GB2149172A - Speech responsive apparatus - Google Patents

Abstract

Apparatus providing a speech recognition facility for a computer (not shown) is installed between a keyboard unit (2) and the computer. In the speech mode a digitised audio input from a microphone (6) is compared by a microprocessor (12) with data held in a speech parameter store (20) to determine the word or phrase spoken. The identity so determined is then used to obtain from a data store (18) a character string, previously entered from the keyboard unit (2), for transmission to the computer on an output (3'). A speech recognition facility may accordingly be provided for any computer without requiring amendment of the computer software. <IMAGE>

Description

SPECIFICATION Speech responsive apparatus The present invention relates to speech responsive apparatus and more particularly to such apparatus for providing a speech recognition facility for data handling apparatus.

Many kinds of speech recognition apparatus are known which may be trained to recognize alpha-numeric characters spoken by a particular user.

Such speech recognition apparatuses are not readily connectable to an existing computer installation which is designed for use with a keyboard input of for example strings of data characters to initiate the running of a particular program. Thus to enable such an installation to respond to simple voice input messages may require extensive software modifications. Such modification requires a considerable amount of time and much effort by a skilled computer programmer.

It is an object of the present invention to provide speech responsive apparatus which is adapted to provide a voice input system to a computer installation without requiring extensive modification of the computer software.

According to the present invention there is provided speech responsive apparatus for insertion between the output of a manually operable keyboard of the kind which provides digitally encoded signals characterising any one of a plurality of keys upon operation thereof and the input of data processing apparatus to provide a speech recognition facility therefor wherein there is comparison means to compare a digital signal derived from an audio input signal in respect of a spoken word or phrase with a plurality of data words held in a preprogrammable speech message data store, said data words characterising valid words or phrases, to determine which, if any, of the words or phrases has been spoken and to provide signals to translation means which signals identify the word or phrase so spoken, said translation means being responsive to such signals from the comparison means to determine from a second preprogrammable data store data characterising a plurality of equivalent key operations of a connected keyboard and to cause signalling means to output a string of digitally encoded data signals which string characterises the plurality of equivalent key operations.

The apparatus may be arranged such that, in response to an audio input signal consisting of a spoken string of valid words, a series of interdependent translations of the valid words in the spoken string determines a said output string of data signals characterising the spoken string.

The interdependence of the translations of valid words in the spoken strings may be at least partly determined by silent pauses of predetermined duration in the audio input. Alternatively or additionally the interdependence of the translations of valid words in the spoken strings may be at least partly determined by predetermined changes of pitch in the audio input.

The apparatus may be arranged such that said output strings of data signals are partly determined by information received from the data processing apparatus.

Switching means may be provided, the apparatus being responsive in a first condition of said switching means to operate in the manner described above and in a second condition of said switching means to permit direct communication between connected data handling apparatus and a connected keyboard.

The apparatus may include a memory arranged to store data defining a representation of the characters output by the apparatus when the switching means is in said first condition, such that when said switching means is switched to said second condition the apparatus may cause display means associated with a connected keyboard to display a correct record of the data transferred to or from connected data handling apparatus.

Speech responsive apparatus in accordance with the invention will now be described by way of example only with reference to the accompanying drawings of which: Figure 1 is a block schematic diagram of a typical computer installation with which the apparatus may be used; Figures 2 and 3 are respective block schematic diagrams of the computer installation of Fig. 1 adapted by use of two differing independent speech responsive apparatuses in accordance with the invention; Figure 4 is a block schematic diagram showing the speech responsive apparatus of Fig. 2 in greater detail; Figure 5 is another block diagram illustrating apparatus according to the invention; and Figure 6 is a flow chart illustrating a feature of the apparatus of Fig. 5.

Referring first to Fig. 1 the apparatus is for use with a typical computer installation 1 to which one or more man/machine interfaces, for example a visual display unit and keyboard 2 are connected. Whilst the visual display unit 2 is shown directly connected by a data bus 3, 3' via a plug and socket connection 4, 4' it will be appreciated that the connection 3, 3' could be via a land line to a remotely located computer or by way of a switched telephone link to any remote data bank capable of being so accessed.

Referring now to Fig. 2 the apparatus essentially comprises a signal conversion unit 5 which is connected in the line 3, 3' a microphone 6 and a speech input control switch 7.

