JP2003036094A - Device for speech dialogue and method for processing speech dialogue - Google Patents

Abstract

PROBLEM TO BE SOLVED: To conduct speech dialogue processing with good efficiency and high credibility by conducting branch judgment and determining post-processing about a recognition result toward inputted speech. SOLUTION: In a speech dialogue device 100 provided with a speech dialogue processing part 105 connected with users' telephone set 101 through a communication line 103, several options for fixing the speech inputted from the telephone set 101 and likelihood toward each option are determined as the recognition result, in conducting the speech dialogue processing and the branch judgment is conducted by comparing the likelihood with a standard value in advance fixed to use for judging. The branch judgment determines how to process the recognition result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a telephone voice interaction device and a voice interaction processing method. According to the present invention, voice interaction processing is performed in a device having a voice recognition function of recognizing a voice of a user and converting it into data, and a voice recording function of recording a voice of the user.

[0002]

2. Description of the Related Art An example of a conventional voice dialogue system is disclosed in Japanese Patent Laid-Open No.
No. 0-70613. According to this document, the user inputs to the voice interaction device from the telephone by means of voice or the like. The information input from the telephone is divided into items for which voice recognition is easy and items for which voice recognition is difficult in the voice interaction device.

The voice dialog device performs a voice recognition process for an item for which voice recognition is easy. On the other hand, for an item for which voice recognition is difficult, a voice is temporarily recorded by a voice dialogue device and converted into data manually (“listen up” in the above document) manually. Specifically, the work of converting the data into the data manually by the offline is as follows.

First, a guidance message is sent from the voice interaction device to the telephone of the user. The user utters an address, family name, first name, etc. according to this guidance message. The voice uttered by the user is
It is input to the voice interaction device via the telephone. The input voice is temporarily recorded in the voice interaction device.

After that, the voice dialogue device performs voice recognition on the recorded voice. Then, based on the recognized voice, a plurality of candidates for the address, name, etc. are determined. The plurality of determined candidates are displayed on the screen of the computer terminal or the like.

[0006] An operator (person) performs a work of listening to a plurality of candidates displayed on the screen or the like of a computer terminal. Then, information such as an address and a name is fixed from the plurality of candidates displayed on the screen of the computer terminal.

[0007] According to the above-mentioned document, the work carried out by listening and raising is as follows. The user's voice once recorded is reproduced by an appropriate means. While listening to the reproduced voice, the operator (person) can select a plurality of candidates displayed on the screen of the computer terminal, etc.
Check the data.

[0008]

In the conventional voice dialog device, the operator does not have to utter all the voices input for the items for which voice recognition is difficult, even if the voice recognition device correctly recognizes the input voice. I was going.

[0009] Further, when the user does not recognize a voice recognition item at one time when he / she inputs the voice recognition easily, the user must repeat the input until the voice recognition device recognizes it. There wasn't.

The conventional voice dialogue system has the above problems. For this reason, it has been desired to develop a voice interaction device that performs voice interaction processing such as recognition processing efficiently and reliably with respect to the input voice.

[0011]

Therefore, the inventors of the present application repeatedly conducted various studies and thoughts in order to solve the above-mentioned problems. Then, by deciding whether or not the user's input becomes the branch condition of the next question and deciding the post-processing, it is concluded that efficient and highly reliable voice dialogue processing can be performed. , Came to this invention.

The voice dialogue system of the present invention comprises a voice dialogue processing section connected to the user's telephone through a communication line. Then, the voice interaction device of the present invention performs voice interaction processing according to the following method.

First, the voice dialogue processing section carries out voice dialogue with the telephone of the user and receives the voice of the user from the telephone. Then, the voice interaction processing unit determines a plurality of candidates for determining the received voice and the likelihood for each candidate as the recognition result. After that, the voice interaction processing unit compares the reference value with the likelihood, and makes a branch decision on the recognition result. As a result of the branch determination, how to process the recognition result is determined. That is, the branch judgment in the present invention is a judgment as to what processing should be performed on the recognition result. The reference value is set in advance in the voice dialogue processing section and is used for branch determination.

As described above, according to the voice interactive apparatus and the voice interactive processing method of the present invention, the branch judgment is performed on the recognition result for the input voice. And
As a result of this branch determination, how to process the recognition result is determined. Therefore, when a correct recognition result is obtained for the input voice by the voice interaction device, it is possible to omit the manual work.

Further, according to the voice interactive apparatus and the voice interactive processing method of the present invention, by performing the branch judgment,
Appropriate processing can be determined even when the input voice is not correctly recognized. As a result, it is possible to reduce the burden of re-input processing by the user as compared with the conventional voice interaction device.

That is, according to the voice interaction device and the voice interaction processing method of the present invention, an accurate branch determination is made for the recognition result based on the comparison between the likelihood and the reference value. Therefore, according to the voice interaction device and the voice interaction processing method of the present invention, it is possible to realize efficient and highly reliable voice interaction processing.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of a voice dialogue apparatus of the present invention will be described below with reference to the drawings.
It should be understood that the drawings used in the following description are merely schematic representations to the extent that the present invention can be understood, and therefore the present invention is not limited to the illustrated examples. Further, in each drawing used for the description, the same constituent components are denoted by the same reference numerals, and the duplicate description may be omitted.

[Structure of Embodiment] The structure of a voice interactive apparatus 100 according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 shows the configuration of a voice dialog device 100 according to an embodiment of the present invention.

The voice interaction apparatus 100 comprises a voice interaction processing section 105 connected to the user's telephone 101 via a communication line 103.

For the voice input from the telephone 101, the voice interaction processing section 105 determines a plurality of candidates for determining the voice and the likelihood for each candidate as the recognition result. After that, the voice interaction processing unit 105
The reference value and the likelihood are compared, and a branch decision is made on the recognition result. As a result of this branch determination, how to process the recognition result is determined.

