FR2738382A1 - VOICE DIALOGUE SYSTEM FOR AUTOMATED INFORMATION PROVIDING - Google Patents

Abstract

The invention relates to a voice dialogue system for the automated provision of information, in particular a telephone number for a caller. The invention provides an alphabet recognition device (30) which can recognize a phrase spelled out by the user and which can select from among a plurality of predetermined phrases similarly spelled phrases. Expression recognition (50), which compares the expression entered by the user with the expressions selected by the alphabet recognition device (30) and which provides at least one expression based on a determination of similarity for supply to the user.

Description

Voice dialogue system for automated delivery

  The invention relates to a voice dialogue method intended for the automated supply of information, in particular a call number for a user, a voice dialogue installation intended for the implementation of the voice dialogue method as well a device for recognition, independent of the speaker, of speech, to be used in particular in an installation of

voice dialogue.

  There are known voice dialogue systems intended for the automated voice supply of telephone numbers, in which the dialogue between a caller, who wishes to obtain certain information concerning a

  phone number, and the system runs over the phone.

  Voice dialogue systems operating to date can, however, recognize only a fixed vocabulary, restricted by means of about 1000 words. Any texts and among them place names, names of

  family and phone numbers are provided through

  diary of a speech synthesizer. But it turned out that errors appear when pronouncing names, especially when the names do not obey the

  usual German pronunciation rules.

  The invention therefore aims to make available a voice dialogue method for the automated supply of information and to create a voice dialogue installation of a form suitable for this method, which can process a vocabulary to be recognized which is very wide, that is to say -to say about 10,000 to 100,000 words, which can nevertheless reach an acceptable recognition rate and reduce the number of errors when vocalizing foreign language terms or even which does not commit these errors at all. The invention solves this technical problem by the following operating steps: a) loading of orthographic-phonetic information by time intervals for a multiplicity of predetermined expressions from a dictionary which can operate online, the information being available in real time, b) voice prompt of the user to enter an expression, c) temporary storage of the expression entered, d) voice invitation of the user to spell the expression entered, e) in response to the spelled expression, recognition and selecting several of the predetermined and spelled reference expressions using the orthographic information stored in step a) on the basis of a determination of similarities, f) sending the expressions selected in step e) and the expression temporarily stored in an expression recognition device, g) recognition and selection of at least one expression from the expressions ons selected on the basis of a comparison of similarities and, h) sequential supply of the expressions found in step g) as well as other information

  associated in the form of synthesized speech.

  In an implementation of the method according to the invention, step h) is repeated until the user sets

  end of the supply in the form of synthesized speech of the expressions.

  In an implementation of the method according to the invention, steps e) and g) are ended after the expiration of a predetermined period of time and the user is invited to re-enter his expression, when the 'no expression was recognized. In an implementation of the method according to the invention, the user identifies one of the expressions pronounced by synthesis as being in agreement with his expression and, in response to this expression, a query of a database is triggered. electronic telephone directory capable of operating in real time in which all the data blocks satisfying the expression criterion are read and given to the user for selection, and the user can identify, on the basis of the name and address, data block, including number

  call must be provided by the facility.

  In an implementation of the method according to the invention, orthographic-phonetic information is loaded for the predetermined expressions at predetermined times from a dictionary that can

operate online.

  The invention solves the technical problem by installing a voice dialogue for implementing the method according to the invention, having the following particularities: a device intended for the input of an expression by a user, at least one synthesizer intended in producing voice signals for the user, a speech input device, an alphabet recognition device, which recognizes a phrase spelled by the user and which can select expressions of similar spelling from a multiplicity of predetermined and spelled reference expressions35, an expression recognition device, which compares the expression entered by the user with

  expressions selected by the recognition device

  birth of the alphabet, and which provides at least one expression based on a determination of similarities for the supply to the user, and at least one dictionary which can operate in

  line, which stores orthographic and

  phonetics for the multiplicity of predetermined expressions, to which the alphabet recognition device, the expression recognition device

  and the synthesizer can access in real time.

  In one embodiment of the installation according to the invention, it comprises an intermediate memory, which temporarily stores the expression entered by the user and a memory which receives the expressions preselected by the device.

recognition of the alphabet.

  In one embodiment of the installation according to the invention, the recognition device

  of expressions works in keyword recognition mode.

  In one embodiment of the installation according to the invention, the data stored in the dictionary is information relating to

  spelling, phonetics and intonation of predetermined expressions.

  In an embodiment of the installation according to the invention, additional information concerning homonyms is stored in the dictionary. In an embodiment of the installation according to the invention, the expression entered by the user can be a place name, a family name or several related words. In one embodiment of the installation according to the invention, there is provided a section of the dictionary which can operate online for the

2738382

  memorization of a general vocabulary, place names

and family names.