The signal conversion unit 5 includes a mode selection switch 8 which is used to select the mode of operation of the unit as hereinafter described. To allow simple interconnection of the signal conversion unit a corresponding plug 9 and socket 10 arrangement is provided for respective interconnection with the socket 4' and plug 4 of the visual display unit 2 and computer installation 1.

In a preferred arrangement of the signal conversion unit 5, shown in Fig. 3 to which reference now is made, mode selection as hereinafter described is physically separated from the translation function of the unit so that a speech circuit console 5' may be positioned adjacent the visual display unit 2 and a translation unit 5" may be positioned remotely therefrom, data transfer being effected by means of a suitable interconnection 11. In this case the mode selection switch 8, the microphone 6 and the speech input control switch 7 are connected to the speech circuit console 5'.

Referring now to Fig. 4 and noting that the combination of the speech circuit console 5' and the translation unit 5" of Fig. 3 will function in a similar manner to the signal conversion unit 5 of Fig. 2 the function of the apparatus will now be described.

By use of the mode switch 8 located on the console 5' the user may select the mode of operation of the apparatus. The selected mode is detected by a microprocessor 12 on inputs 13 and in dependance on the mode selected the microprocessor follows one of six programmes. For convenience of description the six modes of operation of the apparatus 5 are referred to hereinafter as: (i) Transparent, (ii) Macro, (iii) Text, (iv) Train (v) Load and (vi) Save.

It is here noted that the mode switch 8 may be a series of keys on a keypad of the console.

Alternatively, the microprocessor 12 may be arranged to detect a string of alphanumeric characters and/or control characters which select the mode of signal conversion unit 5. In the latter case, the microprocessor 12 may be programmed to inhibit the transfer of the selection string to the computer installation.

In the transparent mode the microprocessor 12 causes an electronic switch 14 to effect a connection between the input thereof and one output which is connected to a lead 15. Thus signals output by the visual display unit 2 on the connection 3 are passed by way of an OR gate 16 to the connection 3' to the computer installation. It is here noted that if the visual display unit 2 is operated using simplex signalling-that is if the character transmitted is automatically displayed-characters received on the connection 3 are also stored in raster scan order in a screen store 17 the purpose of which will become apparent hereinafter. Functions 14 and 16 can be incorporated in the microprocessor system and are not necessarily physical switches or gates.

If the visual display unit 2 is operated using duplex signalling the signal reflected on the connection 3' is stored in the lead 3 for display in the usual manner. Data from the computer terminal regardless of the signalling mode is stored in the screen store 17 and transmitted on the lead 3 in a similar manner.

It will be realised that in the transparent mode the unit 5 has no apparent effect on data transfer between the visual display unit 2 and the computer installation. However, the screen store 17 is constantly updated to ensure that on transfer from any of the other operating modes to the transparent mode the microprocessor 1 2 may cause the screen of the visual display unit 2 to be rewritten with the correct, current data.

It is noted here that the transparent mode is also used for voice input purposes. Thus, if the speech input control switch 7 is operated whilst the unit 5 is in transparent mode a translation from voice input to data output is effected in a manner hereinafter described.

The macro, text and train modes are used to programme the unit 5 for use with the particular computer installation or data bank with which it is associated. In each of these three modes, the microprocessor 12 causes the electronic switch 14 to connect the input connection 3 to the microprocessor 12 by way of a connection 17. Received data is not output to the connection 3'.

The macro mode permits a user to enter data strings by use of the keyboard of the visual display unit 2 to be transmitted to the computer when selected by a voice input at a later time and the constraints for sending each such string. For example if it is necessary to transmit to the computer installation a string such as "1 0.0, 40.0, 1.0, 1 lE6, 'CR"' where 'CR' indicates carriage return, to define, say, default parameters the user inputs this string from the keyboard in the macro mode. The received strings are stored in a data store 18 by the microprocessor 12.

Having entered all such strings or any such strings as are required for the time being, which strings may include both alpha-numeric and/or control characters, the user switches to text mode.

In text mode the microprocessor 12 causes each of the strings to be transmitted from the data store 18 in turn to the visual display unit 2 which enables the user to check each of the strings and to enter from the keyboard the word or phrase to be used to identify the respective strings.

For the example quoted above, the microprocessor 12 would cause the string "10.0, 40.0, 1.0, 1 E6, 'CR"' to be displayed. In response to this display a user may key in the phrase "Default Parameters" to define the string.

