DE1940082A1 - Method for recognizing speech features and recording them using an electrically controlled typewriter - Google Patents

Description

Patent attorney

Stuttgart-Feuerbach

Short street ö

Tl TT η

- OJi A, F ₁ NaWoIi

STANDARD TELEPHONES AND CABLES LIMITED, 190 Strand, London W.G. 2 / England

Method for recognizing speech features and their recording by means of an electrically controlled typewriter

The invention relates to a method of recognition of speech features and their recording by means of a electrically controlled typewriter,

In the case of speech recognition, the individual features fall into the chronological order of their occurrence and thus result in a series node of variable length. Should now these characteristics e.g. recorded by means of an electric typewriter, this must output the information as a parallel code a fixed number of elements are offered. The at The serial code required for recognition is therefore in one suitable, from an electrically controlled Sohreibmasohine . convert processable parallel code, where the conversion appropriately after the end of a word «,

The object of the invention is therefore to provide a method to recognize speech features and to record them using an electrically controlled typewriter, that makes this possible.

According to the invention, the object set is achieved in that the speech vibrations recorded by a microphone and converted into analog SO electrical wave trains

Dr Le / ßr

August 6, 1969 009808 / 12S6

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

the characteristics short or long by a discriminator. Loud, as well as short and long pause intervals or loud dark or lighter timbre, as well as short and long. Pause interval obtained and stored in a memory device as elements of a code group constructed as a serial code when they occur, that the memory device is queried by the long pause interval characterizing the end of a word and the serial code read out through a code converter matrix into a typewriter controlled by an electrically ge, - W 10 usable parallel code is converted, by means of which the typewriter is controlled, that thereupon the memory contents of the memory device is erased, so that the memory device is available for storing the features of the following word.

The invention will now be described in detail with reference to the figures, be written. ''

It shows:

Figure 1 a detection and conversion device,

'20 consisting of a discriminator, a memory

device and a code converter;

FIG. 2 shows an alternative solution for the discriminator;

Figure J shows another discriminator solution for other speech features.

BAD GRiGiNAL

, .. _; . ;. * '**' ■% 1140082

JUGreen - B, J, H & lliwell - A ₁ F ₁ NeWeIl 2 ^ 1-1

It has already been suggested that a length-variable eode group must be composed of a sequence of idle elements of different types and must be terminated by an element indicating the end of the god group. Since speech sounds are identified by such code groups and a speaker inevitably pauses between two speech sounds, the usual element indicating the end of a code group is an interval that extends over a certain minimum duration, in which the amplitude does not exceed a certain visual value. For the conversion into such a code, elements similar to a Morse code can be used, whereby in addition to the end character two other different characters are required. Diesis then represent a short sound followed by a pause and a long t «aut followed by a pause. The speaker generates these two sounds by means of the signs of sounds, e.g. by following a soft palate stop / d / a short or long vowel like / üi / in thick or / da / in there * '

The conversion unit shown in Fig. 1 consists of a discriminator, outlined with a line and with ί bezei-ohneti its 4 exits A * 1 * Q and D with a Godew & Nälep, also-?

if bordered by a dashed line and labeled 2, is allied.

The code converter consists of a shift register, denoted by 3 in a row, which consists of k bistable circuits, and a binary counter, denoted by 4 in a dashed line, which is made up of 2 bistable circuits.

The outputs of all 6 bistable Sehaltkreise lead to a decoding matrix for a parallel code, which is bordered by a line is denoted by 5. The entrance of the discriminator 1 is labeled E and receives its input signals from a microphone.

The discriminator 1 should be short and long sounds as well as short ones and be able to differentiate between long pause intervals. The boundary between short and long sounds is approximate

■ 1540082

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

at 150 ms. Purely the distinction between short and long pause intervals depends on the Spreohgeso speed of the Speaker. For slow speakers, the threshold is around 500 ms and decreases when speaking faster down to the vicinity of 200 ms. The time distinction is made by logic circuits, the "integrating monostable circuitry "should be called and meet the following punctures:

1. The output is 0 when the input is 0.

2. If the input changes to 1, the remains The output initially changes to 0, only when the input signal 1 lasts a certain time t the output on 1 for the duration of the persistence.

A realization of such an integrating monostable Circuit consists of a capacitor, which consists of a Constant current source is charged. A threshold value circuit, e.g. from a Schmitt Trigger exists.

