US3286031A

US3286031A - Voice actuated device

Info

Publication number: US3286031A
Application number: US262407A
Authority: US
Inventors: David A Geddes
Original assignee: ALTO SCIENTIFIC CO Inc
Current assignee: ALTO SCIENTIFIC CO Inc
Priority date: 1963-03-04
Filing date: 1963-03-04
Publication date: 1966-11-15
Anticipated expiration: 1983-11-15

Description

Nov. 15,

Filed March 4, 1965 2 Sheets-Sheet l Fi 51- l /2 [I3 /4 /5 AUDIO ASYMMETRY LOW PASS 0.0.

AMPLIFIER DETECTOR FILTER AMPLIFIER I/ 19 2o /9 /8 [/7 [I6 UTILIZATION SWITCHING LEvEL HIGH PASS LOW PASS DEVICE cIRcuIT DETECTOR FILTER FILTER F 8 UTILIZATION DEVICE INVENTOR.

ADJUSTABLE VOLTAGE L.- SUPPLY DAVID A GEDDES ATTORNEY Nov. 15, 1966 D. A. GEDDES 3,286,031

VOICE ACTUATED DEVI GE Filed March 4. 1963 F 2 Sheets-Sheet 2 Fi 4A WE F 4 c WMM/ Fi 4D Fi 45 J W Fi 4F i /W Fig 4G I x VV ,Wnv \vh n F19 INVENTOR.

DAVID A. GEDDES Y VflfyAM/L W B ATTORNEY United States Patent 3,286,031 VOICE ACTUATED DEVICE David A. Geddes, San Jose, Calif., assignor to Alto Scientific Co., Inc., Palo Alto, Calif., a corporation of California Filed Mar. 4, 1963, Ser. No. 262,407 8 Claims. (Cl. 179-1) This invention relates to voice actuated devices, and more particularly, to a device and a method for converting selected acoustical sounds, and particularly live speech, into electrical output signals for actuating a switching means which device and method is highly effective in discriminating against non-selected acoustical sounds.

There are many applications where it would be desirable to control the operation of a switch means or the like directly by speech rather than in the conventional manner, such as by hand or by timing circuits. This is particularly true in situations where the operator of a machine is disabled from exercising manual control or where he finds manual control inconvenient, and either wishes or is compelled to rely upon his voice to either shut off or turn on a device.

For example, in the case of a machine tool such as a lathe, an operator, whose clothing is caught by the rotating spindle may be disabled from reaching the power switch to turn off the lathe to extricate himself. Enabling him to turn off the machine by means of his own voice command could do much towards protecting him from serious harm. In the case of machine tools such as a punch press, the operator may Wish to utilize both his hands to manipulate the work piece. Even though foot controls have been extensively used heretofore, an audible command to the press to punch would be very desirable since it would permit him to utilize his feet for their natural function of support.

Also in the case of home appliances or signalling devices, the user of such appliances or devices may find it more convenient to operate the same by an audible command, particularly where he is disabled or inconvenienced. This is particularly true in case of patients in a hospital who are disabled from exercising any manual control whatsoever including the depressing of a switch and who may, by means of voice, initiate a call signal for a nurse or the like. a

Most prior art devices, which have attempted to solve the problem of responding to speech, have been found unsatisfactory because of their disability to effectively discriminate against other types .of noise. These devices utilize the energy level of the acoustical input power. Obviously, discrimination based upon input power level is unsatisfactory since there is always the possibility of loud background noise such as the dropping of a work piece or the ringing of a telephone which would actuate the prior art devices unintentionally.

It has been found that human speech (and certain other sounds) have certain unique characteristics which allow discrimination between it and background noise. For example, human speech may be divided, for the purpose of this discussion, into voiced sounds and frictional sounds. Voiced sounds are those that originate in the vocal cords which vibrate when air passes through them. Frictional sounds, on the other hand, arise when the tongue, teeth, or lips form a constriction through which air passes.

It has further been found that voiced sounds have a unique property not possessed by frictional sounds and most other types of noise, namely the ability to assume an asymmetric amplitude characteristic when properly phased. IBy asymmetric amplitude characteristic is meant that the positive and negative peaks of the electrical voice signal envelope are different and, when summed, provide an output signal which will be referred to as the asymmetry signal.