Here, the plurality of candidates in the recognition result are, in detail, for determining the characteristics of the input voice. A plurality of candidates in the recognition result are, for example,
It is determined based on physical characteristics such as a frequency spectrum extracted from the input voice.

The voice dialog device 100 also includes a storage unit 10.
It has 7. In the storage unit 107, voice processing information 137 used for voice dialogue processing is stored in advance.

The voice processing information 137 includes a first information group and a second information group. The first information group has first information and second information used for determining a recognition result. Also,
The second information group has third information used for branch determination. Details of the first to third information included in the first information group and the second information group will be described later.

Next, the configuration of the voice dialogue processing section 105 will be described.

The voice dialogue processing section 105 is provided with a communication line 103.
Line processing unit 109 connected to the telephone 101 via the telephone
And a voice interaction control unit 111 and a voice processing unit 113 that determines a recognition result.

The voice dialogue control unit 111 is connected to the line processing unit 10.
9. It has a function of exchanging voice and required information with the storage unit 107 and the voice processing unit 113, and outputting the result of branch determination. The voice interaction control unit 111 is provided with a basic processing unit 115, a branch determination unit 117, and a memory 119 for storing a reference value.

The basic processing unit 115 reads the first and second information from the storage unit 107 into the memory 119. Then, the basic processing unit 115 uses the first information to perform a voice conversation with the telephone 101 via the line processing unit 109, and receives a voice from the telephone 101.

The voice processing unit 113 receives the voice from the telephone 101 and the second information from the basic processing unit 115. Then, the voice processing unit 113 determines the recognition result using the second information, and transmits this recognition result to the basic processing unit 115.

The branch judgment unit 117 receives the recognition result from the basic processing unit 115. Then, the branch determination unit 117 reads out the third information from the storage unit 107 and the reference value from the memory 119, respectively. Then, the branch determination unit 117
A branch determination is performed based on the comparison between the likelihood and the reference value and the third information.

In this embodiment, the reference value is
It is preferable to set the likelihood in the recognition result. Furthermore, the reference value is optimally set as four reference values of a first threshold value, a second threshold value, a first threshold value difference, and a second threshold value difference.

By the way, it is preferable that the voice processing unit 113 comprises a voice recognition unit 121 and a voice recording unit 123.

The voice recognition unit 121 has a function of receiving a voice from the telephone 101, performing a voice recognition process on the voice, and outputting a recognition result. Therefore, the likelihood in the recognition result has a value that depends on the device configuration of the voice recognition unit 121. The reference value set for the likelihood also depends on the device configuration of the voice recognition unit 121.

On the other hand, the voice recording section 123 is used for the telephone 101.
It has a function of receiving a voice from the user, performing a recording process for the voice, and creating a voice recording file.

As described above, according to the device configuration of the voice dialogue processing unit 105 of this embodiment, the voice recognition unit 1
Reference numeral 21 performs voice recognition on the voice input to the voice interaction device and determines the recognition result. By comparing the likelihood in the recognition result with the reference value, the branch determination unit 117 can evaluate the recognition performance for voice recognition. Details of the comparison between the likelihood and the reference value according to this embodiment will be described later.

Further, each component of the voice dialogue processing section 105 may have any suitable device configuration suitable for voice dialogue processing. At this time, the reference value can be set to a desired value according to the device configuration of the voice recognition unit 121.

The voice dialog device 100 in this embodiment is optimally a computer device composed of a plurality of hardware. Further, it is desirable to use hardware having a program for each component of the voice interaction processing unit 105.

Next, the structure of the storage unit 107 will be described. The storage unit 107 stores the voice processing information 137 and a database unit 135.

The voice processing information 137 has a plurality of files. This multiple files is a spoken dialogue sequence 1
25, a voice message 127, a confirmation recognition grammar 129, a confirmation voice message 131, and a re-input message 133.

Here, it is preferable that the voice dialogue sequence 125 and the voice message 127 are created by an organization that uses the voice dialogue device 100 to provide a voice dialogue service. The voice message 127 includes a plurality of files containing voice messages. The voice interaction sequence 125 will be described in detail later.

The confirmation recognition grammar 129 is used for recognizing the voice when "Yes" or "No" is input as the voice. That is, it is preferable that the confirmation recognition grammar 129 is a file in which a grammar for recognizing an input voice of “Yes” or “No” is recognized.

The confirmation voice message 131 is preferably a file containing a voice message for confirming the recognition result to the user.

Further, the re-input message 133 is optimally a file containing a voice message for prompting the user having the telephone 101 to re-input.

The database unit 135 stores the data necessary for the service provided by the voice dialog device 100, the recognition result, and the like.

Each file 125, 12 in the storage unit 107
In practice, 7, 129, 131, 133 are preferably stored on a computer disk. Also, 1
Each of the files 25, 127, 129, 131, and 133 and the database unit 135 may be stored in the same storage device. Here, the voice interaction device 100 according to the present embodiment is a computer device including each hardware configuring the voice interaction processing unit 105 as described above and a storage device having each file and database unit. Is preferred.

According to the configuration of the storage unit 107 as described above, the voice dialogue sequence 125 and the voice message 127 are used.
Can have any suitable file structure. Therefore, in this embodiment, various voice dialogue services can be provided to the user. Further, in this embodiment, it is also possible to perform voice interaction processing suitable for the content of the voice interaction service.

The files stored in the storage unit 107 are preferably updated as needed.

Next, the voice dialogue sequence 125 will be described with reference to FIG. FIG. 2 is a diagram showing a configuration of the voice dialogue sequence 125. According to this embodiment, the voice interaction sequence 125 is preferably configured to have a plurality of interaction cells 201 and a plurality of global variables 215.

Each dialogue cell 201 is a voice dialogue device 100.
Is the smallest unit of a series of dialogues. Then, in the voice interaction process, the voice interaction control unit 111 repeats the process of each interaction cell after the initial interaction cell. here,
The initial dialogue cell is a dialogue cell which is first processed by the voice dialogue control unit 111 when the voice dialogue processing is started.