The simplest form of the discriminator 1 is shown in FIG. It consists of an envelope detector 11, which is followed by a threshold value circuit 12, the output signal of which corresponds to the sound pressure of speech. This output signal is once to the input of a first integrating monostable circuit 15 and also to the input of a first inverter 14, whose output signal corresponding to a pause in speech is applied directly to the output terminal C and also to the input of a further integrating monostable circuit 15, the output of which is connected to the Output terminal D is connected. The output signal of the first integrating monostable circuit is applied directly to output terminal B and, after passing through a further inverting stage 16, to output terminal \ A. The envelope detector 11 can have a bandpass filter I7 and a 009808/1256

ί 940082

J.U.Qreen - B.J. Halliwell - A.P. Newell 2-1-1

Device for automatic volume control 18 connected upstream be.

A signal at output terminal D therefore shows a long one Pause on and serves as an indicator of the end of a code group 5 "Code element. The appearance of this signal causes from output a signal to the outputs of the decoding matrix which corresponds to the state of the bistable circuits of the shift register 3 and the binary counter 4. An output signal at output B indicates * that a long sound has been received an output signal at output A makes a short sound. These output signals are each at one of the two inputs of the first bistable circuit of the shift register 3 · The occurrence of a signal thus causes a transition from 0 to 1 or 1 to 0. When a signal occurs at output G of the discriminator, which is connected to the storage line 31 1st and on which a pulse occurs after each code element, if a code element indicating the end of a code group is not present at D, the content of the bistable circuits becomes of the shift register 3 relocated to the next following circuit. This storage impulse occurs at the beginning every pause in speech, but it should be remembered that the same beginning also causes the integrating monostable circuit 13 to change to the output signal 0, if this is not already pending, it will therefore be the Reset time of the integrating monostable circuit chosen so that this is greater than the time required for storing in the shift register, so that in the register still has the signal present at output B. The beginning of a break is saved and not that after Resetting at output A occurring »The same transfer pulse is used as a counting pulse for binary counter 4 » Shift register 3 and binary counter 4 are reset by a 0 to 1 transition on the return line 32 · Off the memory contents of the 6 bistable circuits that make up The shift register and the binary counter are set up

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

the information for a 6 bit parallel code is taken, which is decoded by the matrix decoder 5. The matrix decoder has a number of row groups, of which in of Figure two are shown and denoted by 53 on. Each row of the groups is connected to a line from one of the 6 pairs of columns. Each group of lines is closed with an AND circuit 50, the output of which is an output of the Matrix decoder forms. One pair of columns is included in each case assigned to one of the six bistable circuits, with one column in each case at one output of the circuit. Each AND gate 50 has seven inputs, one of them is the call line 51 and the others are each with one Line of a row group and thus a line of the six pairs of columns jr'So only one of the AND circuits 50 Output when a read pulse is applied to the polling line 51. Code converter and decoding matrix are designed together in such a way that an output signal is only output at output D after a code group end signal. For this purpose, the output D is connected to the input of a monostable Circuit 7 connected, the output of which is a pulse of predetermined duration when each code group end signal occurs gives away. The rising plank of this impulse serves as-. Call pulse for the de-coding matrix. By the sloping Planke is reset shift register and binary counter, by placing between the output of the monostable circuit and Reset line 32, a reversing stage 8 interposed will. In those cases where the code converter and the Decoding matrix are not used together, remains the monostable circuit 7 and the inverter 8 as Part of the code converter, so that the rising edge a front code group end signal at output D excited ■ Impulse for reading out the memory contents of the six bistable switching stages is used and the sloping plank to reset them.

00900871256, /.

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

The output signal of the monostable circuit may now be applied directly to the associated inputs of all AND circuits 50. In this case, background noises picked up by the microphone can be evaluated as one of the short code elements. In order to overcome this problem, the decoder can be designed so that the system can be switched from the storage state to the stored state with the aid of two code elements. For this purpose, there is only a direct connection from the output of the monostable circuit J to the call line 51 of the row group 56 / while all the other call lines 51 are connected via a two-input AND circuit 57. The signal for the second input of the AND circuit 57 is supplied by a bistable circuit 58. In this case, * the bistable circuit 58 receives its control voltages from the output of the row group 56 * which switches the system into the storage state, and from the output of the row group 59 which switches it into the storage state. In the storage state, a pulse of the monostable circuit 7 is blocked from all call lines 51 of the decoding matrix with the exception of the call line for the digit group 56. The code element corresponding to the line group 56 switches the system to the unload state, in which all code elements of the other line groups including 59 are stored. wherein the code element 59 switches the system back into the storage state. The code element appearing at line group 56 is, as already described, selected to be so long that it no longer occurs in normal language alternation.