.The reason for the asymmetry characteristic is due to the fact that voiced sound includes prominent even harmonics which, when out-of-phase with their fundamentals (or other even harmonics) give rise to the asymmetry envelope. Frictional sounds and background sounds, on the other hand, have been found to lack the presence of prominent even harmonics so that their envelope does not exhibit asymmetry to any substantial degree. It is therefore the absence of prominent even harmonics from non-voiced sounds which make it possible to discriminate speech from most other sounds and noises.

As has already been stated, the degree of asymmetry of the voiced sound wave envelope depends on the relative phasing and relative amplitude of the even harmonics with respect to their fundamental. One prior art device utilizes phasing for discrimination between voiced and nonvoiced sounds by passing the electrical signal through a phase shifter which is frequency sensitive and which is set to optimize the phase shift for a selected fundamental and its most prominent even harmonics to maximize asymmetry. In other words, this prior art device relies for discrimination on maximizing asymmetry for a selected voiced signal applied to its input. It has been found, however, that such phase shifters are rather expensive particularly when several stages must be used to obtain the desired phase shift and further that there are certain other non-voiced sounds which exhibit sufiicient asymmetry to trigger the device and which are not discriminated against.

It has been found that live speech inherently has envelope asymmetry as it is generated by the vocal cords. Even though maximizing this asymmetry by utilizing phase shifters, set to provide optimum conditions with respect to a selected audible input signal for best discrimination between different sound sources, results in a satisfactory device, such voice actuated devices may be constructed more reliably and more economically by utilizing the inherent asymmetry in the voice signal, at least for certain applications. Furthermore, discrimination against noise and certain other sounds may be obtained more inexpensively by use of specially designed filters without incurring the cost of phase shifters. Finally, discrimination by means of filters has certain additional advantages in that a voice actuated device may be made to discriminate against certain types of voices and noises not discriminated against in other systems.

It is therefore a primary object of this invention to provide a voice actuated device which is reliable and economical to manufacture.

It is another object of this invention to provide a voice actuated device which changes its position upon receiving a live voice command.

It is another object of this invention to provide a voice actuated device which discriminates against all but voiced sounds and which provides an output signal in response to voiced sound which may be utilized to control a switch means or the like.

It is another object of this invention to provide a voice actuated device which responds to a live voice command and which discriminates against other acoustical sounds.

It is also an object of this invention to provide a voice actuated device which is responsive to the rate of change of vowel sounds in speech, recognizes human speech, and discriminates against all other types of sound.

Other objects and a better understanding of the invention may be had by reference to the following description, taken in conjunction with the accompanying drawings, in which:

FIG. 1 is a schematic block diagram of the voice actuated device of this invention connected to a utilization means;

FIG. 2 is a schematic circuit diagram of an embodiment of the asymmetry detector and the first low pass filter portions of the device of FIG. 1;

FIG. 3 is a schematic circuit diagram of an embodiment of the high pass filter, level detector and switching circuit portions of the device of FIG. 1; and

FIGS. 4A to 41 are illustrative wave form diagrams derived from different acoustical input signals useful in explaining the operation of this invention.

Referring now to the drawing, and particularly to FIG. 1 thereof, there is shown a preferred embodiment of a voice actuated device constructed in accordance with this invention. A plurality of microphones 11, conveniently and strategically positioned at selected locations, are suitably connected to the input terminal of an audio amplifier 12. Microphones 11 are preferably selected to provide virtually no relative phase shift of the various component frequency of the applied voice signal so that the transduced signal has the same phasing as the input signal.

The electrical signal developed by microphones 11 is applied to an audio amplifier 12 which provides suitable amplification. Just as microphone 11, amplifier 12 is constructed to provide an amplified signal which has substantially the same phasing as the applied voice signal. After suitable amplification the amplified signal is applied to an asymmetry detector 13 which will be explained hereinafter more fully in connection with the description of FIG. 2. Briefly, asymmetry detector 13 is an amplitude peak detector which develops an asymmetry signal representative of the difference between the positive and negative amplitude peaks of the amplified signal.

The asymmetry signal developed by asymmetry detec tor 13 is passed through a low-pass filter 14 which has a cut-off frequency usually below 10 cycles per second and a roll off or slope factor which depends on the slope factors of similar filters through which the signal is subsequently passed as will be explained. The filtered asymmetry signal is then applied to a DC. amplifier 15 which provides for suitable amplification. In the particular embodiment of this invention shown in FIG. 1, DC. amplifier 15 is a chopper amplifier having a small dynamic range which should be protected from high frequency transients. It is for this reason that low-pass filter 14 is placed between detector 13 and amplifier 15.

