Variable delay line signal processor for sound reproduction

[57] ABSTRACT

An electrical delay line which has variable delay controlled by a signal input thereto is connected in a sound signal channel for signals such as human speech, to compress or expand the sound signal waveform depending on whether the time delay in the line is increased or decreased. By periodically sweeping the delay line from minimum to maximum time delay or vice-versa, repeated segments of a continuous sound signal waveform are processed so that an Output audio signal can be obtained having the original frequency components of the signal and occupying a time duration which is equal to or smaller or larger than the original sound sequence with the successive segments of the signal processed by the variable delay line assembled with regard both to the significant parameters of human speech or other coding and the electrical conditions imposed by the system to produce a composite audio signal which is an intelligible replica of the original and substantially free of annoying aberrations introduced by the delay line processor. Variable delay using analog or digital signal storage is also provided.

Claims, 64 Drawing Figures

<img file="US3786195A_D0001.tif" />

PATENTEDJAK15I974 3.786,195

SHEET OlOF 10

SPEECH COMPRESSION 333-5000 HERTZ

Recorded Speed=S Maximum Accumulated Signal Delay= 6 ms

<img file="US3786195A_D0002.tif" />

-MAX.= IOms

<td> u- X Σ —\ k— 3 Lu v /¹ _^ω r hi ω &lt;. al ΐ ° H n w t S z &lt; σ τ, ¹¹ to il II D - ω _ S S -.u . ° Ξ to 5 ~ 3 * . ....................... —V“ _: &lt;3 U. _E</td><td> ’ 1- X ’ to to &amp; ^ω z ii ii 5 z S ° £ 0 jJj M_ CT X *** INVENTOR MURRAY M. SCHIFFMAN BY ATTORNEY</td>

PATENTED JAN 151974

3,786,195

SHEET 02 OF 10 ro o

o

<img file="US3786195A_D0003.tif" />

PATENTED JAN151974 3,786,195

SHEET 03 OF 10

<img file="US3786195A_D0004.tif" />

Max. Delay Line Length

BY_: .

ATTORNEY

PATENTED JAN 15 1974

3.786,195

SHEET 04 OF 10

DISJUNCTIVE TRANSITION

<img file="US3786195A_D0005.tif" />

ATTORNEY

PATENTEDJM151974

3.786.195

SHEET OS Of 10

<img file="US3786195A_D0006.tif" />

(a) (b)

<img file="US3786195A_D0007.tif" />

Chunk Length = Sample Period - gap (c) (d)

<img file="US3786195A_D0008.tif" />

EXPANSION

INVENTOR

MURRAY M. SCHIFFMAN

BY

ATTORNEY

PATENTED JAN 15 1974

3.786,195

SHEET 06 OF 10

<img file="US3786195A_D0009.tif" />

Sample Period Control----»·

Pulse-Train Gen.

<img file="US3786195A_D0010.tif" />

—Delay 95

T r— ¹ Sample — Period Control j I Changes Pulse Spacing

I I '

L_J

<img file="US3786195A_D0011.tif" />

Blank B, Sam.?

Blanking B____

Speech Samplej

<img file="US3786195A_D0012.tif" />

Manual Control of Pulse Width

Allows Change in Blanking Period

<img file="US3786195A_D0013.tif" />

INVENTOR

MURRAY M. SCHIFFMAN

BY

ATTORNEY

PATENTEDJAN15B74 3,786.195

SHEET 07 OF 10

<img file="US3786195A_D0014.tif" />

<img file="US3786195A_D0015.tif" />

(a )

FIG.I4 4_(b) (c)

<img file="US3786195A_D0016.tif" />

<img file="US3786195A_D0017.tif" />

ATTORNEY

PATENTEOJAM151974 3,786.195

SHEET 08 OF 10

<img file="US3786195A_D0018.tif" />

<img file="US3786195A_D0019.tif" />

<img file="US3786195A_D0020.tif" />

MURRAY M.SCHIFFMAN

BY

<img file="US3786195A_D0021.tif" />

ATTORNEY

PATENTED JAM 151974 3.786,195

SHEET 09 OF 10

<img file="US3786195A_D0022.tif" />

BY iSJLpTp

ATTORNEY'

PATENTED JAN 151974

3,786,195

SHEET 10 OF 10

<img file="US3786195A_D0023.tif" />

<img file="US3786195A_D0024.tif" />

<img file="US3786195A_D0025.tif" />

ATTORNEY

3,786,195

VARIABLE DELAY LINE SIGNAL PROCESSOR FOR SOUND REPRODUCTION

BACKGROUND OF THE INVENTION

The field of this invention is the processing of human speech signals or similar signals for ultimate comprehension by the human listener at substantially the natural or normal frequency component distribution but at time interval durations which are usually different from the original time duration of the speech utterance.

Sound compression and expansion systems which utilize relative motion between a magnetic tape tecord medium and the air gap of the pick-up head which senses the recorded signal on the magnetic medium are well known as exemplified by the patent to Schuller 2,352,023. Devices of this type suffer from the usual operational, cost and weight limitations involved in equipment which utilizes substantial mechanical motion components. A delay line version of time compression and expansion for real time signals is also well known as shown, for example, in the patent to French et al. 1,671,151, where a voice signal is propagated along a delay line and a movable pick-up repeatedly scans the delay line to sense the signal propagating therethrough with the relative velocity between the pick-up head and the propagation velocity of the wave in the medium giving bandwidth compression or expansion for the purpose of transmitting the signal over a narrow band telephone line. Later workers in the field eliminated the mechanical motion portions of systems such as French et al. by substituting electronic switching sequentially along taps on the electrical delay line thereby providing frequency compression or expansion of the signal for transmission over a narrow band line. The application of the sequentially scanned tapped delay line for modifying the time duration of recorded speech signals without altering the frequency components thereof is disclosed by Greenberg et al. U.S. Pat. No. 3,480,737. By relating the speed of scan of the delay line to the velocity of propagation in the delay medium and the relative speed of the reproduction of a recorded message compared to the speech utterance from which it originated Greenberg et al. achieve time expansion or compression of a recorded speech signal without altering the frequency components thereof.

Another form of frequency-time transformation is known in the art using a signal controlled variable time delay line for error correction. Systems of this type detect an unwanted frequency effect due to time irregularities in the pulse train of a repetitive signal or such variations in an audio system where the speed of the record medium past the pick-up head is subject to periodic variations which result in the production of the audible irregularity known as “wow.” In reproducing the original signal these systems eliminate speed errors by servo-control of the time delay of a delay line which is interposed in the signal channel. Audio systems which employ a reference or timing signal track with variable delay line compensation for playback speed are shown for example in FIG. 9 of Coleman, Jr. U.S. Pat. No. 3,202,769 who shows an open loop servo. Woodruff U.S. Pat. No. 3,347,997 shows a closed loop servo which adjusts playback speed to compensate for the relatively low frequency “wow” component whereas imperfections of a high frequency nature (i.e., “flutter”) are compensated by a variable delay line. Such compensation systems depend for their operation on the repetitive character of the error signal and thus by providing a delay line of maximum delay time adequate to compensate the maximum expected error do not encounter the problem which would be presented for a system that required indefinite increase in the delay quantity to continuously modify the time-frequency character of a signal.

The systems described in the patents to Schuller, French et al. and Greenberg et al., when used to reduce the frequency of a speech signal, while compressing the time in which a given segment of speech is reproduced, inevitably involve discarding a portion of the original speech wave. The ratio of the speech signal discarded to that which is used is directly related to the compression ratio and the discard loss is inherently and fundamentally related to this process of reducing the frequency and compressing the time for the processing of a given passage of speech. Since the portion of the speech which is reproduced alternates with portions which are discarded the problem of merging to reproduce sections in continuous time slots presents some problem and various solutions have been offered.

Thus Schuller suggests a skewed air gap in his rotating magnetic pick-up head or skewed tape approach to the point of contact with the rotating air gap so that the arrival and departure of the magnetic tape record relative to the air gap will occur gradually as the skew provides a transition from zero to full air gap contact with the recording medium.

The patent to French et al. suggests a number of alternatives including the use of two spaced transducers rotating in unison with respect to the delay medium such that the message reproduced in compressed time in one transducer has superposed thereon from the other transducer the message which would ordinarily have been discarded by the first transducer.

