JP2006236037A - Voice interaction content creation method, device, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To simplify a creation method for various kinds of contents used in a voice interaction device. <P>SOLUTION: In this voice interaction content creation method, a scenario template decided with display order of images and order of interaction fit for an application to be developed, a plurality of slot names each expressing an input item name decided by the application to be developed and slot IDs attached to the slot names, and templates of a plurality of interaction sentences each previously embedded with the slot ID, each prompting input to an interaction opposite party by substituting the slot name according to an embedded position of the slot ID, and each leading the proper input are prepared. The slot name is substituted according to the slot ID embedded in the template of the interaction sentence, each interaction sentence embedded with the slot name is substituted for the scenario template, and a scenario file and a prompt voice file are generated. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention recognizes a user's voice, understands its contents, performs necessary processing, displays the result on the voice and the screen, and returns a response to the voice dialog system used for developing a program for a voice dialog system The present invention relates to a creation method, an apparatus, a program, and a recording medium on which the program is recorded.

Display an anthropomorphic animated agent character on the screen, recognize the voice spoken as the user speaks to the agent, play the response with pre-recorded voice or synthesized sound, or on the screen There has been proposed an apparatus that advances a dialogue while notifying by displaying it in a text and outputting it as a word, or displaying an nuance of a response as an action of an agent in an animation (Patent Document 1).
In creating a voice dialogue system, it is necessary to describe in detail the operations of various parts such as voice recognition, voice synthesis, voice reproduction, voice understanding, dialogue understanding, and dialogue control.

Usually, in the detailed description work (programming) of this operation, in addition to the design information of the spoken dialogue system to be created, the technical characteristics of technical elements such as speech recognition and speech synthesis used in each part are grasped. It was necessary to have specialized knowledge. In addition, it is necessary to describe a process of linking the parts such as reproducing the voice while recognizing the voice. Because of these requirements, the design and description of a spoken dialogue system has become a very complex task.
Due to this complexity, parts that are often used in a voice dialogue scenario have been conventionally used in a spoken dialogue system program (hereinafter referred to as a voice dialogue scenario) description as shown in (Patent Document 2). The voice dialogue scenario is designed and described in a simplified manner by connecting them in order according to the order of dialogue (dialogue flow).

In addition, in order to perform speech recognition, it is necessary to create a grammar, a language model, a dictionary, and the like for receiving speech uttered by the user and outputting it as a recognition result. In this process (Patent Document 3 and Patent Document 4), a language model with high coverage called a class statistical language model and a dictionary are created from a small number of corpora (computerized speech / language data), and speech recognition is performed. Techniques to perform have been proposed.
In recent years, not only voice modals (voice dialogue format) such as voice output and voice input, but also response contents for voice input are drawn on the screen, and a mouse or the like among items such as links on the drawn screen. A multi-modal system has been proposed that responds by voice to the item selected in (1). In such a system, in the middle of a voice conversation, the status of each item, such as “entered” or “not entered”, for the items necessary for the system to return search results etc. It is possible to provide an easy-to-understand dialogue system for users by expressing the results of dialogues and dialogues on the screen. For this purpose, document files such as images and hypertext are also prepared. At this time, by receiving information about items that have already been input, by creating a device that dynamically generates the above document that displays candidates for items that have not been input (slots), A method for reducing the cost of document creation has also been proposed.
JP 2004-295837 A JP 2004-310628 A JP 2004-69858 A JP 2004-053745 A

  Regarding the development of a voice conversation scenario, as described in (Patent Document 2), the burden of development can be reduced by taking a technique of connecting partial voice conversation scenarios. However, in the flow design of the voice conversation scenario, very high expertise is required, and it is difficult to mass-produce many voice conversation scenarios in a short time. This is because, in general, in what order the input items necessary for the output of the results are input in voice dialogue, and what language expression (prompt) is given as a response from the system to guide the user according to the procedure. This is because a careful design is necessary for the use.

