JP4223841B2 - Spoken dialogue system and method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a speech dialogue system and method that has a database to which metadata is assigned for each column, and that automatically creates a grammar for speech recognition by referring to the database.
[0002]
[Prior art]
Along with the rapid development of computer technology in recent years, dialogue systems using voice have rapidly become widespread. In the spoken dialogue system, in order to accurately recognize the user's utterance, first, acoustic recognition is performed by referring to the speech recognition dictionary for each phoneme, and then by referring to the speech recognition grammar. Grammatic recognition is often done.
[0003]
Conventionally, when creating a speech recognition grammar used in a speech dialogue system, a system developer first designs a dialogue scenario according to the service content, and utterance content (word string) that a user can utter in the dialogue according to the scenario. ) Has been modeled as a language generation mechanism such as a finite state automaton.
[0004]
And for speech recognition grammar, not only system-driven recognition grammar, which is a recognition grammar that restricts the user's utterance content, but also user-driven recognition grammar that allows the user to speak freely. Various ideas have been made so that grammar can be generated.
[0005]
For example, in (Patent Document 1), not only system-driven voice conversation but also guidance output is controlled so that a user's free utterance can be accepted, and recognition is performed according to a response from the user. A spoken dialogue system capable of switching grammar is disclosed.
[0006]
[Patent Document 1]
JP 2002-342065 A
[0007]
[Problems to be solved by the invention]
However, the process of generating recognition grammars is a process that consumes a great deal of man-hours. In particular, when generating user-driven recognition grammars, all of the dialog contents that the user may utter are covered. In practice, there is a problem that a complete recognition grammar cannot be generated.
[0008]
In addition, a speech recognition grammar is constructed by using words (tokens) constituting user utterances and semantic tags for adding semantic information thereto. For example, in an information retrieval system using speech, it is obtained from recognition results. The database to be searched is searched based on the keyword and semantic tag information to be searched. In this case, two processes of the voice recognition process and the information search process are performed to extract the search result that is a response to the user. There was a need.
[0009]
In order to solve the above-described problems, the present invention can easily create a user-driven or system-driven grammar for speech recognition and can quickly extract a response to the user. The purpose is to provide a system.
[0010]
[Means for Solving the Problems]
In order to achieve the above object, a speech dialogue system according to the present invention has a database storing content information to be searched in a speech dialogue service, and recognizes at least a voice input unit for inputting a user utterance and a user utterance. A voice dialogue system including a voice recognition unit, a dialogue management unit that controls response instructions according to recognition results, and a response generation / output unit that generates and outputs a response to a user. A voice tag for creating a speech recognition grammar including a token tag and a data tag, in which a token tag that is identification information indicating a category to be spoken and a data tag that is identification information indicating data to be searched are set for each column. It further includes a recognition grammar creation unit that recognizes user utterances based on the created speech recognition grammar. And wherein the Rukoto.
[0011]
With this configuration, the search target database is prepared by preparing a database to which a token tag that is identification information indicating a category to be uttered by a user and a data tag that is identification information indicating data to be searched are provided. The recognition grammar can be generated simply by referring to the target database, and the search result can be extracted at the time of speech recognition, so the search processing after speech recognition can be omitted and the processing load can be reduced. It becomes possible.
[0012]
In the speech dialogue system according to the present invention, the speech recognition grammar creation unit preferably creates speech recognition grammar corresponding to a user's free speech or speech recognition grammar with restricted speech. This is because it is possible to easily create both user-driven / system-driven grammars for recognition.
[0013]
The speech dialogue system according to the present invention preferably includes means for recording a speech recognition result or dialogue establishment status as log data, and it is preferable to switch a speech recognition grammar to be used or created based on the log data. This is because the response to the user can be dynamically switched according to the progress of the conversation, and a speech recognition dictionary corresponding to the situation can be created.
[0014]
In addition, the speech dialogue system according to the present invention calculates the ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used based on the log data, and calculates the calculated ratio. Is less than a predetermined threshold, the grammar for speech recognition is switched to a grammar for speech recognition with restricted utterances, and if the threshold is exceeded, the grammar for speech recognition is changed to a free utterance. It is preferable to switch to a corresponding speech recognition grammar. This is because the fact that the data that is the final search result has been output indicates that the dialogue has been performed normally, so that it can be determined that the larger the number of such times, the more utterances can be handled.