These efforts to compensate for the discontinuity caused by periodically discarding portions of the continuous speech wave have, in general, been possible only because the output signal is derived from a sampling along the propagating speech signal path since with such sampling the signal stored in the line can be readily accessed at any point along the line. For the above-mentioned variable delay line correction systems, discard has not been considered or compensated.

BRIEF SUMMARY OF THE INVENTION

The present invention provides compressionexpansion systems for speech or other coded signals in which the active frequency-time conversion element is a signal responsive delay line which is directly interposed in the path between the signal source and the ultimate reproducer or utilization device which receives the converted speech wave. Such a system does not avoid the inherent problem in time compression occasioned by the discontinuity resulting from discarding alternate portions of the speech wave. The discard portion of the speech wave may be stored or cancelled in the delay line or diverted from entering the input terminal of the delay line which is directly in the signal channel. In any case this signal and transients produced by line switching must be discarded as the variable delay line is repetitively controlled by the delay signal between its minimum and maximum delay values. Line switching and discard occur coincidentally with the requirement for making contiguous two originally spaced

3,786,195 portions of the speech wave and hence both of these functions are accomplished by a number of alternative embodiments herein disclosed which function in a manner consistent with requirements imposed by the parameters of the speech signal itself.

Accordingly, it is the principal object of the present invention to provide a speech compression-expansion system which utilizes a signal controlled variable delay line located directly in the signal channel between the signal source and sound reproducer which delay line is repeatedly sequenced between maximum and minimum delay values to modify the frequency-time characteristic of the sound reproduced from the original signal.

It is a feature of the invention to control the input and output bandwidth of the system for the speech frequencies which are required to be reproduced for intelligibility in relation to the compression ratio thereby to exclude those frequencies which would produce distortion and intermodulation due to insufficient sampling rate or excessive phase shift per delay stage for the high frequencies and inadequate output chunk length to attain frequency conversion for the lowest frequencies.

A further feature of the invention provides maximum discard intervals as determined by the maximum delay line length which in relation to the compression ratio limits the actual original speech message discard to a value which minimizes the loss of significant cues or transitions in the speech code thereby minimizing the loss of information content transferred to the listener.

A still further feature of the present invention is to provide, in a system utilizing variable delay line speech compression or expansion, for signal processing at the point of juncture of two reproduced speech portions to suppress distracting noise components and also to avoid the introduction of false cues which could modify the information conveyed in the subsequent speech segment. To this end, if required to suppress such noise, the transition between successive reproduced speech samples can be modified by simple transfer function selection or control, or the transition can be eased by the introduction of synthetic or speechderived signal portions to approximate a smooth transition within a time interval which does not lose actual cues and under such conditions that do not introduce false cues.

These and other features and advantages of the invention will be apparent from the following detailed description taken in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGS. 1(a) to 1(h) show diagrams representing the reproduction of a message recorded on magnetic tape and the operation of the system of the present invention at different compression ratios.

FIGS. 2(a) and 2(b) are diagrams for compression and expansion, respectively, showing input-output signal time relations.

FIG. 3(a) shows a set of curves for a specific maximum time delay relating various parameters involved in the processing of speech at various compression ratios greater than one and FIG. 3(b) indicates similar relations for ratios less than one (i.e., expansion).

FIGS. 4(a) to 4(f) show waveforms useful in describing forms of processing of a transition between adjacent reproduced speech samples.

FIGS. 5(a) to 5(d) show a set of curves representing active processing of the transition between adjacent samples.

FIGS. 6(a) to 6(e) show waveforms useful in describing the use of two delay lines to effect transition between adjacent speech samples.

FIG. 7 shows a block diagram of a speech compressor-expander system in accordance with the invention.

FIGS. 8(a) to 8(d) show waveforms useful in describing the operation of the system of FIG. 7.

FIG. 9 shows a block diagram of a dual delay line system in accordance with the invention.

FIGS. 10(a) to 10(d) show waveforms useful in describing the operation of the system of FIG. 9 for compression.

FIGS. 11(a) to 11(d) show waveforms useful in describing the operation of the system of FIG. 9 for expansion.

FIG. 12 is a partial block diagram of a modification.

FIGS. 13(a) to 13(c) show waveforms useful in describing the operation of circuit of FIG. 12 for compression.

FIGS. 14(a) to 14(c) show waveforms useful in describing the operation of the modification of FIG. 12 for expansion.

FIG. 15 is a partial diagram of a dual delay line binaural system.

FIGS. 16(a) to 16(f) show waveforms useful in describing the operation of the system of FIG. 15.

FIG. 17 is a partial block diagram of a speech processor in accordance with the invention using an analog shift register at the variable delay element.

FIG. 18 is a block diagram showing gap filling with signal continuity in a system similar to FIG. 17.

FIG. 19 is a partial showing of an embodiment of the invention with variable delay provided by an r-bit parallel digital shift register.

FIG. 20 shows an embodiment of the invention with variable delay provided by a serial digital shift register.

FIG. 21 shows an embodiment using an analog storage memory matrix for variable delay.

FIG. 22 shows an embodiment using an r-bit digital random access memory.

FIG. 23 shows a logic diagram of a directional zero signal level gating control in a dual line system.

FIG. 24 shows waveforms useful in describing the operation of the circuit of FIG. 23.

FIG. 25 shows graphically the clock frequency and maximum signal frequency for the system of FIG. 17.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

The description of the preferred embodiments will be preceded by a discussion of the parameters of the speech signal particularly as they relate to speech compression for reproducing a given speech message in a shorter period of time. Because of the fundamental and unavoidable limitations involved in speech compression, the following discussion will proceed with primary attention directed to the method and apparatus used in

3,786,195

Compression mode. Compression mode operation in the time domain results in the discard of a fraction of the original information directly proportional to the compression factor which is also the factor by which the time to present a given speech sequence is de- 5 creased. The method and apparatus are usable however in expansion mode and the considerations involved for use of reproduced signals which occupy a greater length of time than the original speech utterance will be described separately hereinafter. The system is also ca- 10 pable of frequency transformation without a corresponding time change to achieve a desired frequency signal, such as may be involved in generating speech in a medium having a propagation velocity different than ^air·......._........................... 15

Referring to FIG. 1, a particular system using a delay line which provides maximum time delay of 6ms for the final portion of the sample is shown. Assuming that the speech signal to be processed is limited to frequency components between 333 and 5,000 Hz, certain param- ²⁰eters of the playback system for compression can be defined. A magnetic tape 21 has the speech signal recorded thereon of which the lowest frequency component at 333 Hz is depicted by sine wave 22 with the tape being drawn past a pick-up transducer as it is ²⁵wound on a take-up reel 24 at speed S. The electrical signal produced by the transducer 23 passes through a compression processor 25 and is reproduced as an audible signal from speaker 26.

................... -.................................._ _ 30

The system 23-26 shown on line (a) of FIG. 1 just described reproduces the recorded signal on the tape 21 without frequency or time change if the speed of takeup reel 24 draws the tape past the transducer 23 at the speed of recording S and for this condition processor 35 25 would introduce a fixed constant delay time of any value. Thus in line (b) of FIG. 1 where c=l, the reproduction of the 333 Hz sinusoidal signal without change other than fixed phase delay (which has been ignored) is shown. ₄₀

For speech compression the tape speed is increased by a factor c and the processor 25 is operated to vary the delay linearly from its minimum to maximum value. As shown in curves (c), (d) and (e) of FIG. 1, a compression ratio of c = 2 restores, for a 6ms final signal ⁴⁵delay requiring 8ms delay line, a 12ms portion of the original recorded wave 22 and in so doing retains half and discards half the amount of signal originally occumaintained at 6ms corresponding to 8ms in the delay line at the end of the sample, a 6ms portion of the original information representing 12ms at the recorded speed is discarded and on line (c) is labelled “discard.” This discard includes cycles 5,6,7 and 8 of the original wave 22 and represents the gap in the information content between successive chunks which are reproduced as audible signals. This audible output is represented on line (d) where the chunk is depicted as a piece Of tape 31 played at speed 2S and containing cycles 1-4 which after processing is effectively stretched into a piece of tape 32 occupying the original 12ms of recorded time and containing the cycles 1-4 at their original recorded frequency. In line (d) it will be noted that the next cycle reproduced is cycle number 9 of the original wave after cycles 5—8 inclusive have been discarded. The representation in line (d) of a smooth transition between the end of cycle 4 and the start of cycle 9 should not be taken as representative of real signal conditions as would be obvious from a consideration of an actual signal as opposed to the idealized signals presented in FIG. 1.