  In addition, after speech recognition, it is necessary to understand utterances (processing to extract keywords to be substituted into items (slots) that need to be heard from the user in order to achieve the purpose of dialogue) However, when registering a word as a keyword in the dictionary at the stage of creating a class language model, (1) notation and (2) reading of the language and assigning the word to which class symbol in the class language model In addition to (3) information (class information), it is necessary to add (4) an assignment destination slot that represents meaning in the speech understanding process. That is, it is necessary to register four types of information, which is complicated.

  In addition, when preparing a device for generating a document for dynamically displaying a screen during a dialogue or at a stage of providing information from a result of the dialogue, another device other than the dialogue processing device (for example, a Web) is usually used. It is necessary to create a CGI program on the server. Creating a program other than a voice conversation scenario is more difficult for a voice conversation scenario creator who is inexperienced than creating a simple document. However, instead of dynamically generating a document, statically preparing a document corresponding to all input states has been a very troublesome task.

As described above, there are the following problems as problems of the multimodal voice conversation scenario, and it has been a problem to reduce the work cost and the difficulty level thereof.
(1) Determining the flow (scenario) of the voice dialogue scenario itself is also difficult because it requires specialized knowledge. (2) It is difficult to write response (prompt) content according to the state during the dialogue ( 3) It takes a lot of work to register a word as a keyword. (4) It is difficult to create a document or program for displaying the result on the screen.

The present invention has been made in view of the above problems.
With regard to (1) above, partial voice conversation scenarios are made into parts, and the developer connects the parts while considering the flow according to the domain (topic) such as applications such as search and online shopping, transportation, retail, and securities. Rather than, for example, an application that asks for input about two items and returns results derived from the two information (displays stores from "location""industry", between "departure station""destinationstation" between By preparing the entire flow (such as displaying route transfer guidance information) as a template, the flow design becomes unnecessary. At this time, it prepares text (displayed in the agent spout and as a telop on the screen) and response voice according to the domain of the application to be created.

  Regarding (2) above, which is a problem when preparing a response tailored to the above domain, the situation in each scene in the scenario flow prepared above (one slot has already been entered and the other slot is still entered) In response to the request text, a template (template) with a response text in which a slot name portion is left blank is prepared. Then, by specifying the name of the slot, the name is embedded in the blank in the template, the prompt sentence is dynamically generated, the prompt sentence is embedded in the scenario template, and if necessary, a speech synthesizer or human Is provided with means for creating a sound file by recording the sound of the text read out by a recording device.

Regarding the problem (3), there is provided a slot / class replacement means for replacing the name of the slot into which the keyword is input as a class as it is.
With respect to the problem (4) above, means for automatically creating a list enumerating all combinations of keywords (including combinations in which a keyword is not included in a slot) in a slot in the voice interaction application to be created Is provided. For a combination that does not contain a keyword in the slot, a document file to be presented to the user can be displayed by displaying a list of keywords that are already in the slot and keywords that are not yet in the slot. A means for automatically creating is provided. The file name of the created document is added to the list in association with the corresponding keyword combination.

  Further, in an environment where a dialogue scenario is executed, that is, in a voice dialogue device, the dialogue scenario is executed, and each time there is an input from the user, the keyword information inputted in the slot is received, and the correspondence between the keyword and the document is listed. A result display executing means and a screen display means are provided for extracting the path and file name of the document associated with the combination of keywords input with reference to the above, and reading and displaying the file.

The following effects can be obtained by the present invention.
1. Since the developer of the development application itself does not need to develop the flow of a voice conversation scenario that requires specialized knowledge of the voice conversation scenario, even a non-specialist of a voice conversation application with an appropriate flow can easily perform the voice conversation. Can develop content.
2. As above, in the composition of the system response prompt content that requires expert knowledge of the voice interaction scenario,
Since there is no need to create a composition for each development application from the beginning, it is automatically generated by a template, which facilitates the development of spoken dialogue content by non-experts who perform appropriate dialogue.
3. Since the intermediate result document file is automatically generated, it is possible to reduce the amount of development work for multimodal speech interactive content.
4). Since it is no longer necessary to assign class information to keywords, which is necessary for creating a class language model, the amount of development work for spoken dialogue content can be reduced.
5. By simply preparing a list of slot names and a list of word information, it is possible to create semi-automatically or entirely of all the elements necessary to execute multimodal spoken dialogue content, so the development work is less than before. Much more.

