FR2554267A1 - SPEECH CONTROL APPARATUS - Google Patents

Abstract

THE INVENTION RELATES TO AN APPARATUS ALLOWING THE TRANSFORMATION OF AN INSTALLATION INCLUDING A COMPUTER INTO A SPEECH RECOGNITION APPARATUS. THE DEVICE IS PLACED BETWEEN A KEYBOARD 2 AND THE COMPUTER. IN SPEECH MODE, A DIGITIZED AUDIO-FREQUENCY SIGNAL FROM A MICROPHONE 6 IS COMPARED BY A MICROPROCESSOR 12 TO DATA FROM A MEMORY 20 IN ORDER TO DETERMINE THE SPOKEN WORD OR EXPRESSION. THE DETERMINED IDENTITY ALLOWS THE OBTAINING OF A CHAIN OF CHARACTERS INTENDED FOR THE COMPUTER. APPLICATION TO SPEECH CONTROL OF DEVICES.

Description

The present invention relates to a speech-controlled apparatus and more

  precisely such a device intended to ensure speech recognition with

  a data processing apparatus.

  Many types of speech recognizers are known that can be trained to recognize alphanumeric characters spoken by a particular user. Such speech recognition apparatus can not be easily connected to an existing installation comprising a computer for use with signals from a keyboard and for example using strings of data for triggering the execution of a computer. a particular program. Thus, the introduced messages may require significant changes to the software so that an installation can be controlled by a single voice. The changes require time and

  a lot of effort from an experienced programmer.

  The invention relates to a speech-controlled apparatus which may constitute a voice input system of an installation comprising a computer and

  not requiring a major change in

  skies. The invention thus relates to such a speech-controlled apparatus intended to be placed between the output of a manually operated keyboard of the type which transmits digitally encoded signals which characterize any one of the keys when it is controlled, and the input of a data processing apparatus so that the assembly constitutes a speech recognition facility in which a comparison device is for comparing a digital signal derived from an audio input signal as a result of of a sentence or word uttered, to several words of data contained in a previously programmable speech message data memory, the data words characterizing valid words or phrases, so that, if appropriate, those of words or phrases that have been stated are determined and

  that signals are transmitted to a translation device.

  these signals identifying the word or phrase thus stated, the translation device being controlled by the signals from the comparison device and being intended to determine, from a second memory

  previously programmable data, data characteris-

  maintaining equivalent operations of keys of a connected keyboard, and causing the transmission, by a signal generating device, of a digitally encoded data signal chain, this chain characterizing the set of equivalent operations

of keys.

  The apparatus may be constructed such that, as a result of an audio input signal comprising

  an articulated chain of valid words, a series of

  interdependent statements of valid words contained in the stated string determines a string of data signals

  output characterizing the stated string. The inter-

  The translation of valid word translations into the stated strings can be determined at least in part by the pauses of predetermined duration contained in the audio input signal. Alternatively or additionally, the interdependence of translations of valid words into the stated strings can be determined at least in part by predetermined changes

  height of the audio input signal.

  The device can be made in such a way that

  the data signal output chains are deter-

  in part by the information received from the

data processing.

  A switching device may also be incorporated, the apparatus being controlled, in a first state of the switching device, so that it operates as previously described and, in a second state of the

  switching device so that it enables communication

  direct cation between the connected device

  data and a connected keyboard.

  The apparatus may include a memory for storing data defining a representation of the characters transmitted by the apparatus when the switching device is in the first state so that when the switching device is switched to the second state, the device may cause the display device associated with the connected keyboard to display a suitable record of the data transferred to the connected data processing apparatus or

from.

  Other features and benefits of a

  speech-controlled apparatus according to the present invention.

  better from the description which will follow, made in

  reference to the accompanying drawings in which:

  FIG. 1 is a diagram of an example of an installation

  computer system with which the device can be used;

  Figures 2 and 3 are synoptic diagrams

  Figure 1, for use with two different independent devices controlled by speech according to the invention; Figure 4 is a more block diagram

  of the speech-controlled apparatus of the

  Figure 2; Figure 5 is a block diagram of another apparatus according to the invention; FIG. 6 is a flowchart illustrating

  a feature of the apparatus of Figure 5.