The conversation cell has a first information group 219 and a second information group 217. As described above, the first information group 219 includes the first information and the second information.
Further, the second information group 217 includes the third information. This third information is preferably the interactive cell mode 203 that determines the processing of the interactive cell. At this time, the branch determination by the branch determination unit 117 is preferably performed based on the comparison between the likelihood and the reference value and the interactive cell mode 203. Details of the interactive cell mode 203 will be described later.

The first information group 219 includes a voice file name 211 as the first information and a recognition grammar 21 as the second information.
3 is preferably included.

The voice file name 211 designates the file name of the voice message to be transmitted to the telephone 101 from the voice message 127. As described above, the voice message 127 includes a plurality of files containing voice messages. A voice message corresponding to the file designated by the voice file name 211 among the plurality of files included in the voice message 127 is transmitted to the telephone 101.

According to this embodiment, the user replies by voice to the voice message transmitted to the telephone set 101. The voice of the user is recognized by the voice interaction device 100 using the recognition grammar 213.

Further, it is optimal that the second information group 217 includes a preprocessing program 205, a postprocessing program 207, and a next cell pointer 209 in addition to the interactive cell mode 203.

The preprocessing program 205 describes the preprocessing required for voice dialogue. The preprocessing is, for example, processing such as acquisition of time and an incoming call account from the user's telephone 101.

Further, the post-processing program 207 describes post-processing required for voice dialogue, such as processing for a voice recognition result.

The next cell pointer 209 is for designating the next interactive cell to be advanced after all the processing in one interactive cell is completed.

Here, the interactive cell mode 203 is preferably composed of a plurality of flag strings. Then, in each dialogue cell, the dialogue cell mode 203 is
It depends on the pre-processing program 205, the post-processing program 207, and the next cell pointer 209.

That is, the plurality of flag strings constituting the conversation cell mode 203 are (a) whether or not the conversation cell includes a voice recognition process, and (b) there is a branch to the next conversation cell in the conversation cell. Whether (c) the post-processing program 207 substitutes the recognition result into the global variable 215, (d) the post-processing program 207
It indicates whether the recognition result is registered in the database unit 135, and (e) whether the post-processing program 207 performs the process of increasing the global variable 215 or the counter value in the database unit 135 for the recognition result. It is desirable to have.

Here, the process of increasing the global variable 215 or the counter value in the database unit 135 will be described. For example, consider a case where the voice interactive device 100 provides a voice interactive service for conducting some sort of questionnaire. At this time, the telephone 10 owned by the user
From 1, the answers to the questionnaire are sequentially input to the voice interaction device 100. For example, in the selection-type questionnaire, the selection items by the user are input to the voice interaction device 100. Then, in the global variable 215 or the database unit 135, the input selection items are counted, that is, totaled.

Next, the global variable 215 will be described. The global variable 215 is a variable commonly used by the dialogue cells 201. For example, the voice interaction device 10
It is assumed that 0 is for providing a reservation service for travel and the like. Then, in this reservation service, it is assumed that there is a dialogue for the user's answer regarding information such as address, name, date of reservation, and the like. Therefore, the dialog cell 201 at this time includes a dialog cell for prompting the input of the address, a dialog cell for prompting the input of the name, a dialog cell for prompting the input of the reserved date, and the like. And
For each dialogue cell, when the voice response of the user is sequentially input to the voice dialogue device 100, it is substituted into each global variable. Since the information assigned to the global variable is commonly used in each dialogue cell, it is necessary to assign an accurate value.

[Operation of Embodiment] Next, FIG. 1 and FIG.
In addition, the operation of each component in the embodiment of the present invention will be described with reference to the flow charts shown in FIGS. In the flowcharts of FIGS. 3 to 10, the operations performed by the voice dialogue control unit 111, the line processing unit 109, the voice recognition unit 121, and the voice recording unit 123 are respectively described as a dialogue control process, a line control process, and
The voice recognition process and the voice recording process will be described separately. The operation performed by the user's telephone 101 will be described in the process of simply naming the user's telephone. Further, in the figure, the step of each process is shown by adding a number to the symbol S.

1. Voice Interaction Process The voice interaction process performed in this embodiment will be described with reference to the flowchart shown in FIG.

First, in the configuration of this embodiment shown in FIG. 1, the service provider activates the voice interaction control unit 111 to the voice interaction processing unit 105 by a conventionally known starting method. When activated, the voice interaction control unit 111 starts voice interaction processing.

(S301) The voice interaction control section 111 activates the line processing section 109. In the voice dialogue control unit 111, specifically, the basic processing unit 115 is the line processing unit 10.
Start 9.

(S302) In the voice dialogue control section 111, the basic processing section 115 receives the line processing section 109,
Waiting for the line status to change. Waiting for a change in line status
That is, it is in a state of waiting for an incoming call from the telephone 101.

(S303) On the other hand, the activated line processing unit 109 starts the operation of monitoring the state of the communication line 103.

(S304) Then, the user sets the telephone 1
A call is made from 01 to the voice interaction device 100 (a call is made). An outgoing call from the telephone 101 is received by the line processing unit 109. Then, the line processing unit 109 notifies the voice interaction control unit 111 of an incoming call. Here, the incoming call notification is a notification that the outgoing call from the telephone set 101 has been received.

(S305) Upon receiving the incoming call notification, the voice dialogue control section 111 processes the initial dialogue cell of the voice dialogue sequence 125.

In the voice interaction control section 111, specifically, the incoming call notification is received by the basic processing section 115. Then, the basic processing unit 115 reads the information of the initial dialogue cell of the voice dialogue sequence 125 from the storage unit 107 into the memory 119 in response to the incoming call notification. Based on the information read in the memory 119, the basic processing unit 11
5 processes the initial dialogue cell.