Lines (f), (g) and (h) of FIG. 1 illustrate the situation which prevails when the compression ratio is equal to five with the tape speed drawn past transducer 23 at five times the recorded speed S. With a final signal delay of 6ms maximum this compression ratio results in a chunk length of 1.5 containing 2 cycles of the 333 Hz wave 25 of line (a) and again a discard interval of 6 ms equal to the final signal delay and corresponding to a delay line length of 10ms at the end of the sample. The information gap, however, has increased to the point where the last half of cycle number 3 and first half of cycle number 13 and all the intervening information in the original recorded wave have been lost in the discard and this gap in the message represents 30ms of the originally recorded speech utterance.

The relations among the parameters of a speech compression system and those pertaining to the information content of the speech coding interrelate in a manner to specify optimum conditions and place outside limits on the mode of operation of the systems of the invention for a given intelligibility factor. These parameters can be examined with respect to a particular system for various compression ratios and for this purpose the system parameters for a system using a delay line with a maximum final signal delay AT_max of 6ms are set forth in the following table.

TABLE I

<td colspan="7"> Typical Parameters for Speech Compressor</td>

<td rowspan="2"> Comp.</td><td rowspan="2"> Line</td><td colspan="5"> Chunk/Discard Ratio</td>

<td> (Playback</td><td> (Recording</td><td> Sample</td><td> Rep</td><td> Cycles</td>

<td> Ratio</td><td> Length</td><td> “c”Time)</td><td> Time</td><td> Period</td><td> Rate</td><td> Sample</td>

<td></td><td> 4· T„,</td><td> τ_(λ/δτ_μ„</td><td> ToHt/cAT^nj.</td><td> T</td><td> 1/T</td><td> (fmin=</td>

<td> c d</td><td> (ms)</td><td></td><td> (ms)</td><td> (ms)</td><td> l/P.</td><td> Hz)</td>

333

<td> 1.25</td><td> 2/9</td><td> 6 %</td><td> 24/6</td><td> 30/7.5</td><td> 30</td><td> 33.3</td><td> 10</td>

<td> 1.5</td><td> 2/5</td><td> 7 1/5</td><td> 12/6</td><td> 18/9</td><td> 18</td><td> 55.6</td><td> 6</td>

<td> 2</td><td> %</td><td> 8</td><td> 6/6</td><td> 12/12</td><td> 12</td><td> 83.3</td><td> 4</td>

<td> 3</td><td> 1</td><td> 9</td><td> 3/6</td><td> 9/18</td><td> 9</td><td> 111</td><td> 3</td>

<td> 4</td><td> 6/5</td><td> 9 3/5</td><td> 2/6</td><td> 8/24</td><td> 8</td><td> 125</td><td> 2 %</td>

<td> 5</td><td> 4/3</td><td> 10</td><td> 1.5/6</td><td> 7.5/30</td><td> 7½</td><td> 113</td><td> 2 ½</td>

pying 24ms of recorded time. This retained portion is designated a “chunk” and is shown in speeded form prior to processing on line (c) of FIG. 1 to include cy-⁶⁵cles number 1,2,3 and 4. By virtue of the compression process and since the maximum final signal delay is

The basis for the frequency-time transformation employed in the present invention can be derived as follows. Consider a sine wave V=E sin ait recorded with a tape recorder. If the tape is played back at c times the original recording rate, the result is

3,786,195

V = E sin c wt (i) where c is called the compression ratio. If c&gt; 1 the time is compressed for any given speech passage and if c&lt; 1 time is expanded by the factor e where e = 1/c.

If the signal is then applied to a delay line in which the delay of the line is caused to increase linearly with time at the rate d, so as to cause the average delay of the signal to be c' which represents the delay any point on the waveform will experience in passing through the line, then the signal (1) becomes

V = E sin (c — c') ωί (2)

The original signal is restored if the delay is Ct C t(f₀ restore) (c 1 )t (3) such that c' = d/2 (c + 1), the average delay rate of the line (4&gt; which is derived as one half the sum of the final and initial delay values of the delay line, which when multiplied by time, t, thus yields the total delay, c't.

FIG. 2(a) shows a plot for a given signal sample of signal output time, t_out, vs. the corresponding input time, t_iB. Thus a line with a slope of 4 represents signal of four times the original frequency or speed of presentation, and one-fourth the periodicity while a line with a slope of 1 represents the resultant restored, or unchanged signal. In order to convert such a signal (as represented by the line I of slope c = 4) to one corresponding to line II with a slope of 1, and with a corresponding frequency decrease, it is necessary to increasingly delay the input signal, ct_ln · by an amount c't [or (c—l)t] as shown at line III. Thus a signal chunk, T_jB24 has an ordinate which intersects III at the ordinate value c'T_in and this value when added to the time abscissa value at the point cT_jB on I, delays the signal to T_oui on line II. The delay dt introduced by the delay line is shown by line IV. Such a delay line has the effect of delaying the instantaneous signal, through a linearly increasing amount dt for the interval from t_ln to t_oul as shown by line IV. Thus as in the case of the end signal at time t=T one-half the sum of the initial delay, dT_fB, and the final delay, dT_oW yields an average delay value on line IV of c'T_(B, the amount required for restoration.

Hence, c'T_in = (dT_iB + dT_mJ2) (c-\)T_in = d/2(T_in + cT_tn) d = 2(c—1/c+l) j

In more general terms the restoration may be achieved by cumulatively delaying an input singal, t_iB, by an amount

J tout f(t) _ „ ,, ^ = c't_in=(c-l)t_in , tin &amp;out — ^fin .

For a linearly varied delay line with a rate of change in delay, d,f(t)=d t and _C'T_in= (c— 1) Tin== P”¹ r ^d_ T d P -]Tou. (&lt;i-l) T_in= ~2 Ti_BJtm 2 (c-l)T_iB (6) from which we obtain (4) c’T_in = d/2(c+V)T_tn¹⁰ Hence d = 2 (c- l/c+ 1) (7)

Thus for c s 1, 0 e d &lt;2 c&lt;l, —2&lt;d&lt;0

In FIG. 2(b) the corresponding relations for signal expansion are shown. Line I_e with a slope of one-fourth represents a signal of one-fourth the original frequency ₂₀ or speed of presentation. In order to convert such a signal to one corresponding to line II with a slope of 1, and with a corresponding frequency increase, it is necessary to decreasingly delay the input signal, ct_in (=( 1/e) t_in = (1/4) t_iB) by an amount c't (=( 1—e/e) t = —%f) from an 25 initial delay of c'T_0U(. This amount of delay, c't', at any point shifts the signal to the corresponding ordinate value on line II. The delay dt' introduced by the delay line is shown by line IV_e. Such a delay line has the effect of delaying the instantaneous signal by a linearly 30 decreasing amount d t' for the interval from t_in to t_outas shown by line IV_e. Thus as in the case of the initial signal at time, Z=0, one half the sum of the initial delay, —d T_in and its final delay, —dT_0U( yields an average delay value on line IV_e of —c’T_t„

Hence, c'T_iB = (dT_in + dT_0Ut/2) (1-e/e) T_in = d/2 [T_(n + (T_ln/e)] d=2(l—e/l+e)

The process of linearly increasing time delay cannot ⁴⁰ continue indefinitely, and from time to time the delay line must be returned to its original length. If this process is repeated at periodic intervals, provided the interval is longer than the period of the lowest frequency component of the signal, chunks of the original signal ⁴⁵ will be played back at the angular frequency (c —ο')ω and the rest discarded. When (3) and (4) are satisfied, the system operates as though sections were cut out of the original tape, pasted together, and played back at normal speed. The sections of signal are heard at the ^υ correct frequency but the information is transmitted in a shorter time (if c&gt;l). The speech has been compressed to 1/c of its original length.