[0015]
In addition, the speech dialogue system according to the present invention calculates the ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used based on the log data, and calculates the calculated ratio. If the threshold is less than or equal to a predetermined threshold, create a speech recognition grammar with limited utterances, and create a speech recognition grammar corresponding to free utterances when the threshold is exceeded. It is preferable. This is because the fact that the data that is the final search result has been output indicates that the dialogue has been performed normally, so that it can be determined that the larger the number of such times, the more utterances can be handled.
[0016]
Further, the present invention is characterized by software that executes the above-described voice interaction system as a processing step of a computer. Specifically, a database storing content information to be searched for in a voice interaction service is stored. A voice interaction method comprising: inputting at least a user utterance; recognizing a user utterance; controlling a response instruction according to a recognition result; and generating and outputting a response to the user. In the database, a token tag that is identification information indicating a category to be uttered by a user and a data tag that is identification information indicating data to be searched are set for each column, including the token tag and the data tag. And generating a speech recognition grammar, which is created in the step of recognizing user utterances. It characterized in that it is a voice dialogue method and a computer-executable program for implementing such a step of recognizing a user's utterance based on the voice recognition grammar.
[0017]
With such a configuration, by loading and executing the program on a computer, a token tag that is identification information indicating a category to be uttered by a user and data that is identification information indicating data to be searched are obtained as a search target database. By preparing a database with a tag attached, it is possible to generate a recognition grammar simply by referring to the database to be searched, and the search results can be extracted during speech recognition. A voice interaction system that can be omitted and can reduce the processing load can be realized.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a voice interaction system according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a voice interaction system according to an embodiment of the present invention.
[0019]
In FIG. 1, an utterance or a character by a user is input from the input receiving unit 1. For example, when the voice interactive system according to the present embodiment is an information search system, a search condition by the user is input as voice data. Here, the user may be an engineer such as a system engineer who creates a grammar for speech recognition and an end user who performs a search task.
[0020]
That is, a technician such as a system engineer who creates a speech recognition grammar generates speech data or commands from the input receiving unit 1 in order to generate a speech recognition grammar used in the speech dialogue system according to the present embodiment. Enter the character data. When the speech recognition grammar is generated using the speech data, the input analysis unit 2 recognizes the speech using a general speech recognition grammar and specifies the input content.
[0021]
When the input content is passed to the dialogue management unit 3, the speech recognition grammar creation unit 12 generates a speech recognition grammar by referring to the search database 11 to be searched together with the basic dictionary 13, and the user-initiated speech The data is stored in the recognition grammar storage unit 14 or the system-driven speech recognition grammar storage unit 15.
[0022]
Further, the end user who performs the search task inputs voice data as a search condition from the input receiving unit 1. The speech data is analyzed in the input analysis unit 2 using the initial speech recognition grammar or the speech recognition grammar already generated in the speech recognition grammar creation unit 12, and the analysis result is passed to the dialogue management unit 3. .
[0023]
Here, in the conventional voice dialogue system, the dialogue management unit 3 inquires the retrieval database 11 according to the input content, thereby obtaining the retrieval result that matches the input condition.
[0024]
On the other hand, according to the present invention, when an end user's utterance is input from the input receiving unit 1, a search result is also acquired when speech recognition is performed using a speech recognition grammar generated and stored in advance. It has a feature in that it can. Of course, in a situation where the speech recognition grammar creation unit 12 does not generate the speech recognition grammar, a search result is obtained by a database query in the dialog management unit 3.
[0025]
First, the basic dictionary 13 is a dictionary for extracting basic information when creating a speech recognition grammar. In the basic dictionary 13, information related to “reading” for each word in the search database 11 and words that are not present in the search database 11 and are necessary for speech recognition grammar are registered. For example, a word such as “san” or “kun”, a particle or the like is registered for the word “name”.
[0026]
Next, the search database 11 is a database corresponding to a task such as a schedule guide or a postal code search, for example, whether the content to be input by the user or the content to be presented to the user for each element of the database. Identification information for identification is added as a tag.
[0027]
For example, FIG. 2 shows an exemplary diagram of a data structure when the search database 11 is a schedule guidance database. In FIG. 2, a tag symbol “(T)” is added to each word in the name column, each word in the date column, and the word “plan”. This is an identifier (tag) indicating that it is used as a “token” in the user's inquiry.
[0028]
Further, a symbol “(D)” is added to each word in the schedule column. This is an identifier (tag) for representing a response to a user's inquiry, that is, used as a search result. The search database 11 is created in advance by a user who executes the voice interaction system or a user who provides the voice interaction system.