Having completed the textual definition of the parameters the mode switch 8 is used to select the train mode. In this mode the microprocessor 12 causes the visual display unit 2 to display each of the previously entered phrases and the user operates the speech input switch 7 and speaks the phrase into the microphone 6. The input from the microphone 6 is converted to digital by an analogue to digital converter 19 and is stored by the microprocessor 12 in a phrase parameter store 20 together with an identifier for the appropriate string in the data store 18.

It will be realised that a new user of the apparatus may switch to the train mode and provide a replacement set of phrase parameters without re-entering new macro strings or text. Additional macro strings and related text may be entered by switching to the appropriate modes and entering the new data strings and existing data strings may be modified without changing the related text or train input.

It is noted that in text mode the microprocessor 12 will compare each text input with all previously stored text inputs to ensure that an identical title is not given to two differing data strings and will forward an appropriate message to the visual display unit 2 should this occur.

The load and save modes are provided so that existing speech parameter data and/or text and string data may be stored either in a backing store (not shown) of the unit 5 or, if available, in storage of the computer installation. Thus if there are several users of the apparatus, on completion of use the present user may switch to save mode and enter, say, save speech parameters Fred" or "save text and speech parameters Fred" such that on commencement of a subsequent terminal session by switching to load mode and entering "load Fred" from the keyboard of the visual display unit 2 the data may be recalled and the user can commence speech input immediately.

Having programmed the unit 5, the user returns the apparatus to transparent mode and the microprocessor 12 causes the screen store 17 to reconstruct the display (if any) on the visual display unit 2.

If now the speech input switch 7 is operated and the phrase "default Parameters" is spoken the microprocessor 12 provides comparison means and compares this input after conversion by the analogue-to-digital converter 19 with the speech parameters held in the store 20. If the spoken phrase is identifiable, the microprocessor 12 provides translation means and uses the associated identifier to recover the appropriate data string from the data store 18 and causes the data to be transmitted to the computer installation or data bank via the connection 3'. The transmitted data string will also be returned to the visual display unit 2.

If the spoken phrase cannot be identified by the microprocessor 12 the signal conversion unit 5 may be preprogrammed to transmit either a text string previously entered in the text mode or a null string by way of the connection 3'. Alternatively or additionally an "invalid speech" entry message may be transmitted by way of the connection 3 to the visual display unit 2.

If the character strings are output independently of characters received from the host computer, then in some circumstances voice commands would be accepted but the screen format would not be well structured. A facility is therefore included for character output to be dependent upon characters output from the host computer. Example of screen well formatted.

Prompt 1 type in then wait for prompt 2 before next part of macro string output otherwise input text may appear on screen before prompt and user will not see a pleasing picture nor a picture equivalent to keyboard input.

It will be appreciated that, apart from the phrase input facility the unit 5 may also provide all of the functions provided by known speech recognition apparatus. Thus if a user speaks, say, '7', '5', '1', 'RETURN', the output on the connection 3' will be "751 'CR"'. It will be noted that all of the speech recognition functions of the unit 5 are provided without requiring any reprogramming of the computer installation or data bank.

It will also be realised that the functions of the unit 5 may be performed by a combination of a translation unit and commercially available speech recognition apparatus and that some of the functions thereof could be incorporated in the computer installation.

Some further explanations of the features defined in the appended claims are given below.

The comparison means described in the appended claims may be explained in more detail as follows: At intervals of time, a short term digital representation of the audio input signal spectrum is calculated, each representation is referred to as a frame. The input is therefore converted to a sequence of frames fi' f' Where each frame details the short term spectrum of the audio input signal at that time.

This input sequence . . f,' . . f,' . . f' .

is compared with a set of "templates' which represent each of the valid words or phrases in the following form: Word Z(WZ) is represented as a sequence of frames: f1Z . . . . fzNZ The comparison means determines which sequence of words best explains the input frame sequence, the sequence that best explains the input frames, is say WXWY . . WZ Ref. Bridle and Brown J S Bridle and M D Brown "Connected word recognition using whole word templates", in Proc. Inst. Acoust. Autumn Conf., Nov. 1979, pp. 25 28 The provision by the comparison means of signals to the translation means as defined in the appended claims may be explained in more detail as below.

The information from the comparison means enables the recogniser processor to output the following information for each identified template (Pitch Code) optional Template Identification Score Start Time Stop Time where the latter may be alternatively encoded as either Start Time and Duration or Duration and Stop time Where the start and stop times are the times at which the algorithm determined the input utterance to contain the identified template and the optional pitch code encodes pitch contour information for the period determined.