The voice interaction content creation apparatus according to the present invention can be configured entirely by hardware. However, in order to realize it more easily than that, the speech conversation content creation program proposed in the present invention is installed in a computer and is provided in the computer. An embodiment in which a program is decrypted by a central processing unit (CPU) and the computer functions as a voice interactive content creation device is the best mode.
When a computer functions as a voice dialogue content creation device according to the present invention, the computer should develop a scenario template that stores a scenario template that defines a dialogue order and an image display order suitable for the application to be developed. A slot name list storage unit storing a plurality of slot names representing input items determined by an application and slot IDs attached to these slot names, and slot IDs are embedded in advance, and the slot names are set according to the embedded positions of the slot IDs. Prompt input to the conversation partner by substituting, the prompt template storing a plurality of dialog sentence templates that guide the appropriate input, and the slot according to the slot ID embedded in the dialog sentence model stored in the prompt template Assign a name and prompt Prompt text creation means for generating a sentence list, and each dialogue sentence in which the slot name is embedded are substituted into the above scenario template, and scenario generation means for generating a scenario file and each dialogue sentence in which the slot name is embedded A prompt voice generation unit that generates a prompt voice file by substituting into the prompt sentence list is constructed and functions as a voice dialog content creation device.

  In the present invention, the computer functioning as the voice interactive content creation device further includes an intermediate result document installation position information storage unit, a word information list storage, and an intermediate result document installation stored in the intermediate result document installation position information storage unit. Result display document list generating means for generating a result display document list for displaying a screen showing the result of the dialogue from the position information and the word list stored in the word list storage unit, and a slot stored in the slot name list storage unit An intermediate result document file generating means for generating an intermediate result document file for expressing a situation in the middle of the dialogue from the name list and the word information list is constructed to function as a voice interactive content generation device.

FIG. 1 shows an embodiment of a voice interaction content creation apparatus according to the present invention. In the figure, reference numeral 100 denotes a voice interactive content creation apparatus according to the present invention. Input information 10 prepared in advance is input to the voice interaction content creation apparatus 100, and the voice interaction content 20 is generated based on the input information 10.
Here, the input information 10 indicates a case where the slot name list 11, the intermediate result document installation position information 12, the word information list 13, the additional example sentence corpus file 14 and the like are used. In addition, the generated voice dialogue content 20 includes a scenario file 21, a prompt voice file 22, a result display document list 23, an intermediate result document file 24, a keyword list 25, a class language model 26, a recognition dictionary 27, and the like. Show.

The spoken dialogue content creation apparatus 100 includes a scenario template group 101 that stores a large number of scenario template templates prepared with various scenarios, a prompt template group 102 that stores a large number of dialogue sentence templates, a scenario generation unit 103, and a prompt text generation unit 104. , Prompt voice generation means 105, result display document list generation means 106, intermediate result document file generation means 107, slot / class replacement means 108, class language model and recognition dictionary generation means 109, example sentence corpus 110, and the like.
When a voice interaction application is developed using this embodiment, first, a scenario template (model of a voice interaction scenario) according to the type of application (hereinafter referred to as a development application) from the scenario template group 101 to be developed ( Figure 2: Scenario template will be explained later). Hereinafter, in the present embodiment, an example will be described in which an application is created in which two items of “league” and “score” are input by voice as professional baseball score information, and a screen corresponding to the input content is displayed. It is assumed that keywords such as “pa-league” and “se-league” can be entered for the league, and “ranking” and “batting rate” can be entered for the results. In this case, the scenario template has two slots (embedding positions of items to be input by dialogue), and a scenario template in which a flow for inputting these two slots is described is selected.