  We first refer to Figure 1; the apparatus is intended for use with a conventional installation 1 comprising a computer to which one or more coupling circuits between the man and the machine are connected, for example a keyboard display unit 2. Although it has been shown the assembly 2 connected directly by a common line 3, 3 'of data and a male-female connector 4, 4', it should be noted that this connection can be carried out by a land line connected to a computer placed at a distance or by a link -switched telephone joining a remote database that can

thus be reached.

  Reference is now made to Figure 2; the apparent

  essentially comprises a signal conversion assembly 5 connected in the line 3, 3 ', a microphone 6

  and a switch 7 for controlling speech input.

  The signal conversion assembly 5 has a mode select switch 8 which is used to select the operating mode of the set as will be described later. A corresponding male connector 9 and a female connector 10, intended to allow a simple interconnection of the signal conversion assembly, are intended to interconnect with the female connector 4 'and the male connector 4 of the assembly of display 2 and installation 1 comprising

the computer respectively.

  In an advantageous arrangement of the conversion assembly shown in FIG. 3 referred to below, the selection of the mode as hereinafter described is physically separated from the translation function of the assembly so well. that a talk circuit console 5 'can be placed near the visual display unit 2 and a translation unit 5 "can be placed at a distance, the data transfer being carried out by a suitable interconnection 11. case, mode selection switch 8, microphone 6 and input control switch 7

  of speech are connected to the console 5 '.

  Referring now to FIG. 4, it should be noted that the combination of the console 5 'and the translation assembly of FIG. 3 has an operation similar to that of the signal conversion assembly of FIG. , and we now describe the operation of a

such device.

  Using the mode switch 8 placed on the 5 'console, the user can select the operating mode of the apparatus. This selected mode is detected by a microprocessor 12 receiving input signals 13 and according to the chosen mode, the microprocessor follows

  one of six programs. To simplify the description,

  the six modes of operation of the device 5 are

  called: i) transparent, ii) macro, iii) text, iv) learning

  weaving v) loading and vi) safeguarding.

  It should be noted that the mode switch 8 can be formed by a series of keys on the console board. In a variant, the microprocessor 12 can detect an alphanumeric character string and / or control characters that select the mode of the set 5. In the latter case, the microprocessor 12 can be programmed so that it inhibits the transfer of characters.

  the selection chain at the facility including the computer

  nator. In transparent mode, the microprocessor 12 causes an electronic switch 14 to connect its input to an output that is connected to a wire 15. Thus, signals transmitted by the unit 2 through the connection 3 are transmitted by an OR gate 16 to the

  3 'connection to the installation including the computer.

  It should be noted that when the visual display unit 2 is controlled by simplex signaling - i.e. when the transmitted character is automatically displayed - characters received by the connection 3 are also preserved in the order frame analysis in the screen memory 17,

  for a purpose described hereinafter.

  The functions 14 and 16 can be incorporated in the microprocessor system and are not necessarily constituted

  by carriers or physical switches.

  When the visual display unit 2 is controlled by duplex signaling, the signal reflected by the connection 3 'is stored in the screen memory 17 and returned to the display unit 2 by the wire 3 so that it is displayed in the usual way. Data from the computer terminal, regardless of the signaling mode, is stored in the memory

  screen 17 and transmitted by the wire 3, in a semi-

  lar. It should be noted that, in transparent mode, the unit or assembly 5 has no apparent effect on the transfer of data between the display unit 2 and the installation comprising the computer. However, the screen memory 17 is constantly updated so that, after passing any of the other modes of operation in transparent mode, the microprocessor 12 can surely cause a rewrite on the screen of the screen. unit 2

  display of the current relevant data.

  It should be noted that the transparent mode is also

  used for speech input. Thus, when switching

  7 is activated while the set 5 is in transparent mode, a translation of the voice input to the data output is performed.

  as described later in this specification.

  The macro, text and learning modes are used for programming unit 5 to

  its use with the particular computer

  the particular databank with which it is associated. In each of these three modes, the microprocessor 12 causes the switch 14 to connect the input connection 3 to the microprocessor 12 by

  17. The data received is not trans-

put towards connection 3 '.

