by initiating the-reading sweep from an amplitude discriminator that indicates when the writing sweep has completed the fraction of its excursion.

As indicated above, a speech wave is almost periodic in character, which means that each of its periods is almost, but not quite, the same in form as the prior period. The departure from exact similarity represents a gradual change in any or all the features of the wave, namely, the phase distribution as between its cornponents, the duration of the period on the time scale, and its amplitude or power. The apparatus described Vabove constructs artificial waves each of which is the properly weighted sum of an earlier wave period and a later one. Hence, it `takes all of these gradually changing features into account.

Experiments have shown that the amplitude or power of the speech wave is its property which changes most rapidly and which, therefore, is chiefly responsible in the change in the wave from each period to the next. In accordance with the invention in another of its aspects this one source of change is utilized to control reconstruction of artificial speech, the others being disregarded. This embodiment, which makes for simplification of the apparatus, may be instrumented as shown in Fig. l0,

where a speech wave originating, for example, in a microphone 1 is passed through a variolosser 80, preceded and followed by amplifiers 81, 82. The output of the second amplifier 82 is passed through a rectifier 83 and a low-pass filter 84 to provide a slowly varying control current for application to the'variolosser to adjust the magnitude of the impedance which it interposes inthe voice path. These elements constitute a conventional volume compressor, and the control current derived from the output terminal of therlow-pass filter 84 may also s The only difference between the reading sweep 10 serve, in the fashion to be described, as a measurepf the instantaneous amplitude or power of the original speech. To this end it is transmitted over an auxiliary channel 85,

The compressed output of the amplifier 82 is then chopped in the fashion described in connection with Fig. l, and through the action of like apparatus, with a chopping factor N as determined `by the construction of the divider 8. The chopped speech may now be stretched in time as by a stretcher 14 orotherwise modified to take advantage of the chopping action. It is next transmitted over a medium 19 to a receiver station. There itis first restored to its original dimensions on the time `scale through the agency of a timescale restorer 2,2 controlled by a marker pulse generator 26 'as described in -connection with Fig. l. At the Voutput point ofthe timescale restorer 22 one period out of every N periods appears, and it is the function of the apparatus to be described to insert, in the remaining N -1 blank intervals, repetitions of the prior received period modified .only as tojtheir amplitudes. The successive repetitions may be .constructed much Vas described in connection with Fig. `l by applying the chopped speech output of `the time scale restorer 22 to a sequence of N -1 delay devices. Thus, in the illustration in which N :3, two such delay devices, 86a, 86h are shown. The delay which each one of `them introduces is equal to a single fundamental period of the speech wave, and these delays `must therefore be continuously varied as the period or pitch of the speech varies. To this end, they are shown as under the control of the output of the marker pulse generator 26, smoothed to a slowly varying pitch voltage by the filter 23. `,The actual apparatus by which this pitch voltage `adjustsfthe delays in proportion to the speech wave period maybe similar to `that shown in Fig. l. t

The output of the time scale restorer Vv22 is applied through an adding network 86 to an amplifier 8,7 (a) without delay, (b) after a delay of one fundamental period and (c) after a delay of two fundamental periods. The output of this amplifier 1,87 vthus includes, following each wave actually received, two successive repetitions of it. These waves are now applied to a volume expandor 88 which operates `to expand the `volume range of the signal applied to it to the same extent as that by which the volume range of the original speech was compressed by the variolosser 80, at the transmitter station. One convenient way of achieving this result is to provide La negative feedback amplifier 39 having in its feedback path a variolosser 90 identical with the variolosser in the forward path of the compressor at the transmitter station. The impedance interposed by this variolosser90 `is adjusted by, and is under `the continuous control of, `the amplitude control signal transmitted by way `of the auxliary channel 85.

The output signal of the amplifier 89 may now be applied by way of a suitable coupling transformer to a sound reproducer 42.

Itis `a feature of the apparatusshown in Fig. lO, and especially of the employment of a compressor-expandor combination therein, that the auxiliary amplitude control signal may be continuously applied to the entire output of the amplifier 89 including receivedl periods as well as repeated periods. If preferred, and at the cost of some further complexity of the apparatus, the compressor 80 and the expander 38 may be dispensed with and each received period may be reproduced without change, the auxiliary volume control signalibeing applied exclusively to the N-l artificial repretitions of each received period.