A name is determined for each slot handled by the selected scenario template, and a slot name list 11 (FIG. 1) is prepared. The name decided here is used as a language expression when responding when the development application asks the user for input during the conversation. An example of the slot name list 11 is shown in FIG. 11A indicates a slot name, and 11B indicates a slot ID. Each slot name 11A is described in pairs with a slot ID 11B which is an ID symbol assigned to identify the slot.
Next, the slot name list 11 is input to the prompt text generating means 104, and a prompt sentence list 104A (FIG. 1) is created in which prompt texts that are system response sentences in each scene of the dialogue are listed. In order to generate the prompt sentence list, the prompt text generation means 104 uses the prompt template group 102 corresponding to this input and the scenario template group 101.

An example of a prompt sentence list stored in the prompt template group 102 is shown in FIG. In the prompt template group 102, templates of response sentences when the system responds in each scene during dialogue are listed as text. One sentence and one sentence listed are called prompt template text.
The part in which the slot name is to be reproduced in each template text is left blank, and slot IDs indicated by <condition 1> and <condition 2> are given.
Each prompt template text is associated with a prompt ID in order to indicate a correspondence with a part to be inserted into the scenario template group 101.

The processing procedure of the prompt text generation means 104 is shown in FIG. Executes reading of step SP6 ₁ in the slot name list shown in FIG. 6, step SP6 and executes reading of ₂ at the prompt template, and insert the name of the slot according to the slot ID on blank portions of the step SP6 ₃ In the prompt template in the text , Prompt text is repeated for the prompt template. In step SP6 ₄ executes an operation of outputting a prompt list file. The prompt list 104A illustrated in FIG. 7 is generated by the processing so far.

Next, the prompt sentence list 104A is input to the scenario generation unit 103, and the scenario file 21 is output. The processing procedure of the scenario generation unit 103 is shown in FIG. In step SP8 ₃ in scenario generation process, if found blank associating the prompt ID scans through the scenario templates 101A, inserted into extraction blank portion than prompt list 104A prompt text corresponding with its corresponding prompt ID Repeat to do. According to the scenario template 101A shown in FIG. 2 and the prompt sentence list 104A shown in FIG. 7, the scenario file 21 shown in the example of FIG. 3 is generated.

Here, a scenario template 101A, part of which is shown in FIG. 2, will be described. The scenario template 101A is a file that finally becomes a template of the scenario file 21 generated by this method.
This scenario template 101A is characterized in that almost all of the control contents on the system side required to execute and complete a dialog for a specific application are described in advance in a programming language that describes a scenario file. The difference from the complete scenario file is that the language expression is used as a response, for example, characters are displayed (when voice playback is performed in synchronization with the agent contract operation, a balloon is drawn on the screen at the same time. When text is displayed inside the balloon), when synthesizing speech, the parts that indicate the text to be displayed and synthesized are blank, and each blank is associated with a prompt ID. .

The prompt ID is described in association with each prompt ID in the prompt template 102A used by the prompt text generation means 104 described above. In FIG. 2, the content of the balloon is displayed simultaneously with the reproduction of the action “RestPose” by the animation by the agent character shown in the third line L3, but it is blank and is associated with the ID of prompt001. (<!-Prompt001-><!-/Prompt001->).
In addition, if you choose to use a pre-recorded file instead of real-time speech synthesis during conversation, due to problems such as the processing capability at the time of system execution, the prompt of the response prompt in the development application will be used. In order to create an audio file, a prompt audio generation processing procedure shown in FIG. 9 is used.

In Figure 9-A, perform step SP9 ₁ and SP9 _2, a case where automatically generates an audio file by using the voice synthesis techniques. Since there is no restriction on the processing time for synthesis compared to the case where synthetic speech is created in real time during conversation, it is possible to create synthesized speech of higher quality. Also, run the Figure 9-B In step SP9 ₃ and SP9 _4, to record a voice pre sentence sentence humans have read a prompt statement list 104A. In this case, it is only necessary to prepare equipment having a function of displaying each prompt text of the prompt sentence list 104A on the screen and a function of collecting sound with a microphone and recording a waveform as a file. When prompt text is displayed on the screen, it is possible to indicate what prosody a person who speaks should utter the text if a brief explanation of the conversation situation is also displayed. This explanatory text may also be described in association with the prompt ID in the prompt template 102A.