The selection of the stages for the shift register and the Binary counter must take into account the terminal used take place. The arrangement described so far was intended for controlling an electric typewriter that Usually around 50 different control signals for all

009808/1256

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

Characters and switchings required. When selecting a four-stage shift register in conjunction with a two-stage Binary counter results in a 6-bit parallel code with 64 different code lines. The surplus Code lines can then be used for other purposes such as electrical control of the typewriter.

The shift register can consist of four groups of codes immediately preceding code elements 30 form possible different code lines, all of which are characteristic are. With longer code groups, represented by the remaining 34 code characters, only the last four code elements remain stored in the shift register before the final character, since the previous ones already go through this and the Have influenced the position of the binary counter.

The envelope curve detector 11 of the discriminator 1 can now be switched off an oil rectifier with a connected filter circuit. In this training, however, a compromise must be made between a sufficient tent constant for the Sieving and a sufficiently small time constant so that a subsequent short pause is still noticeable. This sets a limit for the minimum value of a break that can be taken from the arrangement can still be recognized and therefore also for the following speed of the recognizable individual code elements

The arrangement shown in FIG. 2 bypasses these difficulties and replaces the discriminator L shown in FIG. 1. In this arrangement, a bandpass filter 20 is connected to the microphone input E a differential amplifier 21 'connected to the other input is connected through a rectifier and' filter circuit 22 at the filter output. The differential amplifier

009 8 0 8/1256 Α

J.U. Qreen - B.J. Halllwell - A.F. Newell 2-1-1

a threshold value circuit 25 * is followed by an inverter 25 and an integrating monostable circuit 25. The time constant of this circuit is chosen so that it is slightly larger than the period of the lowest Frequency of the pass band of the filter 20 is. As will be shown later, the output of this arrangement is large for pauses and low for speech sounds. When the output signal is the reverse of that the threshold value circuit 12 of the figure! are those of the integrating monostable switching stage 25 following subassemblies are the same as those of the threshold value circuit 12 downstream with the exception of the other position of the inverter 14. For this purpose, the differential amplifier 21 is biased, this bias source being indicated at 26.

If one initially disregards the differential amplifier 21 and the rectifier and filter circuit 22, i.e. if the output of the filter 20 is thought to be directly connected to the input of the threshold value circuit 25, the threshold value circuit would emit a pulse at its output every time the instantaneous value of the The signal passed through the filter exceeds the threshold value. In the case of a signal of sufficient duration and amplitude, the repetition frequency of these pulses is at values which are within the pass frequency range of the filter 20. After passing through the reversing stage © 2h , pulses are produced at the output which are necessarily short compared to the response time constant of the integrating monostable switching stage 25, so that no output signal is emitted from it. If, however, the peak value of one or more oscillations of the signal passed by the filter does not reach the threshold value, the Unikehr stage 24 emits a pulse that is long enough to control the integrating monostable switching stage 25 Threshold switching emitted pulse train a pulse fails. The integrating monostable switching stage 25 is therefore at

Pauses a signal is given «wtSiyend it otherwise remains blocked. 009808/1258

J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1

This shows that the arrangement of filter, threshold value circuit, Inverse stage and integrating monostable switching stage represents a device that indicates when a certain threshold value is undershot in a signal, even if it is only in the case of a single oscillation. Integration the differential amplifier and the rectifier and filter circuit acts in the sense of an automatic Level regulation without affecting the operation of the threshold value circuit, with the Bias voltage source 26 supplied bias voltage this level oil control makes it ineffective below an adjustable value.