  The macro mode allows a user to enter data strings using the keypad of the visual display unit 2 so that they are transmitted to the computer, when the mode is chosen by a voice signal. input at a later time, and the transmission constraints of each such string. For example, when it is necessary to transmit to the installation a chain such as "10.0, 40.0, 1.0, 1E6," CR "", "CR" denoting a carriage return, for the determination for example default parameters, the user introduces this string from the keyboard in macro mode. The received channels are stored in a data memory 18

by the microprocessor 12.

  After the introduction of all the channels or any number of such strings, as necessary at the moment considered, these strings possibly comprising both alphanumeric characters and / or control characters, the user switches to

the text mode.

  In text mode, the microprocessor 12 controls the transmission of each of the channels of the memory 18 to the unit 2 in turn, allowing the user to check each of the strings and to introduce, from the keyboard, the word or the phrase to use to identify

  the respective chains. In the example shown above,

  microprocessor 12 causes the display of the

  chain "10.0, 40.0, 1.0, 1E6, 'CR'". Following this statement,

  chage, the user can introduce the phrase "parameters

  "default" using the keypad to set the channel.

  When the textual definition of parameters

  is complete, switch 8 is used to select

  learning mode. In this mode, the microproces-

  12 controls the display by unit 2 of each of the

  sentences introduced earlier and the user

  switch 7 and speak the phrase in the microphone 6. The microphone signal 6 is transformed

  by an analog-to-digital converter 19 and is

  served by the microprocessor 12 in a sentence parameter memory 20 with an identifier of the

  appropriate string in memory 18.

  It should be noted that a new user of the appa-

  it can switch to learning mode and give a

  replacement set of phrase parameters without reintro-

  create new macro-strings or new texts.

  Additional macro-strings and the associated text can

  may be introduced by switching into the modes

  and introduction of new data strings and existing data strings can be modified

  without changing the associated signals of text or learning.

weaving.

  It should be noted that, in text mode, the microproces-

  In this example, each of 12 compares each text input signal with all previously retained text input signals so that an identical title is not given to two different data strings and does not transmit a suitable message to the data unit. display 2 in this case. The loading and saving modes are used so that the existing data of the speech parameters and / or the text and cha data can not be stored either in a secondary memory (not shown) of the set 5 or the where appropriate, in the memory of the installation comprising the computer. So, if there is

  several users of the device, at the end of the

  the current user can switch to

  save and introduce, for example, "save as

  Fred's speech meters "or" save Fred's text and speech parameters "so that, at the beginning of a subsequent session, the data can be recalled

  and the user can immediately begin to trans-

  put the lyrics, by switching to loading mode and by introducing "load Fred" from the keyboard

of unit 2.

  After programming unit 5, the user

  brings the device back to transparent mode and the microprocessor

  12 causes the reconstruction in memory 17 of

  the display (if any) of unit 2.

  When the speech input switch 7 is then controlled and when the expression "fault parameters" is stated, the microprocessor 12 forms a device for comparing this signal after transformation in the converter 19, to the speech parameters contained in the 20. When the stated expression is identifiable, the microprocessor 12 powers the translation device and uses the associated identifier to restore the proper channel of data from the memory 18, thereby causing the data to be transmitted to the data bank. or at the installation via the 3 'connection. The transmitted string

  data is also returned to display unit 2.

  Where the expression stated can not be

  identified by the microprocessor 12, the conversion unit 5

  signal can be programmed previously so that it transmits either a string of text

  introduced in text mode, ie a null string

  mediate the connection 3 '. In a variant or in addition, a message indicating "invalid speech" may

  to be transmitted via connection 3 to unit 2.