The values set forth in Table I have been plotted in ₅₅ FIG. 3(a). For any given compression ratio the sample ³ time is given by the curve T_0B1 and the chunk length is shown by the curve T_iB. The difference between these two curves is the discard which is equal to the final delay to the signal at the end of the sample period (6ms ₆₀ in the example shown in FIG. (3a). Entering the curve at any compression ratio, such as c = 5 in FIG. (3a), one obtains the chunk and discard times for the tape running at c times the recorded speed and these values projected to the time axis show the actual original re66 corded time for the respective chunk and discard portions. As indicated for c = 5 the chunk is 1.5 ms long and the discard is 6 ms long representing respectively 7.5 ms of recorded and reproduced information and 30

3,786,195 ms of discarded information. This latter value is represented by the quantity c ΔΤ_ΜΜ which is also plotted in FIG. 3(a)

For a speech signal in which the lowest frequency 333Hz has a period of 3 ms, a chunk length of 1.5 ms at c = 5 corresponding to 7.5 ms of recorded time will contain 2.5 cycles of the 333H. signal. For any higher frequency components in the speech signal more cycles will be contained in the 1.5 ms chunk. The length of the chunk should exceed the period of the lowest frequency to be passed (i.e., should include at least a full cycle) otherwise satisfactory compression will not be obtained. As indicated in FIG. 3(a) below the time axis at 3ms, the 333Hz signal is processed at sample periods approaching 3 ms would with its samples reassembled accordingly produce a compressed output of poor quality since the sampling would then be causing a disruptive discontinuity for nearly every cycle of the 333 Hz signal processed. Sample periods less than 3ms would not permit completion of any one cycle so that the resultant reassembled output would not only contain the said disruption but would also begin to exhibit a basic change in its frequency characteristic in the form of waveform compression by truncation to produce false frequencies. While this condition does not represent a real condition for a speech wave due to the complexity of the waveforms, this principle is controlling and sample periods less than the period of the lowest frequency wave in the speech signal will not provide proper compression.

Sample periods greater than the period of the lowest frequency wave will produce compression and an interval of disruption exists from the region where the sample period is only slightly greater than the period of the lowest frequency wave as indicated on the time axis between 3 ms and 6 ms in FIG. 3(a). The result obtained within this period of disruption is a distorted expanded wave in which the effect of disjunctions between samples becomes extremely severe as the single cycle point is approached and diminishes as the number of cycles in the sample increases. As a practical matter two and one-half cycles per sample is indicated as the desired limit in FIG. 3(a) but in general the more cycles in the sample the less the disturbance factor.

In order to avoid the extreme distortion produced by waves which have a longer wavelength than the sample period, these lower frequencies should be filtered out before the speech signals enters the delay line otherwise these disjointed and highly distorted waves will be propagated down the line and intermodulate with the desired signal and may severly degrade the system performance.

For lower values of the compression ratio than c = 5, and keeping ΔΤ„_αχ = 6ms, the chunk length increases with the result that the actual time sample increases to greater than 7.5 ms and therefore more than the minimum number of cycles for the lowest frequency wave component will be present in the chunk. Thus it would be at the user’s option to operate the line over less than the 6ms indicated delay for ΔΤ„_ΟΙ to reduce the amount of discard.

Considering the discard portion of the sample as a constant 6ms long at the compressed rate of playback, the actual information loss is the compression ratio times 6ms so that with c = 5 the actual information discarded for each sample is 30ms of recorded time. As shown on the time axis of FIG. 2 this is the interval from 7.5 ms to 37.5 ms and the relation of this loss of information to the intelligibility of the reproduced speech signal must be examined.

In general, human speech represents an extremely complex coding of a relatively limited set of sounds called phonemes which taken in context with the various attributes of the speech code such as the voicedunvoiced components, pitch, formant frequencies and the continuum of sound pattern represented by sotlnd energy (and the absence thereof) connected by the all important transititions between the temporal components thereof constitutes an acoustic stream of infinite variety and versatility. The ability of the human ear to receive this acoustic message and the ear-brain system to decode the message is not altogether understood since it appears that the readily comprehended information rate far exceeds the mere acoustic response characteristics of the ear as a receiver.

Fortunately, the ability of the ear-brain system to comprehend the message which is conveyed by human speech signals is sufficiently good to permit large portions of the actual acoustic stream to be lost or discarded without significant loss in the perception and comprehension of the message information content of the acoustic signal. Since the comprehension of message content decreases more rapidly than the recognition of individual words as the message is presented to the listener at increasing rate, the problem associated with the discard of a portion of the signal stream can be resolved in favor of comprehension and short of the point where intelligibility of individual words deteriorates. This latter point is reached where the loss or alteration of transitions or other cues representing the connection between a consonant and vowel sound results effectively from the discard of much or all of a given cue or cues so as to alter the apparent information content of cotiguous concatenated chunks. Even before the point of absolute loss of intelligibility is reached the limit of tolerance due to discomfort for sustained listening occurs as a result of the unnatural sounds and the fatigue which develops in the intense concentration required in attempting to extract the information content in the presence of excessive time clipping.

For the purpose of speech compression the loss of intelligibility can be associated with discarding portions of the message containing significant cues or phonemes which components vary in length with the shortest being approximately 10ms to 20ms long. These short cues do not dominate speech but occur with sufficient regularity to make their systematic loss undesirable and hence a desirable upper limit for the discard period would be considered to be 30 ms and preferably closer to 15ms. With this limit set for intelligibility of the reproduced syllables and words the rate of presenting a given message can be increased to the comprehension limit for any given listener and degree of difficulty of the subject matter with minimum concern for the limitation which would be imposed by permitting loss or distortion of the word content or the generation of false cues from the concatenated message chunks. FIG. 3(a) indicates the recording time discard relation to compression ratio as the linear function cAT„,„_a. with the range from 15 ms to 30 ms designated the discard uncertainty range. Thus the 6ms discard at c = 5 projects to include the real time recording interval from time t = 7.5 to t = 37.5 which approaches the

3,786,195 upper limit permitted for discard without undue loss of intelligibility as required not to contribute significantly to the loss of comprehension in the message perceived. Smaller values of c result in smaller actual discard time and hence the intelligibility is improved especially for those cues which are at the lower end of the time scale, i.e., in the neighborhood of 10ms.

While Table I and FIG. 3(a) represent parameters for a typical speech compression system having a final signal delay of 6ms and define the limits of operation within fairly narrow limits, it will be appreciated that the principles involved can be adapted for use over a wider range of operation. Thus the variation of the actual frequency band of the speech signal and the maximum length of the delay line are both important design factors which influence the selection of the chunk-todiscard ratio and sample period for a given range of the compression ratio c. On the other hand, the actual frequency range of the signal has an important bearing on the design of the delay line which must accommodate the frequency spectrum present in the signal as to such quantitative and qualitative factors as the voice pitch, the presence of all or only some of the format frequencies for an individual Voice and the width of the signal spectrum over which linear phase-frequency properties must be preserved. The ultimate system used however will embody design choices of the factors involved within the broad limits herein defined.

FIG. 3(b) is a plot of corresponding relations for signal expansion showing the initial gap, output chunk and maximum delay line length variation with expansion ratio e for a given input sample interval T_(n. The output gap occurs at the start of each sample period and thereafter for the balance of the sample period the reduced frequency time-expanded output chunk appears. The maximum delay d T_(n required is also shown as a function of the expansion ratio e.

One aspect of the speech compression system described in connection with FIG. 1 has not been treated, namely, the audible output of the transducer 26 when the variable delay processing unit 25 is switched from maximum to minimum delay at the end of the sample period. Just prior to switching the delay line is loaded with the speech signal which is to be discarded and if the line is instantaneously switched to zero delay all of this information unless cancelled or predeleted, will be presented in highly condensed form in the output signal. As a practical matter with conventional delay lines utilizing R and L or C components there will be a time interval required for switching the line from maximum to minimum delay and it has been found that even if the line does not contain signal information this switching of a line has a significant minimum time constant associated with it which produces a disturbing transient audible in the output signal. The repetition rate of this transient is the reciprocal of the sample period. Because of the limitations imposed by the parameters of the system as previously set forth herein, this switching frequency and spectral components of the transient itself will always be within the audio range and thus present as a highly disagreeable intermoduation component in the audio output of the device. The present invention provides a number of implementations for transient suppression and message gap bridging arrangements for the purpose of minimizing the disagreeable noise effects involved. In more elaborate systems the substitution of pseudo or real message components further improves the transition from one sample to the next and can be adapted to fill in a portion of what is discarded in the compression process.