Next, a word information list 13 is prepared (FIG. 10). The word information list 13 is a list of words (keywords) that serve as input values for each slot in the development application, and is used for notation and speech recognition of the words. This is a list of readings for reading and IDs of slots to be input destinations.
The created word information list 13 is first input to the result display document list generation means 106 and processed by the result display document list generation processing procedure shown in steps SP11 ₁ to SP11 ₆ in FIG. This template is generated and becomes a result display document list template 106A (FIG. 12). The result display document list template 106A is a list showing which document the development application displays on the screen in accordance with the combination of keywords entered in each slot. The result display document list generating means 106 automatically creates a list of combinations of this keyword and the intermediate result document setting position information and makes it a file. By using a Web browser or the like specified by the intermediate result document installation position information 12 for screen display, a document stored in a storage device connected to a personal computer executing the development application or a server connected to a network is read. Can be displayed.

When a keyword is input to all slots, a document scattered on a server on the network is displayed. Therefore, for the result display document list template 106A, the result display document list in the form of a URL, for example, the address of the server holding the document, the location (path information) of the file in the storage device, and the file name. This is described by the generation means 106. For the description, the content creator creates in advance a correspondence table of keyword combinations and URLs of documents to be displayed when keywords are inputted to all slots in advance, and inputs to the result display document list generation means 106 And Results document list generating unit 106 is able to describe the URL immediately after step SP11 ₄ based on that list.

On the other hand, in the middle of the dialogue when some slots are not filled in, the results displayed on the screen to indicate which slot keyword is not yet entered and which keyword is entered in which slot The document file generation means 107 generates a halfway result document file. The storage location and file name in the storage device of the developed application are described for the file (steps SP11 ₄ to SP11 _{5 in} FIG. 11). The intermediate result document storage position is prepared as intermediate result document installation position information 12 describing the address and path information on the server network where the intermediate result document file is stored. The result display document list generation unit 106 sets the regular file names of the intermediate result document files generated by the intermediate result document file generation unit 107 described in the next paragraph and the intermediate result document installation position information 12 together with the combinations of keywords. The combined complete path information may be generated and added. Furthermore, when the URL or the like of the document to be displayed can be generated not only for the intermediate result but also for the final result, it may be generated and automatically assigned.

FIG. 13 shows a processing procedure of the intermediate result document file generation means 107. Intermediate result document file generating unit 107 reads the corresponding slot ID and slot name from the slot name list 11 in step SP13 _1, there is no keyword to one or more slots out of a combination of keywords entering each slot in step SP13 ₃ the combination generates, as illustrated in FIG. 14 in each combination in step SP13 _4, generates a document file that lists the keywords to be entered keywords and input candidates for the current state of each slot (intermediate result document file) To do. The resulting file is stored by applying a file name according to a predetermined rule in step SP13 _5. As a rule, there is a method of assigning numbers to the keywords entering each slot in the order in which they appear in the word information list and indicating them by a set of numbers of input keywords of each slot. For example, the file name “C-1-.html” is given in the first line L1 of the result display document list template example of FIG. This is a file name indicating that it is a halfway result document file to be displayed when the first keyword is input in the condition 1 slot and no keyword is input in the condition 2 slot.

  In other words, if there is a number following the hyphen, the keyword indicated by that number is entered, and if there is no number, the corresponding slot has no value. “C--2.html” in the eighth line L8 in FIG. 12 indicates that the second keyword is input in the condition 2 slot, and the condition 1 slot is the screen display content in an uninput state. The file format of the intermediate result document file is described in the HTML format in this embodiment of FIG. 12, but any file format that can be displayed by the screen display means (for example, an image file format) may be used. Furthermore, in the case of HTML format, a CSS (cascading style sheet) file is linked, and the background image and the typeface, size, color, etc. of the background image when the intermediate result document is displayed are prepared as a CSS file in advance. It may be possible to change easily.