As the time at which recognition information is output is indeterminate the recogniser processor also outputs silence as a recognised template at a defined rate so that the comparison means receives information about silence at the end of spoken phrases.

The manner in which signalling means outputs a string of characters as described in the appended claims may be explained in more detail as below.

Associated with each Template is a macro definition; this macro can itself call other macros and most importantly the macro can be dependent upon time.

This arrangement can provide a most important facility in that if one were to say one hundred" one would often wish ''1" ''0" "0" "carriage return" to be output, but that if one were to say "one hundred and twenty four" one would wish "1" "2" ''4'' "carriage return" to be output. This facility would be impossible with known systems which associate unique identifiers with each template, so that one would find in this case "100" associated with one hundred "20" associated with twenty "4" associated with four and "carriage return associated with "enter" (say).With this invention not only can 100 20 4 be expressed to 124, but additionally the need for an additional word to signify the end of the string (ie the equivalent of carriage return on a keyboard) is obviated as all the defined macros contain a final CR which is suppressed by templates following within a preset time, thus when the final 4 is followed by silence of sufficient duration a carriage return or other user defined character can be output to signify the end of input string.

Additionally or alternatively the recogniser can pass pitch codes to the processor, a code indicating falling pitch can then be used to indicate that a carriage return should be appended to the final word.

A more detailed implementation of features of the invention is disclosed in the following description with reference to Figs. 5 and 6.

"THE MACRO FACILITIES IN MA CROSPEA K" INTRODUCTION Up until now speech recognisers have output data dependent only upffn the exact words said.

This leads to problems where phrases like 122 are said. Unless the text associated with the word "100" are "One Hundred" and "20" is "Twenty" the output data is confusing and difficult to use.

With Macrospeak, a powerful instrucion set exists to post process the recognised words, and then to output useful data. This data may be in the form of text characters (including control characters) to the host, voice output messages or data to a 16 bit output port.

Other useful functions exist to load in data from floppy discs, and to examine data being sent from the host to the VDU.

REASONS FOR MACROS To interface a speech recogniser to external facilities (see Fig. 5), intelligent decisions have to be made. For many applications all that will be required is the textual version of what has been said. However, control of external devices (like lights, motors, etc) and voice output may be necessary for other situations.

The ability to change syntax and recogniser vocabulary demanding on what has been spoken is also very powerful, only limiting vocabulary size by the mass storage available.

To realise these functions a special programming language is necessary. The program segments are called Macros, as they effectively expand what has been said into useful commands.

To provide numerical output for things like 102, templates either have to be specified for each of 100, 101, 102, etc. or clever look ahead is necessary to decode "one", "hundred", "and" and "two (See Appendix 2).

MACROS These are effectively lists of commands to macrospeak to tell it how, and what data to output.

In general all charcters are output to the host computer directly, though there are some metacharacters used for special commands.

The metacharacters are @, 8, \ and though the semicolon is user defineable as it is just a command delimiter used between successive commands.

These metacharacters cannot be used directly as output text as they alter the interpretation of the text following them. If the operand of the metacharacter is invalid the whole of the operand and metacharacter is ignored.

"\" This is the "literalise" character. It is used to enable the metacharacter to be used in text output to the host. "\" has no illegal operands, it just swtiches off any special meanings that macrospeak associates with the following character(s).

Hence to output 8 as text \Et is needed. Similarly for the other meta characters \@, \\ and \; must be used.

This is used to call a macro from another macro. This enables those sections of macro output that are common to several templates, only to be defined once.

The usage is: @nnrx e.g. @100r4 nn is the macro number to be executed, and x is the number of times that the macro should successively be executed.

The usage: @nn e.g. @72 is allowed and defaults to execute macro nn once.

The special case: @nnrO has been reserved for:- repeat macro nn forever, or until a new template has been recognised.

This is a leading character used to indicate that the following is a command to macrospeak.

The command functions can be broken down into 6 basic groups (i) Headers (ii) Parallel port O/P (iii) Voice message O/P (iv) Synchronisation (v) Recogniser Parameters (vi) Other.

(i) @@@@ File header string ) for data transfer between macrospeak and floppy @!!! End of file string ) discs or other input devices am = N The characters following define macro N Terminate macro definition.

(ii) The parallel port has its own processor and so the macros actually write programs for each bit in the part. In this way a port may run asynchronously with the macros, and hence it is possible to leave them running forever , without hanging up the recognition and further macro processing.