In addition to the Morse code type code described so far, a number of other serial codes of variable length can be used can be used. Such an alternative, which only changes the discriminator described in FIG conditional, is a ternary serial code that has the same code group key as before, the other two code elements of which do not differ over time. For a code with more than 2 code elements, both the Discriminator as well as the shift register can be modified. If a quinary serial code is used, for each level of the shift register requires two storage elements so that # üa? each of the four code elements except for the code group terminator can be stored in a stage. Under certain circumstances, a quaternary serial code there are sufficient restrictions peculiar to it, so that it is possible to change the individual stages of the shift register to be formed with only one storage element. This is possible if the code group end signal from the others Code elements can only be distinguished by crossing a time threshold. A discriminator for one quaternary serial code is shown in Figures Ja and Jb!

The discriminator shown in these figures is for Deriving a serial code of variable length, which-

■ 0098.08 / 1256-

J.U. Green - B.J. Halliwell - A.F. Newell 2.-1-1

as code elements, sounds with a light tone color, sounds with a dark tone color, pauses and, as a fourth element, a code group terminator contains, which is given by the fact that a pause exceeds a predetermined minimum value. the The first three code elements only contain properties of the input signal at a specific point in time. If the amplitude of the envelope of the input signal does not exceed a certain threshold value, this is evaluated as a pause. If it exceeds this threshold, it is evaluated as one of the other two code elements "loud lighter or darker tone color". As a result, the code contains a system-related restriction option that successive code elements do not can be of the same kind. So there are only two possible transitions from one code element to the other. So A sound with a lighter tone can only be followed by a pause or a tone with a lighter tone. A sound of dark timbre Only a pause or a sound with a lighter tone follows, or a break with a sound with a dark or lighter tone. if the initial state is known, describes a binary one Statement for each subsequent transition the process clearly. The same binary statement makes a single transition as well as the same transition between two code elements represented in a sequence of transitions, since each code group begins after a pause and with a Break ends, it is clearly defined by specifying the transitions.

This code made up of four sound elements, namely from the elements sound with a light or dark tone color, Pause and long pause as code group terminator, does not contain the sound elements themselves, but is suspended binary statements about the transitions from one sound element to another. So this code consists of seven

00 98 08/1256

J.U. Green - B.J. Halllwell - A.P. Newell 2-1-1

different characters, six for the possible transitions and one for the code terminator given by a long pause.

The conversion unit shown in FIG. 1 is modified for this code in that the discriminator 1 of the figure is replaced by a discriminator as shown in FIGS. 1 a and b. This discriminator is shown graphically in two successive parts and is also described in this way. The first part that is in figure Yes. | l IO is shown, recognizes from the input signal whether a sound with a light tone color (F), a sound with a dark tone color (V) or a pause (S) is present, whereby if an element is not present, the negation as (F), (V * ) and (S) is displayed.

In the second part shown in FIG. 3b, the transitions are determined and at outputs A, B, C and D. in a form that can be evaluated by the subsequent code converter made available.

In FIG. 3a, the signal supplied by a microphone and present at input E is first fed to a device for automatic volume control 301, which has an envelope detector 302; and then a threshold value circuit 303 is connected downstream. The output signal of the threshold value circuit 303 is applied directly to the output B * and to the output S via an inverter 304. Furthermore, a high-pass filter 305 with a cut-off frequency of approximately 1,000 Hz and a further envelope curve detector 306 are connected to the output of the device for automatic volume control 301. The output signals of the envelope curve detectors 302 and 306 are now compared with one another in a differential amplifier 307, and its output signal is fed to a further threshold value circuit 308. The differential amplifier 307 thus compares the entire speech signal with the components lying above 1090 Hi. Loud, light timbres differ from those with dark timbres in that in their 009808/1256

.U. Green - B.J. Halllwell - A.P. Newell 2-1-1

Spectrum does not contain low frequencies, whereas high frequencies can also be included in the spectrum of sounds with dark timbres. An output signal from the threshold circuit 308 indicates that "no sound of a light tone color" is present and an output signal of the inverter 309 connected downstream of the threshold value circuit 308 indicates that "no sound of a dark tone color" is present, provided that an input signal of sufficient level is present, which as "no break" was evaluated. These output signals are therefore applied to one input of an AND circuit or 311, to the other input of which the "no pause" signal is applied. The output signal of the AND circuit 310 or * is directly at the output V or F and after reversal in an inverter 312 or 313 tm output V or F.