Referring now to FIG. 4 a portion of the 333 Hz wave at the transition point illustrated in FIG. 1(d) has been reproduced in which cycle 4 and cycle 9 of the original recorded 333 Hz wave are shown as a smooth uninterrupted sine wave. The junction between the end of cycle 4 and the beginning of cycle 9 at point 41, although shown as a continuous portion of the sine wave, is in actuality, as previously stated, almost never so related in the non-selective periodic sampling of independent complex waveforms and thus instead of a smooth transition point 41 a disjunction between the end of one chunk and the beginning of the next chunk in successive samples is to be expected. This disjunction could undoubtedly be accommodated with no loss of intelligibility if the transient from switching the line (either loaded or unloaded) did not have to be dealt with at exactly this point in time. Since this transient is responsible for a highly annoying audible output from the system it must be eliminated and for this purpose a gating signal as indicated in FIG. 4(b) may be applied symmetrically with respect to the transition point 41 to produce the output signal shown in FIG. 4(c). By making the gate long enought to encompass the transient resulting from switching the line, the audible noise so generated is eliminated. The improvement obtained by this expedient, while significant, is not ideal since the introduction of the gate signal within the audio range is itself audible as a repetitive disjunctive gap which intermodulates with the audio signal. This effect can be reduced by using an output filter designed for the particular repetition rate and gate width to smooth the abrupt transition shown in FIG. 4(c) and this output response is indicated in FIG. 4(d).

A further improvement is possible by using the gating signal as a gain control signal and tapering the “off” and perhaps the “on” transitions of the gate so that a gradual transition of the audio output from “off” to “on” is accomplished and a relatively smooth transition as indicated in FIG. 4(f) results. The object is to minimize the gap effect which in itself has an audio characteristic and can act like a cue. Tapering the trailing edge of the gate helps this considerably whereas an anticipating start (or relative delay of the speech signal) would be preferable for gradual onset for the leading edge. With these relatively simple expedients the smooth transition between adjacent chunks which are disjunctively joined by the operation of the compression-discard process are achieved in a manner which is satisfactory for many applications.

Referring now to FIG. 5, the more elaborate arrangements for bridging the gap between adjacent samples will be described. As shown in FIG. 5(a) a disjunctive transition which is the norm to be expected represents a sharp discontinuity in the message signal and has superposed thereon the noise transient from switching the line as previously described. By introducing a gate signal FIG. 5(b) of sufficient width to encompass the line switching transient and conditioning the gate to coincide with a zero level and same direction of change for the adjacent signals being processed a zero level gating transition as shown in FIG. 5(c) can be achieved. This transition which is free of line switching noise and essentially continues an existing zero amplitude signal level during the interval of the gate has been

3,786,195 found to provide little or no disturbance to the average listener.

Because of the nature of the human hearing phenomenon, particularly the ability of the ear to synthesize the message it is concentrating upon even in the presence of noise, it may be useful in certain circumstances to introduce a pseudo or real message component in the zero level interval indicated in FIG. 5(c). For this purpose suitably selected noise or signal components of approximately the same amplitude and frequency can be inserted in what is otherwise a quiet gap interval in the message stream and this arrangement of the invention is indicated in FIG. 5(d). Where the gap is to be filled with noise components, a suitable source and symmetrical switching to introduce noise from the source into the signal channel can be readily applied during the gating interval.

FIG. 6 represents a preferred form of gap filling where two signal controlled delay lines are used. The speech signal is applied to both delay lines designated channel A and channel B in FIGS. 6(a) and 6(b) respectively and these two lines are signal controlled to have symmetrical complementary gain characteristics and overlapping variable delay characteristics as shown in FIGS. 6(c) and 6(d). Here the delay control signals as shown in FIG. 6(d) are phased to overlap at least an amount corresponding to the transition portion of the gain control characteristics of FIG. 6(c). The outputs of both delay channels A and B are combined to produce the combined output shown in FIG. 6(e).

Generally the length of the delay lines used for channels A and B in FIG. 6 will employ one full length delay line and one relatively shorter length delay line for storing the signal used for gap filling purposes. This arrangement will reduce the cost of the equipment represented by the multiple section delay lines necessary to obtain the required maximum delay length for system performance requirements. On the other hand, for systems where cost is not a primary factor, two equal full length variable delay lines can be employed and their control signals can be alternately applied so that the signal channel is through first one and then the other delay line thereby giving a full signal period for switching the inactive delay line back to minimum delay condition prior to its use for signal transmission again. For such symmetrical delay lines it may still be useful to provide some overlap during the transition as indicated in FIG. 6(d) with appropriate gain control signals applied as indicated in FIG. 6(c).

Referring now to FIG. 7, a basic speech compressionexpansion system in accordance with the invention will be described. This system comprises a variable speed playback device 51 which is indicated to be a tape transport with a manual select speed control input 52. The signal derived from transporting the tape past a magnetic transducer is applied to an AGC amplifier 53 which also passes the signal through a band pass filter having an adjustable low and high frequency cutoff. The selection of the cutoff frequencies for the filter may be operated from manual control 52 in conjunction with the selection of playback speed for the playback device 51. The manual control 52 also supplies an amplitude control signal to a fine voice pitch adjust control 54 which supplies on line 55 a signal to control the end amplitude of the linearly increasing waveform which controls the variable delay line as hereinafter described.

The signal after passing through the amplifier and filter 53 enters a variable delay line 56 which can be signal controlled between minimum and maximum delay limits. This control signal applied on line 57 is derived 5 from a ramp level and amplitude charger 58 which receives as its input either a compression triangular waveform on line 59 or an expansion inverse of the waveform on line 59 which appears on line 61 after passing through an inverter 62. One or the other of the lines 59 10 and 61 is energized with a ramp waveform depending upon the setting of a switch 63 which supplies the basic ramp waveform from ramp pulse train generator 64. The repetition period of the ramp waveform is selectable by a manual control 65. A pulse coincident with 15 the reset of the linear portion of the ramp waveform appears on line 66 and is applied to a blanking pulse generator 67 to produce a blanking pulse output the width of which can be controlled by manual adjustment 68 and which is synchronized with the input pulse on line 20 66.

The output of the variable delay line 56 is applied to a blanking circuit and amplifier 71 which transmits or blocks the signal depending upon the blanking pulse B applied on line 72 from generator 67 and when the ²⁵ blanking pulse is not present B the delay signal is applied to a speech bandpass filter 73 the output of which is applied to an audio reproducer 74.

In addition to the amplitude excursion established for the linear voltage ramp signal from generator 64 which ³⁰ is controlled by manual control 52 the absolute level of the voltage applied can be controlled by level adjust means 60. The variable delay line 56 will generally be of any known type and in particular may be 360 RC filter stages where the shunt resistor is provided by a FET 35 or other semiconductor device which varies resistance in response to a controlled voltage or current. Such delay lines generally perform best with respect to distortion of the signal passing therethrough if the phase delay per stage is kept well below the maximum possible value of 90°. Accordingly, the line can be designed to operate with 45° to 60° phase delay per stage maximum and the number of stages is then determined as greater than the quantity: N&gt;(6 or 8) c(fmai) ΔΤ,„Μ.. In the above inequality the digits 6 and 8 represent the ⁵ number of stages per electrical cycle of the highest frequency to be passed corresponding to a phase delay of 60° or 45°, respectively, as the maximum phase shift per stage which is to be utilized; the quantity c is the 50 compression ratio; the quantity f_max is the highest frequency being passed by the line; and AT_moa. is the maxi., mum signal delay desired as dictated by the maximum permissible discard interval previously specified. Many other forms of delay line constructions which are capa₅₅ ble of being signal controlled are known in the art and the present invention is not to be considered as limited to any particular form of delay line.

Referring now to FIGS. 8(a) and 8(b) the operation of the system of FIG. 7 will be described. The sample ₆₀ period waveform 81 has an adjustable period set by control 65 for producing an asymmetrical sawtooth waveform 82 which produces a relatively long negative going linear voltage followed by a shorter positive going linear voltage. This waveform is used directly on ₆₅ line 59 for speech compression while, after inversion in inverter 62, its inverse is used on line 61 for expansion. The expansion waveform is indicated in dotted lines at 83 in FIG. 8(a). For a variable delay line 56 which in3,786,195 creases delay as the control voltage becomes more negative, the waveforms 82 and 83 have the proper sense for controlling the delay interval and the magnitude of the delay is determined by amplitude control 52 relative to a voltage level set by the level adjust 60. Thus the operating point in the excursion of the waveform 82 is selected for a given compression ratio in conjunction with the sample period which will be a predetermined combination for any given compression ratio assuming the maximum delay ΔΤ„_αχ, in the line 56 is a fixed value as obtained by selecting line length according to the value d T_0U1 as given in FIG. 3(a) and Table I for the desired compression ratio. If the maximum delay to the signal is not maintained constant the discard period will change correspondingly as is evident from the description of FIG. 1 and corresponding adjustments in the amplitude of the wave will be required to give the slope d required for a compression ratio c. Similar considerations apply for the slope of curve 83 which must be set at its corresponding value d for an expansion ratio e.