There are 16 bits in the port.

Commands actioned directly from the Macros & = N Write a command list for bit N Terminate command list @d H1H2H3H4 Disable the running of the bit programs for all bits indicated by the 4 hex digits H1 . . .H4 e.g. & = 005 means disable bits 0 and 2 Be = B1 H2H3H4 Enable the running of the bit programs for all bits indicated by the 4 hex digits @s H1H2H3H4 Synchronise (i.e. reset program counters to their starts) of all bits indicated by the 4 hex digits Port bit program control @ + Switch on output channel @ - Switch off output channel @t =T Time delay of T milliseconds @s =H1H2H3H4 See above @d = H1H2H3H4 See above @e = H1H2H3H4 HlH2H3H4 See above @w = N See above See above When actioned by the software, only one command per bit is executed per base level loop pass.Note #w . . #. is classed as one command.

NOTE: H1H2H3H4 represents a 4 character Hex number.

(iii) Sv = N Output voice message N (iv) & = "string" halt at this point in the macro until the specified character string is received from the host (to enable synchronisation with host programs, e.g. where type ahead may be a problem).

Sh =oN halt at this point in the macro until parallel output channel bit N has stopped executing its command list.

Sh =v Halt until all voice messages have finished (v) & = 0 Clear text directory Sx =N (N = 0) create training text entry for template N Sy = 0 Clear syntax directory By = N (N = 0) create syntax for node N Sz = O Clear template directory Sz =N (N = 0) create template N Termination character (vi) @^X Send control X to the host @p = T Check for a pause of T seconds in the recognised templates before continuing with the processing of the macro. If a pause of T seconds or greater is not found, then end macro. (Useful for 102 creation etc.See App 2) 81 =filename Load file "filename" from floppy disc & = T time delay for T milliseconds an = N Check to see if the node at which the template was recognised was N. If so, continue processing the macro, otherwise end macro Sc =N Compare next template ident with N. If it is N, continue with the macro, otherwise end.

(If a pause occurs which exceeds the template end silence, a false result occurs and the macro is ended) & Forget next template in recognised template list-for use with Be where this following template has already been taken care of.

Intercept the data transmission between the host and VDU-used with Sh = "string" where a command triggers the string's creation. To avoid missing the returned string Si should be sent before the trigger character NOTE: When metacharacter commands are used, the preceding Statement Terminator charac ter (default = ;) may be omitted.

THE SPEECH MACRO INTERFACE As Mascrospeak is a speaker trained system, the selection of words or phrases to be recognised must be defined. Each speech unit is known as a template. Associated with each template is a macro number so that when a template is recognised, that numbered macro is executed.

APPENDIX 1 THE MACRO FUNCTION \ Literalise @nnrx Carry out macro nn x times @nn Carry out macro nn once @@@@ File header @!!! End of fle @#General termination (of macros and recogniser parameters) Switch on output channel & - Switch off output channel' Terminate port command list Character control X @c =N Continue macro if the next template ident = N @d = H1H2H3H4 Disable the output port command list(s) indicated @e = H1H2H3H4 Enable the output port command list(s) indicated @f Forget the next template in the recognised template list @h = "string" Halt macro until the specified string is received from the host @h = oN Halt macro until output channel N has stopped executing @h = v Halt macro until all voice messages have been sent Intercept the host to VDU transmissions 81 = filename Load "filename" from floppy disc #m =N The characters following define N #n =N Continue macro if the node at which the template was recognised was N #p =T Continue macro if a pause of greater than T seconds is found between the present template and the next @s = H1H2H3H4 Synchronise output channels indicated @t =T Make a time delay of T milliseconds @v =N Output voice message N @w = N Write a command list for output port bit N @x =N N = 0 Clear text directory N < > 0 0 Create training text for template N By = N N = 0 Clear Syntax directory N < > 0 Create Syntax for node N @z =N N = 0 Clear template directory N < > 0 Create template N only available in port command list H 1 H2H3H4 = 4 bit Hex number "102" decode Speech One hundred and two See Fig. 6 To decide what has been said, the pause between words (or lack of it) must be used.

In Fig. 6 the decision between 102 and 122 must be made at the "and" stage, and hence the need for the compare next template instruction Be = N. "@f" is also needed as the next template is taken care of in this present cycle.

e.g.

Template No. Word 100 hundred 50 and 20 twenty 2 two 1 one For this example a pause of 0.6 seconds is assumed as the limit.