Next, the word information list 13 is input to the slot / class replacement unit 108 to create the keyword list 25, and the class language model 26 and the recognition dictionary 27 are generated by the class language model and recognition dictionary generation unit 109.
In the class language model generation method, in the creation process, the example sentence corpus is morphologically analyzed and decomposed into word strings, and if there is a word to be classified in the word string, the word is replaced with a class. Do. At that time, a correspondence table indicating which word belongs to which class is required. In this embodiment, the input slot ID is used as class information as the word class of each keyword. The keyword list 25-slot class replaced unit has generated 15, also illustrates an example of a slot class rereading procedure indicated at step SP16 ₁ ~SP16 ₃ in FIG. 16.

The example sentence corpus is a collection of utterance example sentences composed of general expressions that do not depend on the topic (domain) targeted by each development application. However, in order to improve the speech recognition rate in the developed application, it is desirable to prepare an utterance example using a domain-dependent expression when creating a language model. In this embodiment, an additional example sentence corpus 14, which is a collection of example sentences dependent on the domain collected by the developer at the class language model creation stage, is taken in, and after morphological analysis of each example sentence, the keyword list is used to classify the keyword part as a class name. Replace with
The class language model and recognition dictionary generation means 109 generates a class language model 26 and a recognition dictionary 27 using the prepared example sentence corpus and the additional example sentence corpus 14. Details of this means are described in detail in Patent Documents 3 and 4 described in the prior art. The example sentence corpus used by this means is prepared according to the scenario template name selected by the developer. Each example sentence in this example corpus is morphologically analyzed in advance, and the class portion is replaced with a class having the same name as the slot ID.

FIG. 17 shows an embodiment of the dialogue scenario execution apparatus. In the figure, reference numeral 200 denotes an interactive scenario execution device. The dialogue scenario execution apparatus 200 includes a microphone 201, a speaker 202, a voice recognition unit 203, an utterance understanding unit 204, a dialogue scenario execution unit 205, a voice reproduction unit 206, a result display execution unit 208, and a screen display unit 209. Composed.
The dialogue scenario execution means 205 interprets and executes the contents of the scenario file 21 generated by the audio dialogue content creation apparatus 100. That is, the dialogue scenario execution means 205 outputs an instruction to each part according to the contents of the scenario file 21. At the time when the voice is to be played back, the voice playback means 206 is instructed to play back the prompt voice 22, and the action of the agent animation is instructed to be displayed in a balloon or the like together. Further, the voice recognition unit 203 is instructed to perform voice recognition using the class language model 26 and the recognition dictionary 27.

For the utterance understanding unit 204, the recognition word string is received from the speech recognition unit 203, the keyword is taken out of the recognition word sequence, and the class name in the keyword list 25 is used as the slot ID to which slot each keyword should be input. Instructs to read and compare and output as an input value to the corresponding slot.
Then, the input state of the slot is transmitted to the result display execution means 208, and the result display execution means 208 takes out information on the file to be displayed on the screen as the screen contents corresponding to the input state from the result display document list 23, and displays the screen. The display unit 209 is instructed to display.

If it is an intermediate result document file, the screen display means 209 reads and displays the intermediate result document file stored in the storage area indicated by the intermediate result document installation position. Then, the result document is read from an external server or the like, and the resulting information is displayed.
The voice conversation content creation device 100 and the dialogue scenario execution device 200 described above install the voice dialogue content creation program and the voice dialogue program described in the program language in the computer, and install them in a CPU (central processing unit) provided in the computer. This is realized by deciphering and executing.

  Each of the voice interactive content creation program and the voice interactive program is recorded on a computer-readable recording medium, and is installed in the computer through the recording medium or a communication line.

  The voice dialogue content creation method, apparatus, and program according to the present invention are each utilized at the voice dialogue content creation site.