The operation of blanking pulse generator 67 is shown to produce a pulse 84 in FIG. 8(b) of predetermined width in response to the start pulse of the sample period signal 81 received on line 66. This pulse may be applied in gain control fashion to the circuit 71 with modified trailing edge as previously described to reduce the transient signal and provide a gradual onset of voice sound signals which are passed to the transducer 74. The width B of the blanking pulse is selected with control 68 and is normally made of sufficient duration to permit the short steep linear portion of the ramp waveform to return the delay line 56 to its zero or minimum delay condition and dissipate the signal energy therein (or the transient caused by switching the line itself) prior to enabling the signal channel which energizes the transducer 74 with the subsequent speech signal segments. _

The blanking period B and enabled period B for expansion mode are shown in FIG. 8(c). The expanded chunks with an initial output gap are shown in FIG. 8(d).

The system of FIG. 7 can also be used to substitute noise or pseudo signal gap filling signals corresponding to the system described in FIG. 5. For this purpose a source 75 of such signals is arranged to supply the input signal to filter 73 during the blanking interval. By means of a switch 76 this gap filling during the blanking interval can be made optional. The gap filling signal 75 can also be derived from the message signal output of amplifier 53.

Referring now to FIG. 9, a modified form of the invention particularly suitable for accomplishing the various gap filling procedures for the speech compression systems previously described will be disclosed. Portions of FIG. 9 which are essentially the same as those described in FIG. 7 have corresponding reference numerals and accordingly only the additions and changes will be further described. In addition to the variable delay line 56 a second variable delay line 91 receives the signal wave from amplifier 53. The output of the delay lines 56 and 91 are applied respectively to complementary blanking circuits 92 and 93. Signals passed by these blanking circuits 92 and 93 are amplified and filtered in element 73 and passed to the acoustic reproducer 74 as heretofore described.

A pulse train generator 94 produces a pulse wave train as shown in FIG. 10(a) having a selectable pulse repetition rate determined by the setting of manual control 65 thereby establishing the basic sample period. The output pulse from generator 94 is delayed in delay unit 95 and applied to a first ramp generator 96 and in undelayed form is applied to a second ramp generator 97. The ramp generators 96 and 97 are subject to waveform level control from manual adjust element 60 and ramp linear wave amplitude control from the manual adjust element 52. As previously stated, the fine pitch adjustment 54 may be provided for slightly modifying the ramp slope as a voice pitch adjustment by effectively altering the frequency conversion over a small range. In addition the blanking width interval of each generator is adjustable with controls 68 and 70 respectively. The outputs of. the ramp generators 96 and 97 are applied respectively to delay lines 56 arid 91 to control the time delay of signals passing through the respective lines in accordance with the control signals applied. By means of c or e select controls the sense of the slope of the ramp waveforms can be selected for compression or expansion.

The level and amplitude controls for setting the respective ramp generators 96 and 97 are preferably relatively adjustable to permit selection of the relationship between the two ramp waveforms. By making the delay and phasing of the unit 95 adjustable any desired delay line overlap can also be achieved. It is also possible to rearrange the components to have the complementary gating at the inputs of the two delay lines 56 and 91 with the outputs switched to be combined in a common channel to amplifier 73. This alternative discards the portion of the speech signal that is not utilized by each line before it enters the line and thus eliminates the necessity for dissipating these portions when the lines are switched between active periods.

Referring now to FIG. 10, the operation of the speech compression system of FIG. 9 will be described. The pulse train generator 94 produces the timing waveform of FIG. 10(a). This pulse triggers the transition of waveform C2 in pulse ramp generator 97 which produces the blanking pulse indicatedjn FIG. 10(c) with the predetermined width of B and B being determined by the blanking pulse width control 68. After the delay indicated in FIG. 10(b) the pulse from generator 94 triggers the ramp generator 96 to produce the waveform Cl shown in FIG. 10(b), With this arrangement the control wave C1 for the delay line 56 is overlapped in time by waveform C2 having slope in the same sense and bridging the steep return slope waveform of ramp wave Cl. With the asymmetrical time intervals shown in FIG. 10, the arrangement for gap filling modes of operations shown in FIGS. 5 and 6 can be practiced. By making the waveforms Cl and C2 have symmetrical rising and falling portions the arrangement is suitable for alternate switching of the lines 56 and 91 to provide alternate compressed (or expanded) chunks of the speech sample. The choice of the relative lengths of sample through line 56 and 91 will generally be dictated by manufacturing costs for the delay line. Thus for a main delay line 56 of adequate length for the compression ratio desired, a relatively shorter line 91 used only for gap filling purposes will generally be more economical. On the other hand, two full length lines which are alternately active to pass speech sample chunks thereby providing adequate time for the non-active line to be returned to its minimum delay condition will provide for smooth transitions, any desired overlap and the

3,786,195 maximum time interval for discharging the line to minimum delay condition prior to its processing the next speech sample. The action of the system of FIG. 9 in the gap filling mode is indicated in FIG. 10(d) and generally corresponds to that previously described with respect to FIG. 5.

The operation of the system of FIG. 9 for speech expansion, i. e., increasing the time duration for a given speech utterance and increasing the frequency components thereof from a reproducer running at a slower than recorded rate is shown in FIG. 11. Here the ramp generators 96 and 97 have inverted outputs to produce the expansion waveforms El and E2 shown in FIGS. 11(a) and 11(c) respectively and the blanking waveform has been made symmetrical such that the delay lines 56 and 91 are used alternately for approximately equal periods. By the nature of speech expansion, a gap in the signal Output will always occur since the lines are controlled to change from maximum delay at the start of the sample to minimum to zero delay at the end of the sample. Thus when the line is switched to maximum delay there will inevitably be a time gap before delayed signal emerges from the output end of the line. Applying the control sequence indicated at FIG. 11 the speech samples processed by lines 56 and 91 are overlapped so as to fill the gap as indicated in FIG. 11(d) by the solid and dotted outlined signal chunks Ef and E₂' . The presence of a slight overlap in the reproduced signal does not significantly interfere with intelligibility since it generally is not noticeable and at worst may result in a slight echo effect of the type commonly encountered in a telephone conversation. The time expanded speech waveform obtained using the mode of operation indicated in FIG. 11 is useful for the recognition and comprehension of difficult passages and for analysis and study of foreign languages and the like.

The system shown in FIG. 12 represents a simplification of the system of FIG. 9 where a fixed delay line 101 is used in place of the second variable delay line 91 of FIG. 9. The control of blanking ciruits 92', 93' is simplified in that the variable width blanking gate B as derived from pulse train generator 94 correspondingly produces gaps in the output signals which have been delayed by passage through variable delay line 56. The fixed delay of line 101 is selected to further delay some portion of the signal emerging from the delay line 56 by an amount sufficient to fill the gap caused by blanking pulse B thereby essentially repeating some portion of each message chunk while the variable delay line 56 is switched back to its minimum delay condition. Again, this repetition is not objectionable and may merely introduce a slight echo effect which is much less objectionable than the presence of the gap in the message signal. This sequence of operation is shown in FIG. 13 where the variable chunk C„ and the fixed chunk C_Falternate in supplying the output.

The expansion mode for operation of the circuit of FIG. 12 is shown in FIG. 14 where the ramp signals are inverted for the expansion waveform which controls the delay line 56 to vary from maximum delay to minimum delay over the linear ramp portion E shown in FIG. 14(a). The blanking waveform B is selected to pass some portion of the signal chunk through the appropriate amount of delay to fill the gap between chunks in the output as indicated in FIG. 14(c). Thus the output is composed of chunks E_F and E„ in alternation for continuous signal.

The system of FIG. 12 could be further simplified by eliminating delay line 101 and conditioning gate 93' to introduce in the gap interval any pseudo or noise signal from a suitable source which would simulate the frequency content of the actual speech signal. While this version would be less desirable than using the actual speech signal for gap filling it would, nevertheless, be better than reproducing the speech signal with the message gaps present since the audible effect of gaps becomes detrimental to recognition of the message content, especially at high compression ratios. This modification gives a mode of operation similar to the optional noise gap filling described for FIG. 7.