(S306) When all the processing in the initial dialogue cell is completed, the voice dialogue control section 111
The presence / absence of the next cell pointer 209 is determined. The presence or absence of the next cell pointer 209 determines whether or not to end a series of dialogues in the voice dialogue processing. Specifically, in the voice interaction control unit 111, the basic processing unit 115 determines whether or not the next cell pointer 209 is present.

When the next cell pointer 209 does not exist in the initial dialogue cell, the basic processing unit 115 considers that all dialogues of the voice dialogue processing are to be ended. Then, the basic processing unit 115 performs the process of S307 described below.

On the other hand, the next cell pointer 2 in the initial dialogue cell
If 09 is present, the operation of the voice interaction control unit 111 returns to the process of S305. Specifically, the basic processing unit 115 reads the next cell pointer 209 into the memory 119. After that, the basic processing unit 115 determines that the next cell pointer 2
The processing of the dialogue cell designated by 09 is performed. The processing of each dialogue cell will be described later.

When the plurality of conversation cells 201 forming the voice conversation sequence 125 each have an ID, the next cell pointer 209 is for designating the ID of each conversation cell. Further, the dialogue cell in which the next cell pointer 209 does not exist is called the last dialogue cell.

(S307) When the processing of the last dialogue cell is completed, the voice dialogue control unit 111 notifies the line processing unit 109 that the line is disconnected, and the incoming call waiting state is again established. Specifically, in the voice interaction control unit 111, the basic processing unit 115 notifies the line processing unit 109 that the line is disconnected. Then, the basic processing unit 115 again waits for an incoming call. Here, the notification of disconnection will be referred to as a disconnection notification.

The line processing unit 109 having received the disconnection notice
The line with the user's telephone set 101 is disconnected, and the system waits for the line state to change again. Here, the voice interaction process for one call of one user is ended.

2. Processing of each dialogue cell Next, in the flowchart shown in FIG. 3, the processing of S305,
That is, the processing of each dialogue cell will be described with reference to the flowchart shown in FIG.

(S401) First, the voice dialogue control section 111.
Executes the preprocessing program 205 for the interactive cell. When the preprocessing program 205 is executed, the preprocessing described in this program is performed.

Specifically, in the voice interaction control section 111, the basic processing section 115 reads the preprocessing program 205 into the memory 119. The preprocessing program 205 read into the memory 119 is executed by the basic processing unit 115.

(S402) Next, the voice dialogue control section 111.
Transmits the voice message specified by the voice file name 211 of the conversation cell to the telephone set 101 of the user via the line processing unit 109.

Specifically, in the voice conversation control unit 111, the basic processing unit 115 causes the voice file name 2 of the conversation cell to be changed.
11 is read from the storage unit 107 into the memory 119. Then, the basic processing unit 115 causes the read audio file name 2
The message specified in 11 is replaced by the voice message 127.
From the inside to the memory 119. After that, the read message is transmitted to the user's telephone 101 by the basic processing unit 115 via the line processing unit 109.

(S403) Telephone 101 is line processing unit 1
From 09, the message is received and reproduced. For example, from the line processing unit 109, "This is XX. Please answer your age as if you were in your 20s or 30s."
Message is received, the telephone set 101 reproduces it.

(S404) After that, the user gives a voice response to the message reproduced by the telephone set 101. The telephone 101 uses the voice interaction device 1 to listen to the user's voice.
To 00.

(S405) The line processing unit 109 receives the voice of the user from the telephone 101 via the communication line 103. The line processing unit 109 transmits the received voice of the user to the voice interaction control unit 111. After that, the voice interaction control unit 111 performs an input process. However, this user's response and the input process of the voice interaction control unit 111 may not exist. For example, this is a case where only a voice message such as a commercial is transmitted and played. The input process will be described later.

(S406) After the input processing is completed, the voice interaction control section 111 executes the post-processing program 207 of the interaction cell and terminates the processing of the interaction cell.

Specifically, in the voice conversation control unit 111, the basic processing unit 115 reads the post-processing program 207 of the conversation cell from the storage unit 107 into the memory 119,
Run. When the post-processing program 207 is executed, the post-processing described in this program is performed.

3. Input Process In the flowchart of FIG. 4 described above, the process of S405, that is, the input process will be described with reference to the flowcharts shown in FIGS. First, the processing from S501 to S506 will be described with reference to the flowchart shown in FIG.

(S501) First, the voice dialogue control section 111
Passes the recognition grammar 213 of the dialogue cell to the voice recognition unit 121 in the voice processing unit 113 and notifies that the voice recognition is started. The voice recognition unit 121 uses the recognition grammar 213 to start the voice recognition process.

Specifically, in the voice conversation control unit 111, the basic processing unit 115 reads the recognition grammar 213 of the conversation cell from the storage unit 107 into the memory 119. And
The recognition grammar 213 read into the memory 119 is transmitted to the voice recognition unit 121 by the basic processing unit 115.

(S502) Also, the voice dialogue control section 111.
In the above, the basic processing unit 115 causes the voice recording unit 123 to
Notify that voice recording has started. The voice recording unit 123 starts the voice recording process.

(S503) After that, the voice dialogue control unit 11
1 is in a state of waiting for a result from the voice recognition unit 121 and the voice recording unit 123. Here, the result waiting state is a state in which the voice recognition unit 121 and the voice recording unit 123 wait for the voice processing result to be transmitted. Here, the voice processing means processing including both the voice recognition performed by the voice recognition unit 121 and the voice recording performed by the voice recording unit 123.

(S504) Next, referring to the flowchart in FIG. 4, the telephone set 101 reproduces the message received from the voice interactive apparatus 100 in S403. The user replies to this message by voice. The telephone 101 transmits the voice of the user to the voice interaction device 100.

(S505) The voice recognition unit 121 outputs the voice of the answer given by the user to the voice dialogue control unit 111.
To receive from. Specifically, in the voice interaction control unit 111, the basic processing unit 115 receives the voice of the user from the line processing unit 109. Then, the basic processing unit 115 transmits the voice of the user to the voice recognition unit 121.