  In one embodiment of the installation according to the invention, a programmed microcomputer takes over the control of the installation. In one embodiment of the installation according to the invention, the expression recognition device and the alphabet recognition device are such that they can reject an expression entered by the user and / or invite the user to pronounce his expression again after a lapse of

prescribed time.

  The invention solves the technical problem by a device for speech recognition, independently of the speaker, intended in particular for use in a voice dialogue installation, comprising a device for recognizing the alphabet, which can recognize an expression spelled by a user and who can select, based on a determination of similarity, several spelled reference expressions from a multiplicity of predetermined spelled reference expressions, and an expression recognition device, which compares, based on a determination of similarities, an expression entered by the user and corresponding to the spelled expression, to the expressions preselected by the device for recognizing the alphabet and which provides as result at least one expression.30 In one embodiment of the device of speech recognition according to the invention, the device for recognizing speech 'expressions works in keyword recognition mode. In one embodiment of the speech recognition device according to the invention, it comprises a dictionary, which stores information

6 2738382

  orthographic and phonetic relating to the multiplicity of predetermined expressions, which can be accessed in real time by the alphabet recognition device and the expression recognition device, to determine expressions sounding similarly or

having a similar spelling.

  The invention is able to process a very large vocabulary with an acceptable recognition rate, by the fact that an expression entered by a user is subject to combined speech recognition. This expression can be a surname, a first name, a street name, a place name or also related words. The combined speech recognition includes an alphabet recognition device, which can recognize a user spelled expression and which can then select similar spelling expressions from a multiplicity of predetermined spelled reference expressions. The term "expression of similar spelling" is always used below when one seeks to express that two or more series of words composed of letters sound in a similar way when they are pronounced (for example "es e es es e el "and" ef e es es el "). The combined speech recognition includes, as a second main component, an expression recognition device, which compares the expression entered directly by the user with the reference expressions, which correspond to the reference expressions spelled and selected by the speech device. recognition of the alphabet. The expression recognition device provides, on the basis of a determination of similarity, as a result of the recognition, at least one word intended to be supplied to the user, this word corresponding to a reference expression similar to the user expression. An online dictionary is used to store orthographic-phonetic information for the multiplicity of predetermined expressions, to which the alphabet recognition device, the expression recognition device and a synthesizer can access in real time. An intermediate memory is suitably provided, which temporarily stores the expression entered directly by the user, before it

  does not reach the expression recognition device.

  In addition, the installation includes another memory, in which the spelled reference expressions preselected by the alphabet recognition device are loaded in the form of a list of candidates for names.

with similar spelling.

  The expression recognition device operates in keyword recognition mode, so that the user can pronounce within certain limits additional expressions before and after the expression itself and that the expression recognition device is despite that able

  extract the expression itself.

  The orthographic-phonetic information stored in the dictionary relates to the way of writing the predetermined expressions that the alphabet recognition device uses to recognize a spelled expression and to carry out a preselection of similar spelling names for the recognition device. 'expressions. In addition, it is stored in the dictionary of phonetic transcriptions, for example for place names and family names. Spelling and phonetic transcription of proper names are transmitted in an offline process from an electronic pronunciation lexicon to the dictionary. In this case, only proper names are transmitted which appear in the electronic telephone directory. The electronic telephone directory is a database which can operate in real time, which includes the addresses and telephone numbers necessary to provide information to the user. To obtain a good quality, even for the pronunciation of names which do not obey the usual rules of pronunciation in German, it is memorized, in addition to the phonetic information, also information relating to the intonation of the terms. These vocal characteristics reproduce the intonation of syllables and endings

  also of foreign language words.

  To prevent the results of combined speech recognition from being influenced randomly due to acoustic similarities between words and / or spoken letters, additional information for homonyms is stored in the dictionary. This additional information makes it possible to supplement a candidate obtained by speech recognition with alternatives pronounced in the same way and thus to increase the recognition rate of

the installation.

  A dictionary section is appropriately provided for general vocabulary, for city names and for surnames that exist in these cities.25 Control of the voice dialogue installation is carried out by a microcomputer program. The control software which is implemented in this microcomputer ensures inter alia that the necessary orthographic and phonetic information coming from the dictionary are made available to the recognition devices and the synthesizer in time and that the installation

  prompts a user to enter expressions by speaking. In addition, it monitors the timers running in the speech recognition devices, processes end and help instructions, and supports recognition and error handling.