The block diagram for demonstrating the Example of the audio | voice dialog content creation apparatus of this invention. The figure which illustrated a part of scenario template used by this invention. The figure which shows an example of the scenario file produced | generated with the audio | voice dialog content creation apparatus by this invention. The figure which shows an example of the slot name list | wrist used by this invention. The figure which shows an example of the prompt template group used by this invention. The flowchart for demonstrating an example of the prompt text production | generation procedure used by this invention. The figure for demonstrating the prompt sentence list produced | generated by the prompt text production | generation procedure shown in FIG. The flowchart for demonstrating the scenario production | generation process procedure used by this invention. The flowchart for demonstrating the prompt audio | voice production | generation process procedure used by this invention. The figure for demonstrating the word information list used by this invention. The flowchart for demonstrating the result display document list production | generation process procedure used by this invention. The figure for demonstrating the result display document list template used by this invention. The flowchart for demonstrating the intermediate result document file generation | occurrence | production process procedure used by this invention. The figure for demonstrating the intermediate result document file used by this invention, and its display result. The figure for demonstrating an example of the keyword list used by this invention. The flowchart for demonstrating the slot class replacement process procedure used by this invention. The block diagram for demonstrating an example of the dialogue scenario execution apparatus which performs a voice dialogue using each content produced | generated by the voice dialogue content creation apparatus of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Input information 100 Voice dialogue content production apparatus 11 Slot name list 101 Scenario template group 12 Intermediate result document installation position information 102 Prompt template group 13 Word information list 103 Scenario generation means 14 Additional example sentence corpus file 104 Prompt text generation means 20 Voice dialogue content 104A Prompt sentence list 21 Scenario file 105 Prompt sound generation means 22 Prompt sound file 106 Result display document list 23 Result display document list generation means 24 Intermediate result document file 107 Intermediate result document 25 Keyword list file generation means 26 Class language model 108 Slot Class 27 recognition dictionary
109 class language model and
Recognition dictionary generation means
110 Example Corpus

Claims (6)

A scenario template for creating a scenario that defines the order of interaction and the display order of images suitable for the application to be developed;
A plurality of slot names representing input item names determined by the application to be developed, and slot IDs attached to these slot names;
Prompt template for creating a plurality of dialogue sentences that promptly input to the dialogue partner by guiding the input by substituting the slot name in accordance with the slot ID embedding position, and guiding appropriate input,
And substituting the slot name in accordance with the slot ID embedded in the prompt template, substituting each of the dialogue sentences in which the slot name is embedded in the scenario template, and generating a scenario file and a prompt voice file. A method for creating spoken dialogue content.   2. The voice dialogue content creation method according to claim 1, wherein a list of words used for an application to be developed in addition to creation of the scenario file and prompt voice file, storage location information of example data representing a result of the dialogue, and the middle of the dialogue A speech dialogue content creation method, comprising: creating a result display document list and an intermediate result file document file using storage position information of example sentence data representing the situation of A scenario template that stores a scenario template that defines the order of interaction and the display order of images suitable for the application to be developed;
A slot name list storage unit storing a plurality of slot names representing input items determined by an application to be developed and slot IDs attached to these slot names;
A prompt template in which a slot ID is embedded in advance, and prompts the dialogue partner by substituting the slot name according to the slot ID embedding position, and stores a plurality of dialogue sentence templates for guiding appropriate input, and
Prompt text creation means for generating a prompt sentence list by substituting the slot name according to the slot ID embedded in the dialog sentence template stored in the prompt template,
A scenario generation means for generating a scenario file by substituting each of the dialogue statements in which the slot names are embedded in the scenario template;
Prompt voice generation means for substituting each of the dialog sentences in which the slot names are embedded in the prompt sentence list and generating a prompt voice file;
A voice interactive content creation device comprising: 4. The interactive content creation apparatus according to claim 3, wherein the intermediate result document installation position information storage unit, the word information list storage unit, and the intermediate result document installation position information and word list stored in the intermediate result document installation position information storage unit. A result display document list generating means for generating a result display document list for displaying a screen representing the result of the dialogue from the word list stored in the storage unit;
A halfway result document file generating means for creating a halfway result document file for representing a situation during the dialogue from the slot name list stored in the slot name list storage unit and the word information list;
A voice interactive content creation device comprising:   5. A spoken dialogue content creation program that is described in a program language that can be read by a computer and causes the computer to function as the spoken dialogue content creation device according to claim 3. 6. A recording medium comprising a recording medium readable by a computer, wherein the audio dialogue content creation program according to claim 5 is recorded on the recording medium.

Images