Next, the voice recognition unit 121 performs voice recognition by a conventionally known method in order to determine the received voice. The result of voice recognition, that is, the recognition result is determined as a plurality of candidates and the likelihood for each candidate. Here, the voice recognition unit 121 ends the voice recognition process. In addition, the recognition result is transmitted to the voice interaction control unit 111 by the voice recognition unit 121.

For example, when the user replies "Sanjudai" by voice, the recognition result for the voice of the user is as follows. That is, the first candidate: "30
Generation ”, likelihood of first candidate 0.987; second candidate:“ 40
Teens, likelihood of second candidate 0.765; third candidate: “10
Generation, the likelihood of the third candidate is 0.543.

Here, in this embodiment, the first candidate has the highest likelihood. Then, in the following, the second candidate, the third candidate, ... Are determined in descending order of likelihood.

(S506) The voice recording unit 123 receives the voice of the answer given by the user from the voice interaction control unit 111. At this time, the specific procedure performed by the voice interaction control unit 111 is the same as that described in S505. Therefore, the description of the overlapping description is omitted here.

Next, the voice recording section 123 performs a recording process on the received voice. Then, the voice recording unit 123
Create a voice recording file for the recorded voice. Here, the voice recording unit 123 ends the voice recording process, and notifies the voice conversation control unit 111 of that.

The voice dialog device 100 of this embodiment is used.
In S, the processing of S501 and S502 described above, and S
It is preferable that the processes of 505 and S506 be performed in parallel.

Next, with reference to the flowchart shown in FIG. 6, the processing after S507 performed in the input processing will be described.

(S507) In the voice interaction control unit 111, the basic processing unit 115 receives the recognition result and the voice recording process end notification from the voice recognition unit 121 and the voice recording unit 123, respectively. Then, the basic processing unit 115 transmits the recognition result to the branch determination unit 117.

Upon receiving the recognition result, the branch judgment unit 117 reads the reference value from the memory 119. As described above, the reference value is the first threshold value, the first threshold difference, the second threshold value,
It is set as four reference values for the second threshold difference. Then, the branch determination unit 117 sets the respective thresholds and the respective threshold differences with respect to the likelihood of the first candidate and the difference between the likelihood of the first candidate and the likelihood of the second candidate among the plurality of candidates in the recognition result. Make a comparison.

Here, it is preferable that the difference between the likelihood of the first candidate and the likelihood of the second candidate and each threshold difference be compared based on the following judgment. That is, the larger the difference between the likelihoods of the first candidate and the second candidate is, the higher the reliability of the first candidate for determining the input voice is determined to be.

Further, in this step, the dialogue cell mode 203 of the dialogue cell is changed by the branch judgment unit 117.
It is read from the storage unit 107 into the memory 119.

Then, the branch determination unit 117 makes a branch determination based on the comparison between each reference value and the likelihood as described above or based on the interactive cell mode 203.

The branch judgment by the branch judgment unit 117 is
Regarding the recognition result, (a) use it as it is or (b)
Confirmation process, (c) Re-input process, (d) Save voice recording file only, (e)
It is preferable to judge whether to discard the recognition result. Details of the branch determination performed by the branch determination unit 117 will be described later.

Next, the processing of S508 to S512 performed as a result of the judgments of (a) to (e) will be described.

(S508) The branch judgment unit 117 (a)
When it is determined that the recognition result is used as it is, the voice interaction control unit 111 uses the first candidate as it is. Then, the input process ends.

(S509) The branch judgment unit 117 determines (b)
When it is determined that the confirmation processing is to be performed, the voice interaction control unit 111 performs the confirmation processing. Details of the confirmation processing will be described later.

(S510) The branch judgment unit 117 determines (c)
When it is determined that the re-input process is performed, the voice interaction control unit 111 performs the re-input process. Details of the re-input process will be described later.

(S511) The branch judgment unit 117 (d)
If it is determined that the voice recording file is to be stored, the basic processing unit 115 in the voice interaction control unit 111 is determined.
Uses the pointer to the voice recording file as the recognition result. Then, the basic processing unit 115 stores the voice recording file in the database unit 135.

The pointer to the voice recording file is
For example, in the flowchart shown in FIG.
The file name of the voice recording file created by the process of FIG. And to substitute this pointer,
For example, in the voice interaction control unit 111, the basic processing unit 11
5 means that the recognition result is discarded, the file name of the voice recording file is used instead of the recognition result, and the input process is terminated.

Further, it is preferable that the voice recording file stored in the database unit 135 is taken out and a confirmation work is performed if necessary. This confirmation work is performed by a desired means such as the above-mentioned listening and raising or voice recognition processing in the voice recognition unit 121.

(S512) The branch judgment unit 117 determines (e)
If it is determined that the recognition result is discarded, the voice interaction control unit 111 determines that there is no recognition result and ends the input process.

4. Branching Determination In the flowchart of FIG. 6 described above, the process of S507 will be described with reference to FIG. 11 and the flowcharts shown in FIGS. 7 and 8.

The process of S507 is a process performed by a branch determination. FIG. 11 divides the procedure of this branch judgment into five cases (A) to (F), and in each case, among the judgments (a) to (e) described above,
It shows the correspondence between which judgments are made. Here, in FIG. 11, FIG. 7, and FIG. 8, the likelihood of the first candidate and the likelihood of the second candidate are referred to as likelihood (1) and likelihood (2), respectively. Further, in FIG. 11, the likelihood difference is
The difference between the likelihood of the first candidate and the likelihood of the second candidate is shown.

First, the processing from S601 to S607 in the flowchart shown in FIG. 7 will be described.

The branch judgment unit 117 receives the recognition result from the basic processing unit 115 and reads the reference value from the memory 119. Furthermore, the branch determination unit 117 reads the interactive cell mode 203 from the storage unit 107 into the memory 119. Then, the branch determination unit 117 starts the branch determination. In addition, the interactive cell mode 203 is configured.
The plurality of flag strings in (e) are as described above.