  In the expression recognition device and in the alphabet recognition device, internal program loops take place which reject an expression entered by the user and which can invite the user after a period of time. prescribed

  has passed, to re-enter its expression.

  The invention is explained in more detail below on the basis of an embodiment in conjunction with the accompanying drawings, which show: Figure 1 a block diagram of a voice dialogue installation comprising the combined recognition according to invention of speech and having an online dictionary, Figure 2 a flowchart for the progress of an automated voice dialogue of name recognition and provision of an associated call number, which is implemented by the installation of voice dialogue according to

figure 1.

  FIG. 1 shows the basic structure of a voice dialogue installation, which can carry out dictionary-controlled recognition of any expressions, for example place names or family names, by a combination of devices of speech recognition and which can provide, on the basis of a determined expression (result of recognition), information associated with the expression (for example a call number). FIG. 1 shows in detail a telephone apparatus, in which a caller can pronounce the name of the place or the surname of a subscriber, from whom he seeks to obtain the telephone number, or other determined expressions. At the operator side of the voice dialogue installation, at least one analog-to-digital 35 converter 80 is provided, which converts the subscriber's analog voice signals into digital signals. The converter output

2738382

  analog-digital can be connected to the input of a speech memory 20 and to the input of a device 30 for recognizing the alphabet or letters. The speech memory 20 is used to temporarily store the expression entered directly by the person calling into the telephone apparatus 10, for example the name "Meier", for later use. The device 30 for recognizing the alphabet receives, via the analog-to-digital converter 80, as a function of the state of the unfolding of the voice dialogue, a spelled version of the expression previously stored in the

  speech memory 20 and input directly. A mic-

  computer 120 programmed ensures that the expression entered directly is loaded into the speech memory 20 and that the spelled expression is sent to the device 30 for recognizing the alphabet. The output of the alphabet recognition device 30 is connected to a memory in which is stored a list of candidate expressions of similar spelling, which were determined by the alphabet recognition device 30 during a preselection. An expression recognition device 50 has three inputs, which are respectively connected to the output of the candidate memory 40, to the speech memory 20 and to an online dictionary 70. The expression recognition device 50 operates in what is called the keyword recognition mode, which still makes it possible to extract the true expression, for example "Meier", even when additional expressions such as "uh" , "please" or similar expressions precede or follow the true expression. The output of the expression recognition device 50 is connected to a recognition result memory 55, in which the expressions of the result, which are names of similar consonance, are stored by the expression recognition device 50. The L! expressions stored in the memory 55 of the recognition results are sent to a synthesizer 60, which transmits, on the basis of the corresponding information from the dictionary, in order, the names in the synthesized language via a digital converter - analog 85 to the subscriber's telephone 10. The synthesizer 60 can also produce the voice invitations made to the caller in connection with a database, not shown, which contains all the texts to be announced by

  installation in orthographic or phonetic form.

  The online dictionary 70 mentioned above is characterized above all by the fact that it can be used at the same time and in real time by the device 30 for recognizing the alphabet for the recognition of letters, by the device 50 for recognizing of expressions and by the synthesizer 60. This is why all the information concerning the expressions to be recognized or to be pronounced by the installation is stored in this dictionary 70. This is in this case information relating to the spelling, pronunciation and intonation, which are loaded in an offline process from a 100 pronunciation lexicon in the 70 dictionary online. In addition, information relating to homonyms is stored in the dictionary 70, to supplement the result of the recognition of the expression recognition device with names sounding in the same way or to supplement the reference expressions spelled out from the recognition device. of the alphabet by names with similar spelling and thus increase the probability of detecting the correct expression. This also ensures an increased success rate when using or improving the overall throughput of the installation, since the expressions to be recognized are more rarely rejected by

12 2738382

  speech recognition devices 30, 50. Information about homonyms allows the expression recognition device, for example

  for an expression "Meier", to find all the ortho-

  graphs present in the electronic telephone directory, such as "Meier", "Mayer", "Maier" and "Meyer" and put them in the list of recognition results. On the other hand, this allows the device for recognizing the alphabet, to reproduce the variants of letters for example appearing often and which can be used wrongly, such as for example "MULLER" or "MUELLER", by the correct reference expression spelled, even when, for example, it appears in the electronic telephone directory only with the spelling having "". The online dictionary 70 described therefore helps for the first time both to recognize the

  speech and speech synthesis.