[0029]
The speech recognition grammar creation unit 12 creates a speech recognition grammar based on data registered in the search database 11 in response to a user request. For example, when the user requests the schedule guidance database shown in FIG. 2 to create a grammar for speech recognition composed of items of “name”, “date”, and “schedule”. Uses the words registered in the name field, date field, and schedule field to create a speech recognition grammar as an automaton.
[0030]
At this time, the word to which the token tag “(T)” is added as an identifier is “token”, and the word to which the data tag “(D)” is added as an identifier is “response content to the user”. Built into speech recognition grammar. Further, by referring to the basic dictionary 13 described above, words that do not exist in the search database 11 such as “san” and “kun” can be incorporated into the speech recognition grammar as necessary.
[0031]
Thus, the speech recognition grammar is constructed according to the automaton model. Specifically, a “state” for outputting each word registered in the search database 11 is defined, and a state network is constructed so that the state is changed by outputting the word. The method for creating the automaton model is not particularly limited, and any conventional method may be used.
[0032]
An example of the speech recognition grammar created based on the schedule guidance database shown in FIG. 2 is shown in FIG. In FIG. 3, ◯ indicates a state, S0 indicates an initial state, and S1 to Sn (n is a natural number) below indicate a state after state transition. Moreover, the arrow has shown the transition between states. That is, an item to which a token tag or a data tag is added in the search database 11 indicates a condition for the state transition.
[0033]
Here, when the user desires to create a user-driven speech recognition grammar, it is necessary to create the speech recognition grammar assuming the user's free speech. Therefore, as shown in FIG. 4, a speech recognition grammar that combines words from the search database 11 and the basic dictionary 13 is automatically created so as to cover each state and transition, and a user-driven speech recognition grammar 14 is created. To store.
[0034]
In FIG. 4, the value of each item extracted from the search database 11 is substituted into the portion surrounded by the symbols “<” and “>”. Therefore, when the grammar for speech recognition is generated, the data itself to which the data tag that is the search result itself is added is included. In addition, suffixes such as “˜san” are extracted from words registered in the basic dictionary 13.
[0035]
FIG. 5 shows the created speech recognition grammar in the form of an automaton. As shown in FIG. 5, the automaton is created in a form including each item value, and in the final state, the data tag is attached, that is, the search result corresponding to the search condition defined in the state transition is shown. become.
[0036]
Also, if the user wishes to use the system-driven speech recognition grammar, the user's utterance to the question prepared by the system, “Who are you planning?” Or “What is your plan?” For this reason, as shown in FIG. 6, a speech recognition grammar that accepts a minimum number of words is automatically created and stored in the system-driven speech recognition grammar storage unit 15.
[0037]
FIG. 7 shows the created speech recognition grammar in the form of an automaton. Similarly to FIG. 5, FIG. 7 also creates an automaton in a form including each item value, and in the final state, the search result corresponding to the data tag attached data, that is, the search condition defined in the state transition is shown. Will be.
[0038]
The speech recognition grammar may be created in advance according to the user's request, or may be dynamically created in response to a request from the dialogue management unit 3 during the voice dialogue.
[0039]
The input analysis unit 2 recognizes the user utterance input by the input receiving unit 1. As the speech recognition grammar used at this time, the user-driven type stored in the user-driven speech recognition grammar storage unit 14 is used. The speech recognition grammar or the system-driven speech recognition grammar stored in the system-driven speech recognition grammar storage unit 15 is referred to and collated with the input speech.
[0040]
The collation with the input speech is executed only for items to which token tags are assigned in the speech recognition grammar. Moreover, since the item to which the data tag extracted in the verification process is assigned is the response content directly to the user, the content of the item is transmitted to the dialog management unit 3 as a response result to the user.
[0041]
The dialogue management unit 3 determines a correspondence policy with the user based on the result of speech recognition. If the input analysis unit 2 cannot obtain a result or if a dialogue is not established, it is determined that the user-initiated speech recognition grammar cannot be used, and the speech recognition grammar is changed to the system-driven speech recognition grammar. The voice recognition grammar switching unit 16 is instructed to switch to
[0042]
In addition, a log data recording unit (not shown) is provided so as to record the voice recognition result and conversation establishment status as log data, and the grammar for voice recognition is switched based on the log data recorded in the log data recording unit. Is also possible.