(S601), (S602) Branch decision unit 1
If the likelihood of the first candidate is larger than the first threshold value in S601 and the difference in likelihood between the first candidate and the second candidate is larger than the first threshold value difference in S602, the first candidate is left unchanged. The decision (a) to use is made.

Referring to FIG. 11, (S601) and (S
In the process of 602), when the determination of (a) is performed, it corresponds to (A).

(S603) Next, in cases other than (A),
And the flag string in the interactive cell mode 203 is (a)
And ((b) or (c)), the branch determination unit 11
7 performs the processing of S604.

Here, the fact that the flag string is (a) and ((b) or (c)) means that there is a branch to the next dialogue cell according to the voice recognition processing, Alternatively, it is a case where the post-processing program 207 indicates that the recognition result is substituted into the global variable 215.

(S604) The branch determination unit 117 compares the likelihood of the first candidate with the first threshold value. Then, when the likelihood of the first candidate is larger than the first threshold value, the branch determination unit 117
The process of S605 is continuously performed.

(S605) The branch determination unit 117 compares the likelihood difference between the first candidate and the second candidate with the second threshold difference.
Then, when the difference between the likelihoods of the first candidate and the second candidate is larger than the second threshold difference, the branch determination unit 117 determines to perform confirmation processing on the first candidate (b).

Note that, unless the likelihood of the first candidate is larger than the first threshold value in (S604), the process proceeds to S606.

(S606) The branch determination unit 117 compares the likelihood of the first candidate with the second threshold, and the likelihood of the first candidate is the second threshold.
When it is larger than the threshold value, the process of S607 is continuously performed.

(S607) The branch determination unit 117 compares the difference in likelihood between the first candidate and the second candidate with the first threshold difference.
Then, when the difference between the likelihoods of the first candidate and the second candidate is larger than the first threshold difference, the branch determination unit 117 determines to perform confirmation processing on the first candidate (b).

Here, referring to FIG. 11, (S60
In the processes of 3) to (S607), when the determination of (b) is performed, it corresponds to (B).

Incidentally, referring to the flow chart shown in FIG. 7, the branch judging section 117 is
The determination (c) of re-inputting the first candidate is performed.
Referring to FIG. 11, the case classification at this time corresponds to (C).

Next, the processing from S608 to S612 will be described with reference to the flowchart shown in FIG.

(S608) In cases other than (C),
And the flag string in the interactive cell mode 203 is (a)
And (e), the branch determination unit 117, S609
Process. The flag strings (a) and (e) indicate that the post-processing program 207 is to increase the counter value of the global variable 215 according to the recognition result.

(S609) The branch determination unit 117 compares the likelihood of the first candidate with the first threshold value, and when the likelihood of the first candidate is larger than the first threshold value, performs the process of S610.

(S610) The branch determination unit 117 compares the likelihood difference between the first candidate and the second candidate with the second threshold difference.
Then, when the difference in likelihood between the first candidate and the second candidate is larger than the second threshold difference, the branch determination unit 117 makes a determination (a) to use the first candidate as it is.

Incidentally, in S609, unless the likelihood of the first candidate is larger than the first threshold value, the process proceeds to S611.

(S611) The branch determination unit 117 compares the likelihood of the first candidate with the second threshold value, and the likelihood of the first candidate is the second likelihood.
If it is larger than the threshold value, the process of S612 is continued.

(S612) The branch determination unit 117 compares the difference in likelihood between the first candidate and the second candidate with the first threshold difference. Then, when the difference between the likelihoods of the first candidate and the second candidate is larger than the first threshold difference, the branch determination unit 117 makes a determination (a) to use the first candidate as it is.

Here, referring to FIG. 11, (S60
In the processes of 8) to (S612), if the determination of (a) is performed, it corresponds to (D).

Then, referring to the flow chart shown in FIG. 8, in cases other than (D), the branch judging section 117 makes a judgment (e) of discarding the recognition result and proceeding to the next process. In FIG. 11, this case corresponds to (E).

Further, in FIG. 8, in the case other than (E), the judgment (d) of discarding the recognition result, substituting the voice recording file as the recognition result, and storing it in the database part is
This is performed by the branch determination unit 117. In this case,
1 is divided into cases (F).

With respect to the procedure of the branch determination performed by the branch determination unit 117 described above with reference to FIGS. 11, 7 and 8, for example, the recognition result is the first candidate: “30s”, the first candidate.
Candidate Likelihood 0.987; Second Candidate: "40's", Second Candidate Likelihood 0.765; Third Candidate: "10's", Third Candidate Likelihood 0.543 Think

Here, with respect to the reference values stored in the memory 119, the first threshold is 0.800 and the second threshold is 0.
600, the first threshold difference is 0.200, the second threshold difference is 0.1
00 is set.

At this time, the likelihood of the first candidate and the difference between the likelihoods of the first candidate and the second candidate are 0.987, 0.
222. Therefore, the likelihood of the first candidate is larger than the first threshold value 0.800, and the difference in likelihood between the first candidate and the second candidate is larger than the first threshold value difference. That is, at this time, in the processing of S601 and S602 in the flowchart shown in FIG. 7, the branch determination unit 117 makes the determination (a). This corresponds to the case (A) already described.

Here, the likelihood of the first candidate and the first candidate
Although it has been described that the difference between the likelihoods of the candidate and the second candidate is compared with each threshold value and each threshold value difference to make the branch judgment, the likelihood of each candidate and the difference between each likelihood and each threshold value are described. , And each threshold difference may be compared.

5. Confirmation Input Process Next, the process of S509 in the flowchart shown in FIG. 6, that is, the confirmation input process will be described with reference to the flowchart shown in FIG.

Here, as described above, in the process of S509 in the flowchart shown in FIG.
If 17 determines that the confirmation processing is to be performed, the confirmation input processing is started in the voice interaction control unit 111.