FIG. 15 shows a modification of the invention for binaural processing. The speech signal from the band pass filter 53 is applied to symmetrical variable delay lines VDL1 and VDL2 controlled by waveform generator 102. The output of VDL1 is applied as an input to gates 103 and 105. The output of VDL₂ is applied as an input to gates 104 and 106. The delay line VDL, is controlled for linear variation of delay according to the waveform of FIG. 16(c). The delay line VDL₂ is controlled for linear variation of delay according to the waveform of FIG. 16(d). Each of these waveforms has its rapid return transition at the mid-point of the linear delay portion Of the other waveform.

The gates 103 and 106 are controlled by gating waveforms B_t and Βχ shown in FIG. 16(e). Gate 103 passes signal during Bi_and is blocked during B,. Gate 106 is blocked during Βχ and passes signal during Βχ . Amplifier 107 combines the outputs of gates 103 and 106 and applies the combined signal to an audio reproducer 108.

The gates 104 and 105 are controlled by the gating waveforms B₂ and B₂ shown in FIG. 16(f). Gate 104 passes signal during B₂_and is blocked during B₂. Gate 105 is blocked during B₂ and passes signal during B₂. Amplifier 109 combines the outputs of gates 104 and 105 and applies the combined signal to an audio reproducer 110.

The system of FIG. 15 operates to reproduce the entire original signal (for compression ratio equal to two) since each delay line processes the portion which is the discard for the other line as is evident from FIGS. 16(a) and 16(b). For compression ratios greater than two some message discard occurs and for ratios less than two the overlap or message duplication increases in the output. By listening binaurally, however, the intelligibility is enhanced since the overall discard is eliminated (or greatly reduced for the higher compression ratios) and the overlap or repeat of message portions is not detrimental to word detection by the listener.

A binaural system without supplemental gap filling (as just described) would be achieved by removing gates 105 and 106 in FIG. 15. The lines VDL, and VDL₂ would supply the processed signal in alternation to the respective output transducers 108 and 110 for binaural output.

FIG. 17 shows the invention using a form of delay line capable of processing speech signals in a manner which greatly reduces the problems associated with discarding stored information in the line. The system shown in FIG. 17 comprises an analog shift register having a plurality of stages ASR,, ASR₂, ASR„, which has a speech signal input applied on line 111 and a compressed or expanded speech signal output on line 112. Alternate stages of the delay line are clocked by

3,786,195 two-phase clocking signals applied at lines 113 and 114 as derived from a shift frequency generator 115. The frequency variation of the generator 115 is such that the inverse of the clocking frequency namely, the pulse to pulse period varies as a linear function of time with the frequency varying from high frequency to low frequency for compression and from low frequency to high frequency for expansion mode operation.

The analog shift register shown in FIG. 17 is of the general type disclosed in the article by F. L. J. Sangster published in the 1970 IEEE International Solid States Circuits Conference Proceedings, pages 74-75 and 185. Such shift registers sample an analog signal and pass the sampled value along the line at the clock rate by a charge or charge deficit storage mode which permits the signal sample to be recovered at the output of the delay line after a time delay which is proportional to the inverse of the clocking rate. In the present invention, with the clocking rate varied so that the inverse thereof is linear function of time, the delay line operates to expand or compress the speech signal and by tailoring the length of the line and the repetition rate of the linear control function in accordance with the principles herein set forth, continuous processing of random speech signals is achieved. In the compression mode at the end of each linear segment of the control function generated by generator 115 all stages of the delay line may be reset if a reset input is available or the line may be simply emptied during the blanking period as the line is reloaded with the beginning of the next segment of the speech signal at the high frequency clock rate. This rate may be made rapid enough to make the required blanking so short as to be aurally imperceptible. Thus the problems associated with gap filling or smoothing or blanking may be minimized in this version of the invention.

The design parameters for the analog shift register (ASR) can be established by applying the criteria previously set forth, as will now be described.

The instantaneous delay r(t) of an analog shift register at time t is r(t) = dt + r₀ where d is the rate of change of delay and t₀ is the initial delay. The delay dt becomes in the ASR for N stages N (p— lip). l/f_t where f₍ is the frequency of the clock shift signal at time t and p is the number of phases. Defining N (p— lip) as Ν', the delay becomes r(t)= N' [(1//) + (1//,)] where/,, is the initial clock shift frequency.

The slopes of the time delay function for restoring the original speech frequencies are the same quantities as before. For compression:

c't=AT_t = total delay to signal entering at time t = (c— 1 )t for compression ratio c, and d = 2 c-l/c+1 = N'/t [(1//) -(1//)]

For expansion:

c't = AT_t — (1—e/e) )t for expansion ratio e, and d=2 (1—e/l+e) = (N'/t) [(1//) - (1//,)]

Hence the inverse of the shift frequency as a linear function of time multiplied by Ν', produces the delay needed to obtain compression or expansion of the speech wave with original speech frequencies restored.

Referring to FIG. 18, the reset time t_N to clear N stages is t_N = N'/2[( l/f_N) + (1 //,)] which represents the time needed for the first N pulses to refill the line. By suppressing the sample-switching transients by filtering or blanking or any of the other methods herein de20 scribed, and holding t_N to less than 0.2ms the effect of gap intermodulation previously discussed becomes virtually imperceptible.

There is a limit to be observed for the sampling clock frequency to accommodate the highest signal frequency f_max to be passed by the line. Referring to FIG. 25 the signal frequency for compression decreases linearly as it passes through the delay line as indicated by line 201. The clock frequency varies as a hyperbolic function indicated at 202 and must throughout the sample period be equal or greater than the values or line 201 for which the relation is given by / ss (2p/p-l)/_mox (which for a two phase ASR in which p = 2 is 4f_max) to provide for at least two samples per cycle of f_maxthroughout.

FIG. 18 shows a modification of the analog shift register embodiment shown in FIG. 17, but with provision for the exact substitution of processed signal at the initiation of reset for the delay line. The analog shift register delay line 121 processes input signals from line 122 in accordance with the variable pulse derived frequency derieved from square wave generator 123 as previously described in connection with FIG. 17. The pulse frequency is such that the pulse spacing varies linearly as indicated from pulse signal source 124 where the reciprocal of frequency is shown to be linear with respect to time. At a point 125 on the analog shift register the line branches to have duplicate shift register stage paths 126 and 127. The number of stages required in each of the boxes represented by 126 and 127 is sufficient to continue processing the signal while the line 121 is being reset. The branch outputs from stages 126 and 127 are subject to complementary gating controls 128 and 129 and when gated to pass signal both supply input signals to a combining amplifier 130.

In addition to the control of the main line 121-126 from generator 123 and 124 the branch line 127 is controlled through gate B 131 from pulse generator 124 which triggers a second square wave generator 132 and when gated during the B interval by gating unit 133 the trigger rate for generator 132 is derived from a fixed pulse generator 134. The pulse generator 134 may also operate at the pulse rate 1//.

The operation of the system of FIG. 18 can be described with reference to the waveform associated with output line 135. For a given sample period the variation in frequency of generator 124 starts and controls the ASR line 121 as previously described. For this condition the gates B are permitting signals to pass and the output of stage 126 is transferred to the input of amplifier 130 thus producing the frequency converted output signal indicated during the sample period of the waveform. At the same time the B control for gate 131 permits the same control pulse signal from generator 124 to trigger generator 132 thereby keeping the branch ASR stages 127 operating in synchronism with the corresponding stages 126. The blanking control B in gate 129, however, prevents the output from stage 127 from reaching the input of amplifier 130. At the blanking or reset period for the main ASR line 121 and generator 124 the B and B gates switch thereby interrupting signal flow from stage 126 to amplifier 130 and permitting signal flow from stage 127 to the input of amplifier 130. Since stages 126 and 127 were in synchronism this switching will be of identical signals and thus imperceptible at the output line 135 of the ampli

3.786.195 fier 130. At the same time the B and B switching in gates 133 and 131 interrupts trigger pulses from generator 124 and passes trigger pulses from generator 134 to the square wave generator 132. This switching of control assures that the generator 132 will continue to process signal in stages 127 while the generator 124 can be reset for the start of the next sample. At the end of the blanking period there will be some discontinuity as the gates B and B switch back to their original state thereby returning control to pulse generator 124 inasmuch as the start of the next sample period will not produce signals which exactly coincide with the signals terminated at the end of the blanking pulse which were under control of the pulse generator 134.