In the arrangement according to FIG. 3b, there is a monostable circuit 314 or 315 or 31 at the input F or T? or S. Your outputs are with AND circuits 31? ... connected to the second inputs of which the input terminals F, V and S are connected. The monostable circuits 314 * 515 and 33.6 are triggered by the rising Flaak® of a negating output, i.e. at the end of a feature to be recognized. Timbre 'from an impulse. If the sound with he Her timbre is followed by a sound with a dark timbre * awm süfcaXfcet öS, β AND circuit 317 for the "item * -4" © Xragulses. So an output signal de ^ '^; .) * yilD Söiia (i «« isg 34f a ^ n ÖBsrgasig from F to V. Ent apvmlmmS, "ktmasiettiÄ Ausgangsslgnale the AND circuits 33ι3 · ■■■ · ^.'3siir; ift» apgasee vo® ** - > s * ▼ ^-:; f »■ --- s, sf

mm i ■ - -. V " pirn Äus ^" n® signals from these AND circuits li ^ öii m -mm%. "ODIE-Schiiitung""- 323 and 324, whereby each time \ ftlw Ubevgs ^ from- %% Tmm to another of the two ai1i8aa # # ln $ OR-circuit lasts. The ÖÜE * Setial | ijng'323 barn. 324 leads see the "Exit- £ .-!» &# ». f * Furthermore, the output is each of the AND

J.Ü. Green - B.J. Halllwell - A.P. Newell 2-1-1

Circuits connected to one input of a further OR circuit J525, so that at each transition at their output a signal is applied to output terminal C. The input terminal S is connected to the output terminal D. via an integrating monostable circuit 326, see above that at D there is an output pulse when a break is a exceeds the prescribed minimum, i.e. a code group key is present.

8 claims

3 sheets of drawings, 3 figs.

009808/1256

Claims (7)