(S701) The basic processing unit 115 transfers the confirmation voice message 131 from the storage unit 107 to the memory 119.
Read to. Subsequently, the basic processing unit 115 adds the confirmation voice message 131 to the first candidate of the recognition result. Then, the confirmation voice message added to the first candidate is transmitted to the user's telephone 101 by the basic processing unit 115 via the line processing unit 109.

Here, the confirmation voice message given to the first candidate confirms whether the first candidate in the recognition result confirms the inputted voice.
For example, if the first candidate in the recognition result is "30s", the confirmation voice message given to this first candidate is "Are you in your 30s? Please answer yes or no." Becomes That is, in the confirmation process performed in this embodiment, the answer from the user is "yes" or "no".

(S702) The telephone set 101 receives the message from the line processing unit 109 and reproduces it.

(S703) In the voice interaction control section 111, the basic processing section 115 determines "Yes" or "No".
The confirmation recognition grammar 129 for recognizing only is read from the storage unit 107 into the memory 119. Then, the basic processing unit 115
Sends the confirmation recognition grammar 129 to the voice recognition unit 121 and notifies that the voice recognition is started. Speech recognition unit 1
21 uses the confirmation recognition grammar 129 to start the voice recognition process.

(S704) After that, the voice dialogue control unit 11
1 is in the waiting state for the result from the voice recognition unit 121. Here, the result waiting state is a state of waiting for the recognition result to be transmitted from the voice recognition unit 121.

(S705) In S702, the telephone set 1
The user answers by voice to the message reproduced in 01. The telephone 101 transmits the voice of the user to the voice interaction device 100.

(S706) The line processing unit 109 receives the voice of the user from the telephone 101. Subsequently, the basic processing unit 115 receives the voice of the user from the line processing unit 109. The basic processing unit 115 transmits the voice of the user to the voice recognition unit 121.

Upon receiving the voice of the user, the voice recognition unit 121 performs voice recognition on this voice. The procedure of voice recognition is the same as that already described, except that the confirmation recognition grammar 129 is used. Then, the determined recognition result is transmitted from the voice recognition unit 121 to the basic processing unit 115.
Will be notified. Notification of the determined recognition result is called result notification.

For example, a voice response from the user is
If "yes", the recognition result determined by the voice recognition unit 121 is as follows: first candidate: "yes", first candidate likelihood 0.990.

(S707) Upon receiving the recognition result, the basic processing section 115 sends it to the branch determination section 117.
If the first candidate of the recognition result is “No”, the branch determination unit 117 determines to perform the process of S708, that is, the re-input process. A detailed description of the process of S708 will be given later.

On the other hand, if the first candidate of the recognition result is "yes", the branch determination section 117 determines to end all the confirmation input processing. At this time, in the voice interaction control unit 111, the first candidate of the recognition result is used as it is and processed.

6. Re-input process Here, the process of S708 in the flowchart shown in FIG. 9, that is, the re-input process will be described with reference to the flowchart shown in FIG. The process of S510 in the flowchart of FIG. 6 is the same as the re-input process described below, and is performed by the same procedure.

(S801) First, the voice dialogue control section 111.
In the basic processing unit 115, the basic processing unit 115
3 is read from the storage unit 107 to the memory 119. Then, the re-input message is transmitted from the basic processing unit 115 to the user's telephone 101 via the line processing unit 109.

The re-input message prompts the user to re-input the voice on the telephone 101, for example, "Please say your answer again."

(S802) The telephone set 101 receives the message from the line processing unit 109 and reproduces it.

(S803) The user replies by voice to the re-input message from the telephone set 101. The telephone 101 transmits the voice of the user to the voice interaction device 100.

(S804) The line processing unit 109 receives the voice of the user from the telephone 101. Then, the basic processing unit 115 receives the voice of the user from the line processing unit 109. After that, the voice dialogue control unit 111 performs an input process for the voice of the user. This input processing is performed by the same procedure as that already described.

After the input processing is completed, the operation of this embodiment moves to the processing of S406 and thereafter as already described in the flowchart shown in FIG. Here, the description of the overlapping description will be omitted.

As described above, according to the voice dialogue processing of this embodiment, each dialogue cell 2 in the voice dialogue sequence 125 is described.
The processing of 01 is repeated. Then, in the processing of the interactive cell, the branch determination unit 117 determines that the interactive cell mode 2
03, and based on the comparison between the likelihood in the recognition result and the reference value, the branch judgment is performed on the input voice. Then, according to this branch determination, a process suitable for the recognition result is performed. That is, according to this embodiment, the most suitable voice dialogue processing is performed in each dialogue cell.

Therefore, in the present embodiment, in some cases, the manual work may be omitted in contrast to the recognition result which conventionally requires the manual raising work.
Further, even when the voice recognition is not successfully performed due to various reasons, a suitable process is determined and performed by the branch determination.

Further, by comparing the likelihood in the recognition result with the reference value, the branch judgment unit 117 can evaluate the recognition performance for the voice recognition. That is, it can be said that the branch determination unit 117 makes an accurate branch determination of the recognition result in consideration of the device configuration of the voice processing unit 113.

[0166]

As described above, according to the voice interaction apparatus and the voice interaction processing method of the present invention, the branch determination section in the voice interaction control section makes a branch determination with respect to the recognition result for the input voice, and the most suitable processing is performed. To decide. At this time, the branch determination unit can evaluate the recognition performance of the voice recognition by comparing the reference value and the likelihood. That is, the branch determination unit makes an accurate branch determination regarding the recognition result in the voice processing unit. Therefore, according to the voice interaction device and the voice interaction processing method of the present invention, it is possible to realize efficient and highly reliable voice interaction processing.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a voice dialogue device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a voice dialogue sequence.

FIG. 3 is a flowchart showing a voice dialogue process in this embodiment.

FIG. 4 is a flowchart showing a process of each dialogue cell according to this embodiment.

FIG. 5 is a flowchart showing an input process according to this embodiment.