The amplified asymmetry signal is then passed to a further low pass filter 16 which is selected to have the same or a similar cut-off frequency as filter 14. Lowpass filter 16 preferably comprises a plurality of stages, such as conventional RC ladder circuits, to provide a roll off or slope factor substantially greater than say, 6 db per octave. In case a DC. amplifier is selected which does not require a filtered input signal, a low-pass filter 14 may be combined with filter 16 and placed either ahead or behind amplifier 15.

After being passed through low-pass filter 16, the asymmetry signal is applied to a high-pass filter 17 which has a cut-off frequency normally less than 1 cycle per second and which essentially removes the DC. component. It should be noted that the combined effect of low-

pass filters

14 and 16 and high pass filter 17 is that of a bandpass filter having a pass band extending from below 1 cycle per second to say 4 cycles per second. It has been found that at the low frequency end of the pass band, a slope factor of about 6 db per octave is sufficient while at the high frequency end of the pass band a much greater slope factor is desirable, say for example, 36 db per octave.

The filtered and amplified asymmetry signal is then applied to a level detector 18 which is set to provide an output signal only if the amplitude of the applied signal exceeds a selected amplitude. As will be explained hereinafter more fully, level detector 18 performs the function of a noise discriminator which permits only voiced sounds of a certain amplitude to initiate an output signal. The level detected asymmetry signal is applied to a switching circuit 19 which is basically an amplifier-relay combination. The function of switching circuit 19 is to provide an output signal in the form of a predetermined relay position in response to an output signal from level detector 18.

The output signal from switching circuit 19 is then utilized to control a utilization device 20 which may be the power switch of a machine tool or the synchro controlling the angular position of the channel selector of a television set or the like.

Referring to FIG. 2, there is shown a schematic circuit diagram of asymmetry detector 13 and low-pass filter 14. The amplified audio signal is applied to input terminal 25 of detector 13 which is connected, through a suitable coupling capacitor 26, to the cathode of a rectifying diode 27 and to the anode of a rectifying diode 28. There is also provided a suitable ground connection via resistive impedance 29 for a DC. return path for the diode current.

The anode of rectifier diode 27 is connected to an RC circuit comprising resistive impedance 30 and capacitive impedance 31 suitably grounded. Similarly the cathode of diode 28 is connected to an RC circuit comprising resistive impedance 33 and capacitive impedance 34 suitably grounded. The RC circuits are connected to detector output terminal 36 through summing impedance 32 and 35 respectively. Finally, detector output terminal 36 is connected to the low-pass filter output terminal 37 and grounded through a parallel RC circuit comprising resistive impedance 38 and capacitive impedance 39.

The combination of diode 27 and its associated RC circuit provides a diode peak detector in which capacitive impedance 31 is charged up to the voltage that is almost, but not quite, equal to the negative peak of the amplified audio signal. The actual voltage to which capacitive impedance 31 is charged is equal to the amplified audio signal less the voltage drop across diode 27. The same is true for capacitive impedance 34 which is charged substantially to the positive peak amplitude of the amplified audio signal.

In order that the detected signal can follow the variation in the envelope of the applied signal, it is necessary that the time constant of the RC circuit be selected sulficiently small so that the voltage across capacitive impedance 31 can decrease as fast as the envelope diminishes in amplitude. Since, as will be explained hereinafter, the envelope amplitude variations depend on the duration of a voiced sound such as a vowel or the rate of change of the use of vowels in ordinary speech, it has been found that an RC time constant may lie in the range from about 3 to 50 milliseconds. This time constant is also large enough to prevent detector 13 from following the voice frequency themselves.

Summing impedances 32 and 35 develop the asymmetry signal which is equal to the difference between the peak detector signals and therefore represent the degree of envelope asymmetry of the applied audio signal. The combination of summing impedance 32, filter impedance 38 and capacitive impedance 39 form a low-pass filter having a cut-off frequency below 6 cycles per second and a slope factor of 6 db per octave. It is to be remembered that filter 14 only forms a portion of the total low-pass filter of the device 10.

The following table shows representative circuit values Capacitor 26 microfarads 20 Resistor 29 ohms 620 Diodes 27 and 28 l N 60 Resistors 30 and 33 kilo ohms 10

Capacitors

31 and 34 microfarads .5 Resistors 32 and 35 kilo ohms" 27 Resistor 38 do 4.7 Capacitor 39 microfarads 2 Low pass filter 16 may comprise a plurality of simple low-pass RC filter sections each of which has the time constants of the order of magnitude between 10 and 20 milliseconds. The number of filter sections utilized depends on the desired slope factor and may vary between 3 to 8. It has been found that, in practicing this invention, a slope factor of approximately 20 to 40 db per octave is desirable for good noise discrimination as will be explained in connection with the description of FIG. 4.