[0043]
For example, while searching log data, calculate the ratio between the number of times a column with a data tag was output and the number of times the speech recognition grammar was used, and based on the calculated ratio, the speech recognition grammar Switch. In this case, when the calculated ratio is less than or equal to a predetermined threshold, the dialog may not be performed normally due to the fact that the ratio of the data that is the final search result is low. Therefore, it is possible to facilitate normal conversation by switching the speech recognition grammar to a system-driven speech recognition grammar in which speech is restricted.
[0044]
On the other hand, if the threshold value is exceeded, the fact that the final search result data is output is high, which indicates that the dialogue is not performed normally. The grammar is switched to a grammar for user-initiated speech recognition corresponding to free speech.
[0045]
It is also conceivable to give an instruction to the speech recognition grammar creation unit 12 to create the speech recognition grammar again by adding the information already extracted in the dialogue. In this way, by dynamically switching between user-driven speech recognition grammar and system-driven speech recognition grammar according to the state of dialogue, a mixed-driven speech recognition grammar having both features is used. A similar dialogue can be realized.
[0046]
The speech recognition grammar switching unit 16 switches the speech recognition grammar used in the input analysis unit 2 in accordance with an instruction from the dialogue management unit 3, and the user-driven speech recognition grammar storage unit 14 and the system-driven speech recognition A grammar for recognition corresponding to each instruction is extracted from the grammar storage unit 15.
[0047]
Finally, the system response generation / output unit 4 generates a response content to the user and outputs the generated content to the user. In the present embodiment, when the speech recognition grammar is created based on the search database 11, the data that is the response content itself is also incorporated into the speech recognition grammar. Therefore, unlike the conventional speech dialogue system, it is not necessary to create a grammar for speech recognition and recognize input speech, and then search the database based on the recognition result, thereby reducing the processing load of computer resources. It becomes possible.
[0048]
Next, the flow of processing of a program that realizes the voice interaction system according to the embodiment of the present invention will be described. First, FIG. 8 shows a flowchart of processing of a program for executing recognition grammar generation in the speech dialogue system according to the exemplary embodiment of the present invention.
[0049]
In FIG. 8, first, identification information for identifying whether each item in the search database 11 is an item to be input by the user or an item including contents to be presented to the user is added (step S801). Then, a command input such as speech data or character data for generating speech recognition grammar is received from the creator of the speech recognition grammar (step S802).
[0050]
Next, the ratio of the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated based on the log data in which the speech recognition result and the conversation establishment status are recorded. (Step S803).
[0051]
If the calculated ratio exceeds a predetermined threshold (step S804: Yes), the user-initiated speech recognition grammar corresponding to a free utterance is generated or updated (step S804: Yes). In step S805), if the calculated ratio is equal to or less than a predetermined threshold (step S804: No), a system-driven speech recognition grammar in which speech is restricted is generated or updated. (Step S806). Although omitted in FIG. 8, the speech recognition grammar can be generated or updated according to the request of the grammar creator regardless of the log data.
[0052]
Then, the contents of the processing are recorded as log data (step S807).
[0053]
Next, FIG. 9 shows a flowchart of the processing of a program that executes search processing using the generated recognition grammar in the voice interaction system according to the embodiment of the present invention.
[0054]
In FIG. 9, first, input of search voice data is received from the end user (step S901).
[0055]
Next, the ratio of the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated based on the log data in which the speech recognition result and the conversation establishment status are recorded. (Step S902).
[0056]
If the calculated ratio exceeds a predetermined threshold (step S903: Yes), the speech recognition processing is performed using the user-driven speech recognition grammar corresponding to a free utterance as the speech recognition grammar. (Step S904), and if the calculated ratio is equal to or less than a predetermined threshold (step S903: No), the system-driven speech recognition grammar with restricted utterance is used as the speech recognition grammar. Then, voice recognition processing is performed (step S905). Search results are also obtained in the speech recognition process.
[0057]
Then, a response corresponding to the input condition is generated and output (step S906), and the voice recognition result and conversation establishment status in the processing are recorded as log data (step S907).
[0058]
As described above, according to the present embodiment, as a search target database, a token tag that is identification information indicating a category to be uttered by a user and a data tag that is identification information indicating data to be searched are provided. By preparing a database, it is possible to generate a grammar for speech recognition simply by referring to the database to be searched, and to extract search results during speech recognition, so that search processing after speech recognition can be omitted. It is possible to reduce the processing load.
[0059]
As shown in FIG. 10, the program for realizing the voice interaction system according to the embodiment of the present invention is not only a portable recording medium 102 such as a CD-ROM 102-1 and a flexible disk 102-2, but also a communication line. It may be stored in any of the other storage device 101 provided in front of the computer 101 or a recording medium 104 such as a hard disk or a RAM of the computer 103. When the program is executed, the program is loaded and stored in the main memory. Executed.