  When character strings are transmitted independently of characters received from the computer in a hurry, in some cases, voice commands may be accepted but the screen format is not well structured. An installation is therefore incorporated so that the transmission of characters depends on the transmission of characters by the host computer. We

  consider now an example of a screen in

  nable. Guiding introduction type 1 Wait then guiding type 2 Before the next part

a macro-string introduced

  Once extracted or extracted, text may appear on the screen before guidance and the user does not see a pleasing image or image equivalent to the signals inputted by the keyboard. It should be noted that, aside from the phrase or phrase introduction facility, the set 5 may

  also fulfill all the functions performed by the

  known revelations of speech recognition. Thus, when a user pronounces for example '7', '5', '1', 'RETURN', the output signal of connection 3 'is "751'CR'". It should be noted that all the speech recognition functions of the set 5 are transmitted without any new programming of the installation or of the data bank being necessary. It should also be noted that the functions of the unit may be fulfilled by a combination of a translation set and a commercially available speech recognition apparatus, and that some of these

  functions may be incorporated into the installation

the computer.

  We now consider in more detail some

  features of the invention.

  The comparison device has the following properties. At certain intervals, a representation

  the short-term digital spectrum of the audiofre-

  quences introduced is computed, each representation being called "frame" .- The signal is thus transformed into a sequence of frames

I I I

  20 .... f. f each frame representing the short-term spectrum of the

  audiofrequency signal transmitted at this time.

  This sequence introduced .... f1 fI fI is compared to a set of "references" which each represent valid expressions or words, in the following form: the word Z (W Z) is represented in the form of a sequence

of frames: -

  The comparison device determines which sequence of words best explains the sequence of frames introduced, the sequence which best explains the frames introduced.

  is for example WXWY..WZ as described in J.S.

  Bridle and M D Brown "Connected Word Recognition Using Whole Templates", Proc. Inst. Acoust. Autumn Conf., Nov. 1979, p. 25-28 The transmission of the signals by the comparison device to the translation device is carried out

as follows.

  Information from the comparison device

  reason allows the recognition processor to trans-

  put the following information for each identified reference (pitch code), if any Scale identification Start time End time reference The last elements can also be called start times and duration or duration and stop time. The start times and downtimes are those to which the algorithm determined that the received broadcast contained the reference

  identified and the possible height code encodes the information

  height profile for the specified period.

  As the moment at which recognition information

  birth is transmitted is undetermined, the processor

  recognition system also transmits silences as a reference

  recognized frequency at a given frequency so that the comparator receives the silence information at the end of the stated sentences. We now consider how the signal generating device transmits a string of characters. A macro-definition is associated with each reference, and it can itself call other macro-instructions and above all it can

depend on time.

  This arrangement can give very important possibilities because if a person indicates for example "cent", it can often wish the transmission of the signal "1" "0" "0" "carriage return", but if a person says "one hundred and twenty- four ", it may be desired to transmit signals" 1 "" 2 "" 4 "" carriage return ". This possibility is impossible to obtain with the known systems having unique identifiers associated with each reference so that one would find in this case "100" associated with "cent", "20" associated with "twenty", "4" associated with "four" and "carriage return" associated with "input" (for example). According to the invention not only 100 4 can be expressed in form 124 but further the need for the use of an extra word to mean the end of the chain is avoided (i.e. equivalent of the carriage return by the keyboard is avoided) because all the determined macro-instructions contain a final signal CR which is deleted by the references after a certain time preset if

  although, when the final figure 4 is followed by a

  lence of sufficient duration, a user-defined carriage return character or other may be transmitted

  and can mean the end of the introduced chain.

  In addition or alternatively, the recognition apparatus may transmit height codes to the processor, a code indicating a decreasing height may then be used to indicate that a

  carriage return sign must be added to the final word.

  We now describe in more detail the characteristics

  the invention with reference to FIGS. 5 and 6.

  "MACRO-POSSIBILITIES IN MACRO-LANGUAGE"

INTRODUCTION

  So far, speech recognition devices have transmitted data depending only on the exact words said. This posed problems when expressions such as 121 were stated. Unless the text associated with the word "100" is "a hundred" and "20" is "twenty", the output signals lead to

  confusion and are difficult to use.

  In the case of the macro-language, a powerful set of instructions exists and ensures the subsequent processing

  recognized words before the transmission of useful data.

  This data can be in the form of text characters

  (including control characters) intended for

  host, voice messages or data intended to

to a 16-bit output channel.