Referring now to FIG. 3 there is shown, by way of example, a schematic circuit diagram of high pass filter 17, level detector 18 and switching circuit 19. The asymmetry signal, suitably amplified and filtered by low-

pass filters

14 and 16, is applied to input terminal 45 of highpass filter 17 which comprises a single filter section made up of capacitive impedance 46 and resistive impedance 47. As already stated, the prime purpose of high-pass filter 17 is to remove DC. from the asymmetry signal. A DC. asymmetry signal may be generated by certain sounds which have an asymmetry characteristic and which are of constant acoustical power. Such powers are not due to speech since primarily only vowels give rise to voiced sounds and consonants to frictional sounds which alternate with the frequency usually above /2 and below 6 cycles per second. By Way of example the time constant of high-pass filter 17 may be selected to be 600 milliseconds which is made up of a capacitive impedance 46 of 60 microfarads and a resistive impedance of 10 kilo ohms.

The amplified asymmetry signal, having removed therefrom all frequencies except those falling within the selected pass band, is applied to the base electrode of a transistor 48 which is connected as an emitter follower and which accomplishes the function of level detection. For this purpose resistive impedance 47 forming the shunt impedance of the high-pass filter 17 is connected to an adjustable voltage source 49 which provides a base electrode bias adjustable between, say, and +6 volts. The bias on the base electrode determines the minimum amplitude of the filtered asymmetry signal necessary to override the bias to forward bias transistor 48 for emitter current flow.

The emitter electrode of transistor 48 is connected to emitter follower resistor 50 which develops a negative voltage when emitter current flows. The emitter is also connected, through a resistive impedance 51, to the base electrode of a transistor 52. Transistor 52 essentially forms the amplifier portion of switching circuit 19 and supplies collector current to a relay actuating circuit 60. Relay actuating circuit includes relay winding 54 and a rectifier diode 55 and resistive impedance 56 connected there across. Relay winding is also connected between a source of biasing potential 53 and the emitter electrode of transistor 52. Relay winding 54 controls the motion of movable relay element 57 which, when energized, applies a trigger voltage from a source 59 to relay circuit output lead 58. Relay circuit output lead may be cou pled to utilization device 20.

The operation of the voice actuator device of this invention will now be explained in connection with the oscillograph traces shown in FIGS. 4A to 41. The horizontal axis in FIGS. 4A to 41 denotes time, the scale being the same and approximately 2 seconds per centimeter. Each of the nine figures shows, to the same time scale, siX different races, each trace corresponding to a different acousti- 6. cal input signal applied to microphone 11. These different input signals are designated S1 to S6 and are vertically aligned for comparison. The vertical axis of FIGS. 4A to 4I denotes the amplitude scale factor which is different for the various traces. The various acoustical signals applied to microphone 11 are as follows: S1 identifies general shop noise; S2 identifies three successive hand claps; S3 identifies three successive percussions such as hitting a metal object with the hammer or dropping a heavy object to the floor; S4 identifies the successive pronunciation of the five vowels by a first male voice; S5 is the same as S4 but spoken by a second male voice; and S6 is also the same as S4 but spoken by a female voice.

FIG. 4A shows the amplified audio signals at the output terminal of audio amplifier 12 obtained in response to applying the acoustical input signals S1 to S6 to microphone 11. FIG. 4B shows the asymmetry signals at the output terminal 36, FIG. 2, obtained in response to applying S1 to S6, as shown in FIG. 4A, to input terminal 25. This asymmetry signal is obtained when filter 14 is entirely removed from terminal 36. that each of applied signals S1 to S6 shows some degree of asymmetry, however, the asymmetry signal due to input signals S4, S5 and S6 is very much larger than the asymmetry signal due to input signals S1, S2 and S3. This is, of course, due to the fact that voiced sounds of live speech have prominent even harmonics which are initially phased to provide an asymmetry envelope. All other input may be classified as noise which has been found to lack prominent even harmonics or else are phased to show little asymmetry.