  The operating mode of the installation of

  voice dialogue is explained more explicitly below

  after on the basis of the recognition of a name. It is assumed that the voice dialogue installation already knows the name of the place where the person lives and from whom a caller would like to obtain the telephone number. To this end,

  the installation first invited the user of the device

  telephone number 10, to be entered directly, that is to say in non-spelled form, the name of the place (for example Darmstadt). Suitably, the microcomputer 120 controls the installation so that the place name is sent only to the expression recognition device 50 to recognize the expression. As already mentioned, the expression recognition system is able to tolerate additional expressions like "uh" or "please" and to extract as information only the name of the city.35 The voice dialogue system can also be formed so that a pre-selection of the names of

13 2738382

  places of similar spelling by the device 30 of recognition of the alphabet for the device 50 of recognition of expressions, when it was provided by the device 50 of recognition of expressions a result of false recognition or even no result of recognition at all. After the name of the city has been recognized, the voice dialogue installation supplies all the surnames stored in an electronic telephone directory 90 for these city names from the online dictionary 70. It is further assumed that the dictionary online 70 contains the spelling of all proper names, which are necessary for the recognition of letters in the alphabet recognition device 30, a series of sound symbols for all proper names, which are necessary for the recognition of speech in the expression recognition device, as well as a series of sound symbols including intonation information necessary for speech synthesis. In addition, the electronic telephone directory 90 which contains the surnames of the subscribers with telephone numbers and corresponding addresses, includes references to the entries.

  corresponding in the online dictionary.

  The caller is now guided by a dialogue, during which he learns, on the basis of the indication of the place name and the name of the subscriber, the

desired phone number.

  The following voice dialogue between the caller of the telephone apparatus 10 and the installation of

  voice dialogue is explained in the flowchart according to figure 2.

  The caller is first invited by voice by the installation, via the synthesizer, to directly enter the desired name, for example "Meier". Then, what has been entered in the speech memory 20 is temporarily stored. We record in

14 2738382

  the occurrence also of additional expressions like "uh" and "please" in speech memory. Then the caller is invited by voice, through the synthesizer 60, to spell the name entered directly before. Then, the subscriber entered the series of letters M, E, I, E, R. In conjunction with the orthographic information which is recorded in the

  dictionary 70 online, the recognition device 30

  sance of the alphabet makes a determination of similarities and makes a preselection in the list of available surnames stored in the online dictionary under the name of the place. Due to recognition uncertainties, the device 30 for recognizing the alphabet determines several candidates, for example "Neier", "Meier", "Meter", "Mieter", "Neter", "Nieter", "Meiter" , "Meider", etc. This list of candidates found in memory 40 is recorded. The programmed microcomputer 120 causes the expression recognition device 50 to read the expression "Meier" of the user temporarily stored beforehand in the speech memory 20 and to load the preselected candidates found in the memory 40. On the basis of a determination of similarities, the expression recognition device 50 compares the expression of the name entered directly "Meier" to the list of candidates, thereby that it uses the phonetic information stored in the online dictionary 70. The expression recognition device 50 supplies as a result of the recognition, for example, the names "Neier" and "Meier" and stores them in the memory 55 of results. The vocal dialogue installation knows, thanks to the information relating to phonetics and intonation which are stored in the online dictionary 70, how to pronounce the results35 of the recognition found and the intonation to be given to them. . Then we successively transmit the names

2738382

  found, in this case the names "Neier" and "Meier", through the synthesizer 60 to the telephone apparatus 10 of the calling person. The caller can then select the correct name. This family name and the recognized place name then trigger an interrogation of a database of the electronic telephone directory 90. The names and addresses found are read aloud under the command of the user, ie that is, the user can influence the time when the supply of the names and addresses found is stopped and the number of times a list is read aloud or the name for which additional information must be provided. In the cases posing problem, one can envisage the possibility of connecting the person calling with an operator. As soon as the user of the voice dialogue installation indicates that the data supplied via the speech synthesizer 60 (first name, last name, street, house number) correspond to the data of the person

  whose phone number, micro-

  computer 120 causes the installation to read the corresponding telephone number in the telephone directory 90 and to communicate it in voice form to the calling person. Thanks to the dictionary-controlled recognition of any expressions by the combination of the alphabet recognition device and the expression recognition device 50, it is possible to process, with an acceptable recognition rate, a vocabulary markedly more extensive than installations. common ones that only use a speech recognition device. The reason for this resides in the fact that the device 30 for recognizing the alphabet pre-selects the words to be recognized and that it is only sent to the device 50 for recognizing expressions for the actual recognition that this selection

  relatively small of relevant words.