  Other useful functions exist for the char-

  data from flexible disks and for examining the data transmitted from the host computer

to the display unit.

  REASONS FOR USING MACRO INSTRUCTIONS

  Smart decisions must be made when pairing a speech recognition device with external facilities (see Figure 5). In many applications, all that is needed is the textual version of what has been said. However, the control of external devices (such as lamps,

  engines, etc.) and vocal utterance may be required.

in other situations.

  The possibility of changing the syntax and vocabulary of the recognition device according to

  what has been stated is also very powerful because the size

  vocabulary is limited only by the ability to

moire available.

  A special programming language is required

  to perform these functions. Segments of

  program are called macro-instructions because they trans-

  effectively form what has been said in useful orders.

  The transmission of digital signals for things such as references must be specified for each expression 100, 101, 102, etc.,

  or a smarter pre-consultation is required

  for the decoding of "cent", "twenty", "and" and "one".

(See Appendix 2).

MACRO-INSTRUCTIONS

  There are actually lists of orders indicating how to use the macro-language and what data must be transmitted. In general, all characters are transmitted directly to the host computer, although there are some metacharacters used

for special orders.

  The metacharacters are, &, \ and; although the can be defined by the user since it is simply an order separator used between successive orders. These metacharacters can not be used

  directly as output text because they modify the inter-

  the text that follows them. When the operand of the metacharacter is invalid, the whole operand

and metacharacter is ignored.

"\"

  This is the character of literalization.

  It is used to validate the use of the metacharacter in the text transmitted to the computer hurriedly. This sign has no illegal operands, it just ensures that special meanings associated with

  following characters in macro-language.

  As a result, the transmission of & as a text requires \ &. Similarly, for

  other metacharacters, we must use \, \\ and \;

-, ,,

  This is used for calling a macro-instruction

  from another macro-instruction. This allows the one-time definition of common macro-instruction

to several references.

  This use is in the form: nnrx for example 100r4 nn being the macro-number to execute and x the number of times during which the macro-instruction must be executed successively. The use of: nn for example 72 is allowed and the defects to be executed prevent the execution

once of macro-nn.

  The special case: nnr0 has been reserved for: - undefined macro-nn repetition,

  or until recognition of a new reference.

  This is a head character used for

  the indication that the sequence is a macro-language order.

  The control functions can be divided

in six basic groups.

  i) Header ii) Parallel channel O / P iii) Voicemail O / P iv) Synchronization v) Recognition parameters vi) Other i &&&& File header string for transfer of end of file string data between & !! End of macro-language file string and floppies or other input devices & m = N The following characters determine macro-N &# Terminates a macro-definition ii) The parallel channel has its own processor so that the macro-instructions write actually programs for each channel bit. Of this

  way, a channel can operate asynchronously

  chrone with macro-instructions and so it is possible to let them run continuously without hung up recognition and processing

  additional macro-instruction.

There are 16 bits in the channel.

  Orders ordered directly from the macro instructions are as follows: & w = N writes an order list for the bit N &. finish an order list

  & d = H1H2H3H4 ceases to validate the execution of the pro-

  grams of bit for all the bits indi-

  indicated by the hexadecimal digits Hi .- .. H4, for example & d = 0005 indicates to stop the validation of bits 0 and 2 & e = H1H2H3H4 allows bit programs to be executed for all the bits indicated by the four hexadecimal digits & s = H1H2H3H4 synchronizes all the bits indicated by the four hexadecimal digits

  (that is, resets the

program).

  Channel bit control & + switches online channel & - switches off-service channel & t = T delay by milliseconds & s = H1H2H3H4 see before & d = H1H2H3H4 see before & e = H1H2H3H4 see before & e = H1H2H3H4 see before & w = N see before &. see before After ordering by the software, a single order

  bit is executed by passing through the loop at

  basic calf. It should be noted that & w ... &. is

classified as an order.

  NOTE: H1H2H3H4 represents a hexadecimal number with four characters. iii) & v = N voice message transmission N

  iv) & h = "cha ne" stops at this point in the macro-

  until the specified string of characters is received from the host computer (so that the synchronization with the programs

  allowed, for example when

  that the anterior character can pose

a problem).