J.U. Green - B.J. Halliwell - A.F. Hewell 2-i ~ l Claims \ Ιλ Method for recognizing speech features and their recording by means of an electrically controlled typewriter, characterized in that the features short and long sounds, as well as short and long pause interval or sound of dark or light tone color, as well as short and long pause interval and when they occur are stored in a memory device (2) as elements of a code group structured as a serial code so that the memory device (2 ) queried and the serial code read out is converted by a code converter matrix (5) into a parallel code that can be used by an electrically controlled typewriter, by which the typewriter is controlled so that the memory content of the memory device (2) is erased so that di e memory device (2) is available for feeding in the features of the following word. 2. A method for recognizing speech features according to claim 1, characterized in that for recognizing the speech features "short or long sound" and "short or long pause interval" the electric wave trains (E) emitted by the microphone, analogous to the speech vibrations, after a dynamic narrowing an amplifier with automatic volume control (18) and limitation to the characteristic frequency content by a bandpass filter (17) are fed to an envelope detector (11) so that the output signal corresponding to the envelope of the speech signal is fed to a threshold value circuit (12) which only lets signal parts through, Dr Le / Gr 00 9 8 08/1256
6th August 1969 . - 16 - JUGreen - BJHalliwell - .AFNewell 2-1-1 _ic -, - I. whose amplitude exceeds a predetermined value, no longer associated with a pause interval as a noise, so that the output signal of this threshold value circuit directly controls a first integrating monostable circuit (1>) which, when the input signal is present, only emits an output signal when the input signal exceeds a predetermined period of time that the output signal of this integrating monostable circuit (13) is fed directly as a feature for a long sound or after inversion in a first inverter (16) as a feature for a short sound to the output B or A Threshold circuit (12) "inverted in a second reversing stage (14) once directly as a feature for a pause interval output C and after a predetermined minimum duration has been determined by a further integrating monostable circuit (15) as a feature of a long pause interval and code group terminator n is fed to output D. 3. A method for detecting speech features NaOH to claim 1, characterized in that interval breaks the cast for recognizing the speech features "short or long volume" and "short or long ¹¹ of the microphone, the Sprachsohwingungen analogue electrical waveforms (E) via a Bandpass filter (20) can be applied directly to one input of a differential amplifier (21), at the other input of which the envelope curve obtained in a rectifier and filter circuit (22) and an adjustable fixed bias voltage (26) are applied so that the output signal of the differential amplifier (21 ) Leached via a threshold value circuit fej) and a first reversing stage (24) to the input of a first integrating aonostable Sohaltkreises (25), whose output signal once directly as a feature "pause interval" to the output 0 after determination of a predetermined minimum duration through a second integrating monostable circuit (15) 009808/1256, J.U. Green - B.J. Halliwell - A.P. Newell 2-1-1 the output D as the characteristic "long pause interval" and Godegruppensußzeiohen and after inversion in one second reversing stage (14) is fed to the input of a third integrating monostable Sqhaltstufe (13), their output signal directly to output B as a feature "long sound" and, after inversion in a further reversing stage (16), fed to output A. as a "short sound" feature will. 4. A method for recognizing speech features according to claim 1, characterized in that for recognizing the speech features "loud lighter or darker timbre" and "short or long speech interval" from the output of the microphone, the speech vibrations analog electrical wave trains (E) after a Dynamic reduction through an amplifier with automatic volume control (301) initially criteria for the presence and the non-presence of each of the language features can then be derived from these criteria by means of logic circuits, indicators for the transition from one feature to another are formed, which in stored in the memory device (2) as a serial code and converted into a parallel code in the deodification matrix (5) « 5. The method for recognizing speech features according to claim 4, characterized in that »d & S for deriving the criteria for the respective presence and absence of one of the speech features dfts output signal of the amplifier with automatic volume control (301), once directly to a first envelope detector (302), the other time to an external envelope detector (306) via an Itoch pass (305) that only allows frequencies above about 1000 Hz. is supplied that the output signal of the first envelope detector _:. -y_ t | nm &'taefkft an- »int, first threshold value circuit - (392) * das λ s.« **, «- **** ' --ijtotop. a "differens ^ alijylrei * 0Ql) * in this BAD ORiGfNAL - 18 J.U. Green - B.J. Halliwell - A.P. Newell 2-1-1? Λ; second input, the output signal of the second envelope curve detector (306) is applied to a / second / f threshold value circuit (30b) is applied that from the output signal the first threshold value circuit (303) directly the criterion "no pause interval" (S) and after inversion in a first reversing stage (304) the criterion "pause interval" (S) and by comparison with the output signal of the second threshold value circuit which is inverted in a second inverting stage (309) (30ö) in a first AND circuit (311) directly the criterion "Loud, bright timbre" (P) and after inversion in one third inversion stage (304) the criterion "no sound, lighter tone color" (P) and by comparison with the non-inverted one Output signal of the second threshold value circuit (30ö) in a second AND circuit (310) the criterion "loud dark tone" (V) and after inversion in a fourth inversion stage (312) the criterion "no sound of dark timbre" (?) Is derived. 6. The method for recognizing speech features according to claim 4 or 5 *, characterized in that each of the speech features indicators for the transition from one feature to the other are obtained from the criteria for the presence and absence of a feature in that the criteria for the presence of a feature directly to the first inputs of AND circuits (317 ... 322) and the criteria for the non-existence of second inputs of these AND circuits are fed in such a way that signals for the transition from one feature to the other occur at their outputs, that from these signals, based on the fact that the initial and final states of each code group are known as the "long pause interval ¹ *" and that there are only two options for changing a speech feature »Identifiers A, B and C derived and are stored in the memory device (2), while integrating from the criterion "pause interval" (S) by means of an * The code group terminator is derived from the monostable circuit (326) J.U. Green - B.J. Halliwell - A.F. Newell 2-1-1 7. A method for recognizing speech features according to claim 1 and one of the subordinate claims 2 ... 6, characterized in that the memory device (2) consists of a four-stage shift register (3) with a two-stage binary counter (4) connected «that the .output signals A and B of the discriminator (1) are stored in the first stage of the shift register (3) so that this shift register continues to store by a C signal, that the binary counter counts the number of further stores, that the memory is read by a D signal and is cleared by an inverted D signal thereon. b. Method for recognizing speech features according to Claim 7 » characterized in that the outputs of the four-stage shift register (3) and the two-stage binary counter (4) are followed by a deodulation matrix that each of the outputs is connected to a line of b groups of lines (53), which are each completed with a 7-input AND circuit (50) that these AND circuits (50) emit an output signal when a call signal is applied to their 7th input (51), the call signal by means of an AND Circuit (57) and a bistable circuit (5 ») is triggered by the code group final character (D). 009808/1256 ok Blank page