[0060]
Also, the speech recognition grammar generated by the speech dialogue system according to the embodiment of the present invention is also portable recording medium 102 such as CD-ROM 102-1 or flexible disk 102-2 as shown in FIG. It may be stored not only in the other storage device 101 provided at the end of the communication line, but also in the recording medium 104 such as the hard disk or RAM of the computer 103. For example, the voice dialogue according to the present invention It is read by the computer 103 when using the system.
[0061]
The following variations are also noted.
[0062]
(Supplementary note 1) Having a database storing content information to be searched in the voice dialogue service,
A voice input unit for inputting at least a user utterance;
A voice recognition unit for recognizing the user utterance;
A dialogue manager that controls response instructions according to the recognition results;
A spoken dialogue system including a response generation / output unit that generates and outputs a response to a user,
In the database, a token tag that is identification information indicating a category that the user should speak and a data tag that is identification information indicating data to be searched are set for each column,
A speech recognition grammar creation unit that creates a speech recognition grammar including the token tag and the data tag;
In the voice recognition system, the voice recognition unit recognizes the user utterance based on the created voice recognition grammar.
[0063]
(Supplementary note 2) The speech dialogue system according to supplementary note 1, wherein the speech recognition grammar creation unit creates speech recognition grammar corresponding to a user's free speech or speech recognition grammar in which speech is restricted.
[0064]
(Supplementary note 3) The voice dialogue system according to supplementary note 1, including means for recording a voice recognition result and a dialogue establishment status as log data, and switching a grammar for voice recognition to be used or created based on the log data.
[0065]
(Supplementary Note 4) Based on the log data, a ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated. If the threshold is less than or equal to the threshold, the grammar for speech recognition is switched to a grammar for speech recognition with restricted utterances, and if the threshold is exceeded, the grammar for speech recognition is compatible with free utterances. The spoken dialogue system according to appendix 3, wherein the grammar for voice recognition is switched.
[0066]
(Supplementary Note 5) Based on the log data, a ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated. If the threshold is less than or equal to the threshold value, a speech recognition grammar with limited utterances is created, and if the threshold is exceeded, a speech recognition grammar corresponding to a free utterance is created. The spoken dialogue system described.
[0067]
(Additional remark 6) It has the database which stored the content information used as the search object in a voice interaction service,
Inputting at least a user utterance;
Recognizing the user utterance;
A step of controlling a response instruction according to the recognition result;
A voice interaction method including a step of generating and outputting a response to a user,
In the database, a token tag that is identification information indicating a category that the user should speak and a data tag that is identification information indicating data to be searched are set for each column,
Further comprising the step of creating a speech recognition grammar including the token tag and the data tag,
In the step of recognizing the user utterance, the user utterance is recognized based on the created speech recognition grammar.
[0068]
(Supplementary note 7) Having a database storing content information to be searched in the voice dialogue service,
Inputting at least a user utterance;
Recognizing the user utterance;
Controlling response instructions according to recognition results;
A computer-executable program that embodies a spoken dialogue method that includes generating and outputting a response to a user,
In the database, a token tag that is identification information indicating a category that the user should speak and a data tag that is identification information indicating data to be searched are set for each column,
Creating a speech recognition grammar including the token tag and the data tag;
A computer-executable program for recognizing the user utterance based on the created speech recognition grammar in the step of recognizing the user utterance.
[0069]
【The invention's effect】
As described above, according to the spoken dialogue system of the present invention, the search target database includes a token tag that is identification information indicating a category to be uttered by a user and a data tag that is identification information indicating data to be searched. By preparing the assigned database, it is possible to generate a recognition grammar simply by referring to the database to be searched, and the search result can be extracted at the time of speech recognition, so the search processing after speech recognition is omitted. And the processing load can be reduced.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a voice interaction system according to an embodiment of the present invention.
FIG. 2 is a data configuration example diagram of a search database in the voice interaction system according to the embodiment of the present invention.
FIG. 3 is an exemplary diagram of a grammar for speech recognition in the speech dialogue system according to the embodiment of the present invention.
FIG. 4 is a view showing an example of a user-driven speech recognition grammar in the speech dialogue system according to the embodiment of the present invention.