  & h = oN stop at this point in the macro-

  instruction until the N parallel output channel bit has stopped executing its command list & h = -v stops until the end of all

voice messages.

* v) & x = 0 clears text directory

  & x = N (N = 0) creates a text input of

  prentissage for reference N & y = 0 clears synthaxe directory & y = N (N = 0) creates the syntax for node N & z = 0 clears reference directory & z = N (N = 0) creates reference N

&7th; termination character.

  vi) & ^ X Pass X to host computer & p = T Check for a T seconds pause in recognized references before continuing processing of macro instruction. When a pause T seconds or more is not found, end of the macro-instruction (useful for creating 102, etc., see appendix 2) & 1 = file name Load file "file name" from the floppy disk & t = T Delay of T milliseconds & n = N Check that the node for which the reference was recognized was N. If yes, continue processing the

  macro-instruction, if not interrupt it.

  & c = N Compare the identifier of the reference

  following reference to N. If it is N,

  follow the macro instruction and stop otherwise. (When a pause appears beyond the end of reference silence, an erroneous result appears and the macro instruction is complete). & f forget the following reference in the list of recognized references intended to be used with c when this next reference has

already been treated.

  & i Intercept data transmission between the computer, host and display unit, used with & h = "chatne", an order triggering the creation of the channel. & i must be passed before the trigger character so that

  the returned string is not lost.

  NOTE: When metacharacter orders are used, the preceding statement termination character

(default =;) can be deleted.

MACRO SPEECH INTERFACE

  When the macro-language is executed in a learning system of the person speaking, the selection of

  words or phrases to be recognized must be determined.

  Each unit of speech is known as a reference. A macro-number is associated with each reference

  so that, when a reference is recognized, the macro-

  numbered statement is executed.

APPENDIX 1

MACRO FUNCTION

  Literate nnrx Execute macro-nn x times nn Execute macro-nn once &&&& File header End of file &# General end (macro-instructions and recognition parameters) & + Go online output path * & - Set out of service exit lane * &. Finish Channel Command List & ^ X Command Character X

  & c = N Continue macro-instruction when identifying

  next reference = N & d = HiH2H3H4 Stop validation of output channel control list (s) indicated

  & e = H1H2H3H4 Valid list of output channel orders

  Failed & f Forget next reference in list of recognized references & h = "string" Stops macro-statement until specified string is received from host computer & h = oN Stops macro statement until stopping execution of output N & h = v Stop macro-instruction until all voicemail messages are transmitted & i Intercept host-display communications & nca file Load "file name" of the floppy disk & m = N The following characters determine N & n = N Continue the macro-instruction if the node to which the reference was recognized was N & p = T Continue the macro-instruction when a pause greater than T seconds exists between the current reference and the next one & s = H1H2H3H4 Synchronizes the indicated output channels & t = T Introduces a delay of T milliseconds & v = N Transmits a voice message N & w = N Writes an order list for the output channel bit N & x = NN = 0 clears text directory

  N <> 0 creates learning text to refer to

  N & y = N N = 0 clear syntax directory N <> 0 create synthax for node N & z = N N = 0 clear reference directory N <> 0 create reference N * available only in the channel order list

  H1H2H3H4 = 4-bit hexadecimal number.

APPENDIX 2

  Decoding "101" Word Cent One See Figure 6 The pause between words (or no pause) must

  be used for determining what has been said.

  In FIG. 6, the decision between 101 and 121 must be made according to the presence of "and" so that the following reference instruction & c = N. "& f" must also be compared when the following reference is processed in this current cycle.

for example

  Reference n word cent 20 twenty-one In this example, we assume the presence of a pause of

0.6 seconds to the limit.

  Macro-instruction 11, & p = 0.6; Print a 1 and if the person stops; &; to speak, the transmission of a carriage return Mcro-instruction 20 2; & p = 0, 6; 0 The decision is about twenty,; &# that is, if one must read 20 or 20

and something.