The first form of the invention shown in Fig. 1, with or without the refinements thereof shown in Figs. 3 to 9, inclusive, comprises a linear or first order interpolator. The interpolation principle embodied in these: figures may evidently be extended in two' or more different ways:

First, it may be extended to include interpolation of the second or a higher order simply by appropriate modifications and extensions of the weighting network 4). Second, it may be extended to include interpolation based, not merely on the immediately preceding and following periods, but on still earlier and later periods as well. This extension, which provides higher precision of the reconstructed wave, may be effected by appropriate modi cations and extensions of the delay and control circuits, as well as the weighting network.

Furthermore, the two principal forms of the invention shown in Figs. l and l0, respectively, are not to be taken as mutually exclusive. Situations may arise in which it is of advantage to supplement the actio-n of an interpolator, for example, a first order interpolator as shown in Fig. l, with an auxiliary amplitude control signal as shown in Fig. 10. Such a combination is within the contemplation of the invention.

What is claimed is:

l. Speech transmission apparatus which comprises a source of a speech wave consisting of a sequence of wave portions, means for eliminating certain individual ones of said portions to leave blank intervals, means for transmitting to a receiver station wave portions preceding and following each blank interval, means for developing, at said receiver station, a sequence of artificial wave portions of which the amplitudes follow the amplitude variations of the original speech waves, and means for intercalating at least one of said artificial wave portions in each blank interval, thereby to reconstruct said original wave.

2. Speech transmission apparatus which comprises a source of a speech wave consisting of a sequence of wave periods, means for eliminating certain individual ones of said periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means for developing, at said receiver station, a sequence of artificial wave periods, means for causing the amplitudes of said artificial wave periods to follow the amplitude variations of the original speech wave, and means for intercalating said sequence in each blank interval, thereby to reconstruct said original wave.

3. Speech transmission apparatus which comprises a source of a speech wave consisting of a sequence of wave periods, means for eliminating certain individual ones of said periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means for developing, at said receiver station, artificial, wave periods each of which is a weighted sum of a preceding and a following trans mitted wave period, and means for intercalating at least one wave of said sequence in each blank interval, thereby to reconstruct said original wave.

4. Speech transmission apparatus which comprises a source of a speech wave consisting of a sequence of wave periods, means for eliminating N-l of each group of N successive ones of said periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means for developing, at said receiver station, a sequence of N-l artificial wave periods Pa, Pb, Pc PN 1, means for proportioning said artificial wave periods in accordance with the formulae:

wherein P1 represents the period immediately preceding a blank interval and P2 represents the period immediately following said blank interval, and means for intercalating said artificial wave periods sequentially in each blank interval, thereby to reconstruct said original wave.

5. Speech transmission apparatus which comprises a source of a speech wave consisting of a sequence of wave periods, means for developing an auxiliary signal proportional to the amplitude of said speech wave, means for eliminating certain individual ones of said periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means for also transmitting said auxiliary signal, means for developing, at said receiver station, a sequence of artificial wave periods, each of which is a substantial repetition of the preceding transmitted wave period, means for varying the amplitudes of said repetitions under control of said auxiliary signal, and means for intercalating at least one of said repetitions in each blank interval, thereby to reconstruct said original wave.

6. Speech transmission apparatus which comprises a source of a speech wave comprising a sequence of wave periods, means for eliminating from said sequence N-l of each group of N successive ones of said wave periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means, at said receiver station, for developing for each blank interval and from the transmitted waves which precede and follow said blank interval a group of N-l artificial wave periods, means for proportioning said several artificial wave periods as variously weighted sums of said preceding and following wave periods, and means for intercalating said artificial wave periods in said blank interval.

7. Speech transmission apparatus which comprises a source of a speech wave comprising a sequence of wave periods, means for eliminating from said sequence N -l of each group of N successive ones of said wave periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means, at said receiver station7 for mixing the transmitted waves which precede and follow each of said blank intervals in various proportions, to form a group of N *l artificial wave periods, means for arranging said artificial wave periods in order of increasing emphasis of the following wave and decreasing emphasis of the preceding wave, and means for intercalating said ordered artificial wave periods in said blank interval.