Macro 1 1 ; & = 0.6; print out a 1 and if the person then stops speaking, a carriage return line feed aswell Macro 2 2;@p = 0.6; As macro 1 ;@# Macro 20 2;@p = 0.6;0 decision whether 20 or 22 has to be made, i.e. i.e.whether "twenty" means "2" or "2", "0" Macro 100 @p = 0.6;00 Additional characters are needed only if hundred hundredis the last part of the digit Macro 50 @105;@106;@# do macros 105 and 106 (each once) Macro 105 & = 2;;02 If "two" was said after the "and", it is necessary to output 02, As the "two" template has been dealt with, & is used to remove "two" from the recognised templates list Macro 106 @c = 20;@f;@20;@# If "twenty" was said the decisions necessary to process this are made under Macro 20 and so this is called. Note, the "twenty" template ident must be removed from the recognised template list (RTLIST) before calling macro 20 as it has been completely dealt with at this point and macro 20 needs the data held lower down in the RTLIST.

These Macros implement the following rules Utterance Macro Output one 1 two 2 twenty 20 one hundred 100 two hundred 200 one hundred and one 101 One hundred and two 102 one hundred and twenty one 121 one hundred and twenty two 122 two hundred and two 202 two hundred and twenty 220 two hundred and twenty one 221 two hundred and twenty two 222

Claims (13)

1. Speech responsive apparatus for insertion between the output of a manually operable keyboard of the kind which provides digitally encoded signals characterising any one of a plurality of keys upon operation thereof and the input of data processing apparatus to provide a speech recognition facility therefor, wherein there is comparison means to compare a digital signal derived from an audio input signal in respect of a spoken word or phrase with a plurality of data words held in a preprogrammable speech message data store, said data words characterising valid words or phrases, to determine which, if any, of the words or phrases has been spoken and to provide signals to translation means which signals identify the word or phrase so spoken, said translation means being responsive to such signals from the comparison means to determine from a second preprogrammable data store data characterising a plurality of equivalent key operations of a connected keyboard and to cause signalling means to output a string of digitally encoded data signals which string characterises the plurality of equivalent key operations.

2. Speech responsive apparatus as claimed in Claim 1 in which, in response to an audio input signal consisting of a spoken string of valid words, a series of interdependent translations of the valid words in the spoken string determines a said output string of data signals characterising the spoken string.

3. Speech responsive apparatus as claimed in Claim 2 in which the interdependence of the translations of valid words in spoken strings is at least partly determined by silent pauses of predetermined duration in the audio input.

4. Speech responsive apparatus as claimed in Claim 2 or Claim 3, in which the interdependence of the translations of valid words in spoken strings is at least partly determined by changes of pitch in the audio input.

5. Speech responsive apparatus as claimed in any preceding Claim, in which said output strings of data signals are partly determined by information received from the data processing apparatus.

6. Speech responsive apparatus as claimed in any preceding Claim, wherein switching means is provided, the apparatus being responsive in a first condition of said switching means to operate in the manner aforesaid and in a second condition of said switching means to permit direct communication between connected data handling apparatus and a connected keyboard.

7. Speech responsive apparatus according to Claim 6, wherein when the switching means is in said first condition a memory is arranged to store data defining a representation of the characters output by the apparatus such that when said switching means is switched to said second condition the apparatus may cause display means associated with a connected keyboard to display a correct record of the data transferred to or from connected data handling apparatus.

8. Speech responsive apparatus according to any preceding Claim, comprising a speech input unit and a translation unit which are electrically interconnected such that the translation unit may be remotely located with respect to a connected keyboard.

9. Speech responsive apparatus according to any preceding Claim wherein selection means is responsive to a predetermined string of alphanumeric and/or control characters from a connected keyboard to select modes of operation of the apparatus in which the two data stores may be preprogrammed.

10. Speech responsive apparatus according to Claim 9 wherein said selection means is arranged to inhibit the transfer of any selection string so received to a connected computer.

11. Speech responsive apparatus according to any preceding Claim wherein the apparatus is also responsive to spoken alphanumeric characters and/or control words to forward digitally encoded data signals identifying any such spoken character to connected data handling apparatus.

12. Speech responsive apparatus according to Claim 8 wherein said translation unit is a part of a computer installation capable of acting as connected data handling apparatus.

13. Speech responsive apparatus substantially as hereinbefore described with reference to the accompanying drawings.