  Macro-instruction 100 & p = 0.6; 00 Additional characters are &#; & necessary if one hundred is the last part of the number Macro Instruction 50 105; 106; &#; Execute the macro-instructions 105 and 106 (once each) Macro-instruction 105 & c = 2; 02 If "one" was said after "and", one must &f;&;# transmit 01 and, as the reference "a "has been processed, the sign & f is used for the deletion of" one "from the list of recognized references Macro-instruction 106 & c = 20; &f; If "twenty" has been said, the decisions necessary for processing during the macro instruction are executed and called. It should be noted

that the identifier of the ref-

  "twenty" must be removed from the list of recognized references

(RTLIST) before calling the macro-

  instruction 20 because it has been fully processed at this time and macroinstruction 20 requires the data contained further

in the RTLIST list.

  These macro-instructions follow the following rules.

  Issue Macro-instruction one 1 two 2 twenty 20 one hundred two hundred 200 hundred one 101 one hundred two 102 one hundred twenty-one 121 one hundred twenty-two 122 two hundred two 202 two hundred twenty 220 two hundred and twenty-one 221 two hundred and twenty-two 222

Claims (12)

  Speech-controlled apparatus, characterized in that it is intended to be placed between the output of a manual keyboard (2) of the type which transmits digitally encoded signals characterizing keys when they are operated, and the input of a data processing apparatus (1) so that the whole forms   a speech recognition system, a   for comparing a digital signal derived from an audio signal received as a result of an expression or a spoken word, to several words   stored in a previously programmed memory.   mable of speech message data (20), the data words characterizing valid expressions or words, so that those of the words or expressions that have been stated are determined and signals are   transmitted to a translation device, these identi-   the expression or word so uttered, the translation device being controlled by the signals of the comparison device and being intended to determine, from   a second data memory (18) previously produced   grammatically, data characterizing several operations   equivalent keys on a connected keyboard, and to   transmission by a generating device   signals of a digitally encoded data signal string   this chain, which characterizes equiva- slow keys.   2. Apparatus according to claim 1, characterized in that, following an audio input signal formed of a pronounced string of valid words, a series of interdependent translations of valid words in the pronounced string determines an exit string from   data signals characterizing the pronounced string.   3. Apparatus according to claim 2, characterized   in that the interdependence of translations of valid words in pronounced strings is at least   partially determined by pauses of   terminated in the received audio signal.   4. Apparatus according to one of claims 2   and 3, characterized in that the interdependence of   The values of the valid words in the pronounced strings is at least partially determined by pitch variations of the received audio signal.   5. Apparatus according to any one of the claims   tions, characterized in that the chains   of transmitted data signals are determined in part   by the information received from the treatment device of data.   6. Apparatus according to any one of the claims   previous arrangements, characterized in that it comprises a switching device (7) and the apparatus, for a first state of the switching device, operates   in the aforementioned manner and, for the second state of the   switching device, allows direct communication between the connected data a connected keyboard.   Apparatus according to claim 6, characterized in that, when the switching device (7) is in its first state, a memory (17) is for storing data determining a representation of the characters transmitted by the apparatus so well. that when the switching device is switched to its second state, the apparatus may cause the display associated with a connected keyboard to display a proper record of the data exchanged with the connected device data processing.   Apparatus according to any of the claims   tions, characterized in that it comprises a   speech set (5 ') and a set of translat-   (5 ") which are electrically interconnected so that the translation assembly can be placed at a distance connected keyboard.   9. Apparatus according to any one of the claims   characterized in that the selection device is controlled by a predetermined string of alphanumeric and / or control characters from a connected keyboard so that operating modes of the apparatus in which the two data memories   can be pre-programmed or selected.   Apparatus according to claim 9, characterized in that the selection device is for preventing the transfer of any selection string.   received to a connected computer.   Apparatus according to any of the claims   prior art, characterized in that it is also controlled by stated alphanumeric characters, by command words or by such characters and words so that they transmit digitally encoded data signals, identifying each of the characters so stated , to the connected data processing device.   Apparatus according to claim 8, characterized   in that the translation set is part of an installation comprising a computer and which may constitute   connected device for data processing.