FIGS. 4C to 4G each shows the effect of exposing the various asymmetry signals, as developed by asymmetry detector 13 in response to the signals shown in FIG. 4A, to different low-pass filters such as the combination of

filters

14 and 16. More specifically, FIG. 4C shows the result of a low-pass filter having a cut-off frequency of 2 cycles per second and a slope factor of 6 db per octave. FIG. 4D shows the result of a low-pass filter having a cut-off frequency of 0.05 cycle per second and a slope factor of 6 db per octave. FIG. 4B shows the effect of a low-pass filter having a cut-off frequency of 0.5 cycles per second and a slope factor of 36 db per octave. FIG. 4F shows the effect of a low-pass filter having a cut-off frequency of 2 cycles per second with a slope factor of 36 db peroctave. And finally, FIG. 4G shows the effect of a low-pass filter having a cut-off frequency of 6 cycles per second and a slope factor of 36 db per octave.

It is significant to note that the greater the slope factor of the low-pass filter the more prominent will be the voice signals such as S4, S5 and S6 in comparison with the noise signals such as S1, S2 and S3. Accordingly, for maximum discrimination between noise and voice, a large slope factor is extremely desirable as best seen by comparing the filtered asymmetry signals of FIGS. 4F and 4G with those of FIGS. 4C, 4D and 4E. Further, the cutoff frequency of the filter for optimum discrimination against noise depends on the particular noise. For example, a cut-off frequency of 0.5 cycle per second provides good discrimination against signals S2 and S3 but not against signal S1. For best discrimination, a cut-off frequency somewhere between 2 and 6 cycles per second has been found best suited for most types of noise as best seen in FIGS. 4C, 4F and 4G. It has been found experimentally that a cut off frequency of about 4 cycles per second for a low-pass filter having a slope factor in excess of 30 db per octave is very effective to discriminate against noises such as S2 and FIGS. 4H and 4I respectively show the effect of a highpass filter on the asymmetry signals shown in FIG. 4F, that is, the asymmetry signals obtained from acoustical input signal S1 to S6 filtered by a low-pass filter having a cut-off frequency of 2 cycles per second and a slope factor of 36 db per octave. FIG. 4H shows the results of a high-pass filter having a cut-off frequency of 0.05 cycle It is interesting to note I per second, a slope factor of 6 db per octave. FIG. 41 shows the result of a high-pass filter having a cut-off frequency of 0.5 cycle per second and a slope factor of 6 db per octave. It is worth noting by comparing FIGS. 4H and 41 that a cut-off frequency of 0.5 cycle per second provides very effective discrimination against signal S1 which is, of course, to be expected since this signal is substantially D.C. Accordingly, a high-pass filter with a cutoff frequency of 0.5 cycle per second is admirably suitable to supplement to the discrimination provided by the lowpass filter to make the voice actuated device of this invention selectively responsive to live speech.

By way of summary then, the desirable discrimination against all sounds except speech is obtained by passing the asymmetry signal through a band pass filter having a high cut-off frequency of about 4 cycles per second with a slope factor of over 24 db per octave and a low cut-off frequency of about 0.5 cycle per second with a slope factor of about 6 db per octave or more.

One of the underlying principles to consider in practicing the instant invention is the frequency with which voiced sounds are used in normal speech. This will, at least to some degree, determine the limits of the frequency band which has to be passed by the band-pass filter. Since voiced sounds do have a rate of reoccurrence somewhere between 0.5 and 4 cycles per second in normal speech, a device having a band-pass filter passing these frequencies is ideally suitable to respond to live speech. Of course such device would also be responsive to a command having a single vowel sound as for example the word stop. If the word stop is shouted into microphone 11, even though there is no repetition of vowels, an output signal would be provided since the asymmetry signal includes many frequency components within the band pass range.

Furthermore, the level detector provides further discrimination in that it permits the switching device of this invention to be set to be responsive only to asymmetry signals, suitably filtered by a band pass filter, which have an amplitude above the selected amplitude. A level detector provides further assurances that other types of noise such as, for example, background noise, percussion noise and hand-clapping would not actuate the device.

There has been described hereinabove a voice actuated device triggered only by sounds having an asymmetric envelope characteristic. Further, the device is not triggered unless the frequency of the asymmetry signal falls within a narrow pass-band and has an amplitude greater than a selected amplitude.

What is claimed is:

1. A voice actuated device for providing a trigger signal in response to selected voice sound comprising:

a transducer means for converting acoustic vibrations to a corresponding electrical signal;

amplitude peak detector means responsive to the positive and negative portions of said electrical signal and operative to derive a positive and a negative envelope peak signal;

means for comparing the positive and negative envelope peak signals and for developing an asymmetry signal commensurate with their difference; and

filter means responsive to said asymmetry signal and operative to develop said trigger signal, said filter means being selected to pass only those portions of said asymmetry which correspond to said selected voice sounds.