FIG. 19 shows a modification of the present invention in which the variable delay line under the control of the variable frequency generator 136 operates with a 1//repetition rate control function exactly analagous to that described with reference to FIG. 17. In FIG. 19 instead of carrying the analog signal through the successive stages of the shift register, the input signal on line 137 is first converted into a digital word in an A/D converter 138, the parallel output of which applies a parallel word to the input registers of the first stage 139 with this digital value transferred sequentially through the serial stages until it reaches an output D/A converter 140 where it is converted into an analog signal on output line 141. This operation is entirely analagous to the system described with reference to FIG. 17 except for the coding of the information as it passes through the serial stages which are driven with the variable clock frequency to provide the frequency conversion required. One advantage of the system of FIG. 19 is in the provision of a reset signal from generator 136 on line 142 which can be applied to all the registers of all the stages simultaneously thereby effecting an instantaneous clearing and resetting of the line at the end of the sample period.

FIG. 20 shows a modification of the invention which is analogous to that shown in FIG. 19 except that the digital signal is processed serially through a serial shift register 150 after the digital output of A/D 138 has been serialized in a serializer 151. The shift register 150 is under the control of the shift frequency generator 136 including a reset input line 142. The output of the serial digital shift register 150 is applied to a parallelizer 152 which converts the serial bit string into a parallel digital word for conversion by D/A 140 into the required analog output signal on line 141.

FIG. 21 shows an embodiment of the invention in which an analog storage matrix with provision for addressible read-in and read-out signal storage is provided. A charge storage matrix 161 is shown having a plurality of X write lines 162 and a second plurality of Y write lines 163 the intersections of which define the matrix address at which are located analog storage elements. Typically, an analog storage matrix will have a capacitor charge storage device at each intersection of the X and Y lines defining the matrix for storing an analog value represented by the charge on the capacitor. Each such memory location is also accessed by a plurality of X read lines 164 and corresponding plurality of Y read lines 165 where the intersection of the lines 164 and 165 correspond with the location of the charge storage elements located at the intersections of the write lines 162 and 163,

To store an analog signal in a charge storage memory 161 an analog signal input on line 166 is applied and the instantaneous value thereof is stored in the charge storage element associated with the coincidentally en5 ergized X and Y write line intersection from an X counter-write enable 167 and a Y counter-write enable 168. Typically the X and Y counters 167 and 168 will be operating at a predetermined pulse rate derived from a pulse generator 169 with the number of pulses in the X 10 line sequencing the X write lines 162 after which the Y counter 168 is stepped arid the next row of X intersections with the then active Y line being energized by the next sequence of pulses from the pulse generator 169. Accordingly, the memory 161 has a storage capacity of 15 X x Y number of storage elements corresponding to the number of intersections of the X and Y lines. With pulse generator 169 operating at a constant frequency this write-in of the analog signal on line 166 occurs at a predetermined rate and storage capacity is selected 20 to store a signal sampled in accordance with the general requirements hereinbefore set forth.

A frequency converted output signal is derived from output line 171 which receives serially from the charge storage elements of the memory 161 the analog values 25 stored therein as the matrix intersections are sequenced in regular order by the operation of X counter-read enable control 172 and Y counter-read enable control 173. The pulse rate for counters 172 and 173 is determined in accordance with a ramp voltage gener³⁰ ator 174 controlling a voltage control oscillator 175 with the variable rate selected to produce the desired signal compression or expansion in accordance with the principals of this invention. For this purpose a rate control potentiometer 176 is provided for selection of ³⁵ the ramp voltage slope in generator 174 and this slope will generally be a gang control with the tape playback speed control 52 described with reference to the embodiment shown in FIG. 7. This dual control function is indicated by line 177. A further control from the rate ⁴⁰ control potentiometer 176 is applied on line 178 to the pulse generator 169 to control its pulse repetition rate in relation to the maximum read-out rate established by control of the ramp voltage generator 174 and oscillator 175. In particular the writing pulse rate must be maintained higher than the maximum read-out pulse rate in order to avoid having the read-out sequencing overtake the write-in entry. As soon as any storage element has been read it is available for storing the value _5o on the next signal sequence and can be either reset upon read-out or upon entry of the next write signal. The ramp generator 174 provides a reset on line 179 for resetting the counters at the end of each ramp voltage period for the start of the next signal sample stor₅₅ age sequence.

FIG. 22 shows a system employing a random access memory 181 with write and read controls 182 and 183 which operates in a manner analogous to that described for the system of FIG. 21. Since the memory 181 stores ₆₀ a binary information however the input signal on line 184 is required to be converted in A/D converter 185 and the corresponding output must be converted in D/A converter 186. The sequencing of write-in and read-out for the memory matrix generally corresponds ₆₅ to that previously described for FIG. 21.

Where gap filling is employed a particular means for minimizing the disturbance caused by the discontinuity at the beginnings and/or the ends of the signal samples

3,786,195 ^λ is illustrated in FIG. 23 with gating signal control indicated in FIG. 24. The logic control and sequencing elements therein are so configured as to cause the primary signal sample 191 to end at a zero crossing and the supplementary gap filling signal 192 to start at its next zero 5 crossing in the same direction and then at the end of the blanking period for the primary signal for said supplementary signal to end in a zero crossing to be followed by the new primary signal sample 191 at its next zero crossing in the same direction. Thus, the source 10 signals 193 and 194 after filtering through low pass filters 195 and 196 to remove processing and spurious high frequency components are fed into their respective gates 197 and 198 and voltage comparators 199 and 200, the latter units so connected to ground by di- 15 rectional circuitry 201 and 202 as to trigger pulse generator (PG) 203 or 204 whenever a positive going zero crossing by the respective signal 193 or 194 occurs. Gates 197 and 198 are actuated so as to pass signals 193 and 194 by the reset output lines 205 and 206 from 20 flip flops 207 and 208. Flip flop 207 is set by a pulse on line 209 from gate 211 as conditioned by the reset output line 213 from flip flop 208 and by the inverted output 216 of the sample period pulse train generator 219. Flip flop 207 is reset by the pulse output of gate 217 as ²⁵conditioned by the direct output 215 of the PTG 219. Similarly Flip Flop 208 is set by pulse 210 from gate 212 as conditioned by the reset output of line 214 from flip flop 208 and by the direct output 215 of PTG 219.

Flip flop 208 is reset by the pulse output of gate 218 as ³⁰conditioned by the inverted output on line 216 of PTG 219. Gate pairs 211 and 217, or 212 and 218 are pulsed by the output of pulse generator 203 or 204, respectively, whenever a positive going signal zero crossing occurs as described above. Thus with flip flop 207 in ³⁵the set condition allowing the primary signal 191 to pass and with flip flop 208 in the reset condition blocking the filler signal 192, when PTG output line 215 goes positive as at 220, gate 217 allows the next pulse from PG 203 to reset flip flop 207, thus blocking primary sig- ⁴⁰nal to line 191. At the same time, gate 212 is conditioned to pass the next pulse from PG2 204 to set flip flop 208 and allow the supplementary signal to pass until the end of the gap period. At this time the inverted PTG output 216 goes positive as at 221 allowing gate ⁴⁵218 to pass the next pulse from PG2 204 and reset flip flop 208, cutting off the supplementary signal 192 and conditioning gate 211 to pass the next pulse from PG 203. This sets flip flop 207 allowing primary signal 191 to pass to amplifier 222 and be outputted at 223. The ⁵⁰process then repeats in the sequence described above.

While the invention has been described with reference to frequency-time transformations of the original ₅₅signal, the disclosed embodiments are also useful for ⁰frequency transformation, where required, due to other factors such as a change in the velocity of propagation of the sound waves. For example, a human breathing in an artificial atmosphere such as one having a high he- ₆θ lium content speaks with a higher than normal voice pitch but with other parameters substantially unchanged. By using the speech compression modes provided by the present invention, such speech can be restored to its normal frequency range without change in ₆₅time scale.

Obviously, the methods and apparatus herein disclosed can be used for coded audible signals other than speech such as music, for example, by taking due consideration of the corresponding parameters significant to comprehension as herein disclosed. Many other modifications can also be made without departing from the scope of the invention as defined in the appended claims.