FIG. 5 is a diagram showing an example of an automaton of a user-driven grammar for speech recognition in the speech dialogue system according to the embodiment of the present invention;
FIG. 6 is an exemplary diagram of a system-driven speech recognition grammar in the speech dialogue system according to the embodiment of the present invention.
FIG. 7 is a diagram showing an example of an automaton of a system-driven grammar for speech recognition in the speech dialogue system according to the embodiment of the present invention.
FIG. 8 is a flowchart of speech recognition grammar generation processing in the speech dialogue system according to the embodiment of the present invention.
FIG. 9 is a flowchart of search processing in the voice interaction system according to the embodiment of the present invention.
FIG. 10 is an exemplary diagram of a computer environment.
[Explanation of symbols]
1 Input receiver
2 Input analysis unit
3 Dialogue Management Department
4 Response generation / output section
11 Search database
12 Grammar creation for speech recognition
13 Basic dictionary
14 Grammar storage for user-driven speech recognition
15 Grammar storage for system-driven speech recognition
16 Grammar switching part for voice recognition
91 Line destination storage device
92 Portable recording media such as CD-ROM and flexible disk
92-1 CD-ROM
92-2 Flexible disk
93 computers
94 Recording medium such as RAM / hard disk on computer

Claims (7)

An audio input unit for inputting a user's utterance even without low,
A speech dialogue system comprising a dialog manager that controls the response indication in response to the recognition result of the inputted user utterance,
User identification information indicating a category to be uttered token tag, or a database data tag is stored in a state of being set for each column is identification information indicating the data to be retrieved in the voice conversation service,
A speech recognition grammar generating unit configured to generate a speech recognition grammar including said token tag and the data tag,
A speech recognition unit that recognizes the user utterance by comparing the item to which the token tag is given in the created speech recognition grammar with the input user utterance ;
A voice dialogue system comprising: a response generation / output unit that outputs an item to which a data tag corresponding to the token tag extracted in the verification process is attached as a response content to the user .   The speech dialogue system according to claim 1, wherein the speech recognition grammar creation unit creates speech recognition grammar corresponding to a user's free speech or speech recognition grammar in which speech is restricted.   The voice dialogue system according to claim 1, further comprising means for recording a voice recognition result and dialogue establishment status as log data, and switching a voice recognition grammar to be used or created based on the log data.   Based on the log data, a ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated, and the calculated ratio is equal to or less than a predetermined threshold value. In some cases, the speech recognition grammar is switched to a speech recognition grammar with restricted utterances, and when the threshold is exceeded, the speech recognition grammar is used for speech recognition corresponding to a free utterance. The spoken dialogue system according to claim 3, wherein the system is switched to grammar.   Based on the log data, a ratio between the number of times the column with the data tag is output and the number of times the grammar for speech recognition is used is calculated, and the calculated ratio is equal to or less than a predetermined threshold value. 4. The speech recognition grammar according to claim 3, wherein a speech recognition grammar in which utterances are restricted is created in some cases, and a speech recognition grammar corresponding to a free speech is created if the threshold value is exceeded. Dialog system. A voice input unit for inputting at least a user utterance;
A dialogue management unit for controlling a response instruction according to the input recognition result of the user utterance;
A token tag that is identification information indicating a category to be uttered by a user, or a database in which a data tag that is identification information indicating data to be searched in a voice interaction service is set in each column. A voice interaction method executed by a computer,
A step of creating a speech recognition grammar including the token tag and the data tag, the speech recognition grammar creation unit provided in the computer;
A step of recognizing the user utterance by collating the input user utterance with the item to which the token tag is added in the created speech recognition grammar, the voice recognition unit provided in the computer;
A response generation / output unit included in the computer includes a step of outputting an item to which a data tag corresponding to the token tag extracted in the verification process is added as a response content to the user. Spoken dialogue method. A voice input unit for inputting at least a user utterance;
A dialogue management unit for controlling a response instruction according to the input recognition result of the user utterance;
A token tag that is identification information indicating a category to be uttered by a user, or a database in which a data tag that is identification information indicating data to be searched in a voice interaction service is set in each column. Audio that causes a computer to perform processing An interactive program,
A speech recognition grammar creation process for creating a speech recognition grammar including the token tag and the data tag;
A speech recognition process for recognizing the user utterance by comparing the item to which the token tag is assigned in the created grammar for speech recognition with the input user utterance;
A spoken dialogue program for causing the computer to execute response generation / output processing for outputting an item to which a data tag corresponding to the token tag extracted in the verification process is added as a response content to the user .

Images