8. Speech transmission apparatus which comprises a source of a speech wave comprising a sequence of wave periods, means for eliminating from said sequence N -l of each group of N successive ones of said wave periods to leave blank intervals, means for transmitting to a receiver station wave periods preceding and following each blank interval, means, at said receiver station, for developing for each blank interval and from the transmitted wave which precede and follow said blank interval a group of N l artificial wave periods, means for proportioning said several artificial wave periods as variously weighted sums of said preceding and following wave periods, means for variously delaying each of said artificial wave periods by an integral number of fundamental pitch periods of said speech wave to form a sequence of artificial wave periods, the members of which sequence differ progressively from end to end of said sequence, and means for intercalating said sequence in said blank interval.

9. Apparatus as defined in claim 8 wherein said delayed means comprises 2N-2 delay devices, coupled together in tandem, means for applying each received wave to the input terminal of the first of said devices, means for deriving a delayed wave from the output terminal of each of said devices, and means for additively combining said delayed waves.

10. In combination with apparatus as defined in claim 9, means for deriving a period control signal from received Waves, and means for adjusting the propagation length of each of said delay devices under control of said period control signal to introduce a delay equal to the fundamental period of said speech Wave.

11. Speech transmission apparatus Which comprises a source of a speech wave comprising a sequence of wave periods, means for eliminating from said sequence N -1 of each group of N successive ones of said Wave periods to leave blank intervals, means for developing an auxiliary signal representative of the changing character of said speech Wave, means for transmitting to a receiver station wave periods preceding and following each blank interval, means for also transmitting said auxiliary signal to said receiver station, Ameans, at said receiver station, for developing for each blank interval and from the transmitted wave which precedes said blank interval a group of N -l artificial repetitions of said preceding transmitted wave, means for variously delaying each of said repetitions by an integral number of fundamental pitch periods of said speech wave to arrange said repetitions in an ordered sequence, means for varying the amplitudes of said repetitions under control of said auxiliary signal, whereby the members of said sequence difer progressively from end to end of said sequence, and means for intercalating said sequence in said blank interval.

12. Apparatus as dened in claim 11 wherein said delaying means comprises N-1 delay devices, coupled together in tandem, means for applying each received wave to the input terminal of the iirst of said devices, means for deriving a delayed wave from the output terminal of each of said devices, and means for additively combining said delayed waves.

13. In combination with apparatus as defined in claim 12, means for deriving a period control signal from received waves and means for adjusting the propagation length of each of said delay devices under control of said period control signal to introduce a delay equal to the fundamental period of said speech Wave.

14. Speech transmission apparatus which comprises a source of a speech wave comprising a sequence of wave periods, a volume-range compressor for said waves, means for deriving from said compressor an auxiliary signal representative of the amplitude variations of said speech wave, means for eliminating from said sequence N-1 of each group of N successive ones of said wave periods to leave blank intervals, means for transmitting to a receiver station Wave periods preceding and following each blank interval, means, at said receiver station, for developing for each blank interval and from the transmitted Wave which precedes said blank interval, a group of N -1 repetitions of said preceding wave, means for variously delaying said repetitions by an integral number of fundamental pitch periods of said speech wave to form a sequence of artificial Wave periods, means for intercalating said sequence in said blank interval to form a reconstructed Wave of compressed volume range without blank intervals, and means for varying the.` amplitude of said reconstructed wave under control of said auxiliary signal.

References Cited in the tile of this patent UNITED STATES PATENTS 2,098,956 Dudley Nov. 16, 1937 2,732,424 Oliver Jan. 24, 1956 2,766,325 DiToro Oct. 9, 1956 UNITED STATES PATENT OFFICE Certificate of Correction Patent No. 2,860,187 November 11, 1958 Edward E. David, Jr., et al.

It is hereby certified that error appears in the printed specification of the above numbered patent requiring correction and that the said Letters Patent should readA f as corrected below.

Column l1, lines 68 and 69, the formulaJ should appear as shown below instead of Signed and. sealed this 17th day of March 1959.

[SEAL] Attest: KARL H. AXLINE, Attestz'ng 0770e1'.

ROBERT C. WATSON, ovmnssoner of Patents.

Images