16 2738382

Claims (17)

  1. Method of voice dialogue for the automated supply of information, in particular of a telephone number for a user, comprising the following steps: a) loading by time intervals of orthographic-phonetic information for a multiplicity of expressions predetermined from a dictionary (70) which can operate online, the information being available in real time, b) voice invitation of the user to enter an expression, c) temporary storage of the expression entered, d) voice invitation of the user user to spell the entered expression, e) in response to the spelled expression, recognizing and selecting more than one of the predetermined and spelled reference expressions using the orthographic information stored in step a) based on a determination of similarities, f) sending the expressions selected in step e) and the expression temporarily stored to a recognition device (50) of expression, g) recognition and selection of at least one expression from the expressions selected on the basis of a comparison of similarities and, h) sequential supply of the expressions found in step g) as well as other information 17 2738382   associated in the form of synthesized speech.   2. Voice dialogue method according to claim 1, characterized in that step h) is repeated until the user ends the supply of synthesized speech expressions. 3. Voice dialogue method according to claim 1 or 2, characterized in that steps e) and g) are terminated after the expiration of a predetermined period of time and that the user is invited to re-enter its expression, when no expression has been recognized. 4. Method of voice dialogue according to claim 2 or 3, characterized in that the user identifies one of the expressions pronounced by synthesis as being consistent with his expression and, that in response to this expression, a query is triggered. a database of an electronic telephone directory (90) capable of operating in real time in which all the data blocks satisfying the expression criterion are read and given to the user for selection, and the user can identify, on the basis of the name and address, the data block, including the number   call must be provided by the facility.   5. Method of voice dialogue according to one of the   claims 1 to 4, characterized in that one charges   orthophonic-phonetic information for predetermined expressions at predetermined times from a dictionary (70) operable online.   6. Installation of voice dialogue for the implementation of the method according to one of claims 1 to 5,   having the following features: a device intended for the input (10) of an expression by a user, at least one synthesizer (60) intended for the production of voice signals for the user, 18 2738382   a speech input device (110), characterized by an alphabet recognition device (30), which recognizes a phrase spelled by the user and which can select expressions of similar spelling from a multiplicity of expressions of predetermined and spelled references, an expression recognition device (50) which compares the expression entered by the user with   expressions selected by the recognition device (30)   birth of the alphabet, and which provides at least one expression based on a determination of similarities for the supply to the user, and at least one dictionary (70) which can operate online, which stores orthographic-phonetic information for the multiplicity of expressions   to which the alphabet recognition device (30), the expression recognition device (50) and the synthesizer (60) 20 can access in real time.   7. Installation of voice dialogue according to claim 6, characterized by a memory   intermediary (20), which temporarily stores the expression entered by the user and by a memory (40) which receives the expressions preselected by the device (30) for recognizing the alphabet.   8. Voice dialogue installation according to claim 6 or 7, characterized in that the expression recognition device (50) operates in the keyword recognition.   9. Installation of voice dialogue according to one of claims 6 to 8, characterized in that the data   stored in the dictionary (70) is information relating to spelling, phonetics and intonation of predetermined expressions.   10. Installation of voice dialogue according to claim 9, characterized in that it is stored in the dictionary (70) of additional information concerning homonyms.   11. Installation of voice dialogue according to one of the   claims 6 to 10, characterized in that the expression   user input can be a place name, a name   of family or several words relating to it.   12. Installation of voice dialogue according to one of the   claims 6 to 11, characterized in that it is provided   a section of the dictionary that can function online for memorizing a general vocabulary, names of places and surnames.   13. Installation of voice dialogue according to one of the   claims 6 to 12, characterized in that a micro-   computer (120) programmed supports the control of the installation. 14. Installation of voice dialogue according to one of the   claims 6 to 13, characterized in that the   expression recognition device (50) and the alphabet recognition device (30) are such that they can reject an expression entered by the user and / or invite the user to pronounce his expression again after the flow of a lapse of prescribed time.   15. Device for speech recognition, independently of the speaker, intended in particular to be used in a voice dialogue installation according to   one of claims 6 to 14, comprising   an alphabet recognition device (30), which can recognize an expression spelled by a user and which can select, based on a determination of similarity, several reference expressions spelled out of a multiplicity of reference expressions spelled and an expression recognition device (50) which compares, based on a determination of similarities, an expression entered by the user and   corresponding to the spelled expression, to the expressions preselected by the device (30) for recognizing the alphabet and which results in at least one expression.   16. Device for speech recognition according to claim 5, characterized in that the device (50) for expression recognition operates in keyword recognition mode. 17. Device for speech recognition according to claim 15 or 16, characterized by a dictionary (70), which stores orthographic and phonetic information relating to the multiplicity of predetermined expressions, to which the device (30) can access in real time. for recognizing the alphabet and the recognition device (50)   expressions, to determine expressions that sound similarly or have similar spelling.