2. A voice actuated device for providing a trigger signal in response to selected voiced sound comprising:

a transducer means for converting voice acoustic vibrations to a corresponding electrical signal;

filter means responsive to said asymmetry signal and operative to develop said trigger signal, said filter means having a high cut-off frequency below 8 cycles per second with a slope factor greater than 18 db per octave.

3. A voice actuated device for providing a trigger signal in response to selected voiced sounds comprising:

a transducer means for converting acoustical vibrations to a corresponding electrical signal having substantially the same phasing;

amplitude peak detector means, having a time constant selected to follow said selected voiced sounds, responsive to the positive and negative portions of said electrical signal and operative to derive a positive and a negative envelope peak signal;

filter means responsive to said asymmetry signal and operative to develop said trigger signal, said filter means having a low cut-off frequency selected from the range of 0.05 to 1.0 cycle per second and a high cut-off frequency selected from the range of 2 to 8 cycles per second.

4. A voice actuated device for controlling a switching means comprising:

amplitude peak detector means, having a time constant between 3 and 50 milliseconds, responsive to the positive and negative portions of said electrical signal and operative to derive a positive and a negative envelope peak signal;

means for comparing the positive and negative envelope peak signals and for developing an asymmetry signal commensurate with their difference;

amplifier means responsive to said asymmetry signal and operative to provide an amplified asymmetry signal;

filter means responsive to said amplified asymmetry signal and operative to develop said trigger signal, said filter means having a low cut-off frequency selected from the range of 0.05 to 1.0 cycle per second and a high cut-off frequency selected from the range of 2 to 8 cycles per second; and

switching circuit means responsive to said trigger signal, and switching circuit means assuming a selected switching state when triggered.

5. A voice actuated device for controlling a switching means comprising:

a transducer means for converting voice to a corresponding electrical signal having substantially the same phasing;

filter means responsive to said amplified asymmetry signal and operative to develop a filtered asymmetry signal, said filter means having a low cut-off frequency selected from the range of 0.05 to 1.0 cycle per second and a high cut-off frequency selected from the range of 2 to 6 cycles per second;

level detector means responsive to said filtered asymmetry signal and operative to provide a trigger signal only when the amplitude of said filtered asymmetry signal exceeds a preselected level, said level detector means being adjustable to change said preselected level; and

switching circuit means responsive to said trigger signal, said switching circuit means assuming a selected switching state when triggered.

6. A voice actuated device in accordance with claim 5 in which said filter means has a slope factor at the high cut-off frequency which exceeds the slope factor at the low cut-off frequency.

7. A voice actuated device in accordance with claim 5 in which said filter means has a slope factor of the high cut-off frequency which exceeds '18 db per octave.

5 octave and at the low cut-off frequency of approximately 6 db per octave.

References Cited by the Examiner UNITED STATES PATENTS 8/1965 Dersch 179-1 KATHLEEN H. CLAFFY, Primary Examiner.

R. MURRAY, Assistant Examiner.

Claims

1. A VOICE ACTUATED DEVICE FOR PROVIDING A TRIGGER SIGNAL IN RESPONSE TO SELECTED VOICE SOUND COMPRISING: A TRANSDUCER MEANS FOR CONVERTING ACOUSTIC VIBRATIONS TO A CORRESPONDING ELECTRICAL SIGNAL; AMPLITUDE PEAK DETECTOR MEANS RESPONSIVE TO THE POSITIVE AND NEGATIVE PORTIONS OF SAID ELECTRICAL SIGNAL AND OPERATIVE TO DERIVE A POSITIVE AND A NEGATIVE ENVELOPE PEAK SIGNAL; MEANS FOR COMPARING THE POSITIVE AND NEGATIVE ENVELOPE PEAK SIGNALS AND FOR DEVELOPING AN ASYMMETRY SIGNAL COMMENSURATE WITH THEIR DIFFERENCE; AND FILTER MEANS RESPONSIVE TO SAID ASYMMETRY SIGNAL AND OPERATIVE TO DEVELOP SAID TRIGGER SIGNAL, SAID FILTER MEANS BEING SELECTED TO PASS ONLY THOSE PORTIONS OF SAID ASYMMETRY WHICH CORRESPOND TO SAID SELECTED VOICE SOUNDS.