FIG. 6 is a flowchart showing an input process according to this embodiment.

FIG. 7 is a flowchart showing a process in branch determination according to this embodiment.

FIG. 8 is a flowchart showing a process in branch determination according to this embodiment.

FIG. 9 is a flowchart showing a confirmation process according to this embodiment.

FIG. 10 is a flowchart showing a re-input process according to this embodiment.

FIG. 11 is a diagram for explaining a branch determination in a branch determination unit according to this embodiment.

[Explanation of symbols] 100: Spoken dialogue device 101: User's telephone 103: communication line 105: voice interaction processing unit 107: storage unit 109: Line processing unit 111: Spoken dialogue control unit 113: Voice processing unit 115: Basic processing unit 117: Branch judgment unit 119: Memory 121: Speech recognition unit 123: Voice recording section 125: Spoken dialogue sequence 127: Voice message 129: Confirmation recognition grammar 131: Confirmation voice message 133: Re-enter message 135: Database part 137: Voice processing information 201: Dialog cell 203: Interactive cell mode (3rd information) 205: Preprocessing program 207: Post-processing program 209: Next cell pointer 211: Voice file name (first information) 213: Recognition grammar (second information) 215: Global variable 217: Second information group 219: First information group

Continued front page    (72) Inventor Shinji Hayakawa             1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric             Industry Co., Ltd. F-term (reference) 5D015 KK02 KK04 LL10                 5D045 AB04 AB26                 5K027 BB01 HH20

Claims (18)

[Claims] 1. A voice interactive apparatus connected to a user's telephone via a communication line, receiving voice from the telephone, and performing voice interactive processing, wherein the voice is fixed for the voice input from the telephone. A plurality of candidates for determining the likelihood of each candidate as a recognition result, and by comparing the likelihood with a reference value set in advance for use in making the determination, the branch determination A voice interaction apparatus, comprising: a voice interaction processing unit that determines a method of performing the recognition result. 2. The voice interactive apparatus according to claim 1, wherein a first information group including first information and second information used for determining the recognition result, and a third information group used for the branch determination. A voice interactive apparatus comprising a storage unit in which voice processing information having a second information group including information is stored in advance. 3. The voice interaction device according to claim 2, wherein the voice interaction processing unit stores a line processing unit connected to the telephone, a basic processing unit, a branch determination unit, and the reference value. And a voice processing unit that determines the recognition result, wherein the basic processing unit includes the first information and the second information,
Reading from the storage unit into the memory, using the first information, performing a voice conversation with the telephone through the line processing unit, and receiving a voice from the telephone, the voice processing unit, The voice from the telephone and the second information are received from the basic processing unit, the recognition result is determined using the second information, and the recognition result is transmitted to the basic processing unit. The judgment unit receives the recognition result from the basic processing unit, reads the third information from the storage unit, and the reference value from the memory, respectively, and compares the likelihood with the reference value, And a voice interaction device that makes the branch determination based on the third information. 4. The voice interaction device according to claim 3, wherein the voice processing unit performs a recording process on a voice recognition unit for determining the recognition result, and a voice recording process for voice from the telephone. A voice interaction device, comprising: a voice recording unit for creating a recording file. 5. The voice interaction device according to claim 4, wherein the voice processing information stored in advance in the storage unit includes a voice interaction sequence having a plurality of interaction cells, and the interaction cell is , The first information group, and the second information group including a dialogue cell mode as the third information for determining processing of the dialogue cell, wherein the branching determination includes the likelihood and the criterion. A voice interaction device characterized by being compared with a value and based on the interaction cell mode. 6. The voice interactive apparatus according to claim 5, wherein the voice processing information includes a voice message having a plurality of files in which a voice message is recorded, and the first information group includes: A file used when performing the voice interaction with the telephone is a voice file name as the first information that specifies from the plurality of files, and a second file that is used when determining the recognition result. And a recognition grammar. The second information group describes, in addition to the interaction cell mode, a preprocessing program that describes a preprocessing required for the voice interaction processing, and a postprocessing required for the voice interaction processing. A post-processing program, and a next cell pointer for designating a dialogue cell to proceed to after the processing in the dialogue cell is completed. A ram, and the postprocessor, voice dialogue system, characterized in that depending on said next cell pointer. 7. The voice interactive apparatus according to claim 6, wherein the voice interactive sequence has a global variable set corresponding to the interactive cell, and the interactive cell mode has a plurality of global variables. (B) whether the dialogue cell mode includes voice recognition processing, (b) whether there is a branch to the next dialogue cell in the dialogue cell, C) Whether the post-processing program substitutes the recognition result into the global variable, and (d) Whether the post-processing program registers the recognition result in the database unit installed in the storage unit. Or (e) The post-processing program performs a process of increasing the global variable or the counter value in the database unit for the recognition result. Voice dialogue system, characterized in that illustrates how a. 8. The voice interaction device according to claim 7, wherein the branch determination unit uses the result of the branch determination as it is, discards it, or performs confirmation processing,
A voice interactive apparatus, characterized in that either a re-input process is performed or the voice recording file is substituted for the recognition result and stored in the database unit. 9. The voice interactive apparatus according to claim 8, wherein the recognition result is used as it is, discarded, confirmation processing, re-input processing, or the voice recording file is recorded. The reference value used when determining whether to substitute the recognition result and store it in the database unit is four reference values of a first threshold value, a second threshold value, a first threshold value difference, and a second threshold value difference. A voice interaction device characterized by being set as. 10. The voice interaction device according to claim 9, wherein the branch determination unit determines the likelihood of a first candidate among the plurality of candidates, and the first candidate and the second candidate in the branch determination. (A) the likelihood of the first candidate is larger than the first threshold,
When the difference in likelihood is larger than the first threshold difference, it is determined that the first candidate is used as it is. (B) In cases other than (A), and the flag string in the interactive cell mode is For the dialogue cell, it indicates that there is a branch to the next dialogue cell according to the voice recognition processing, or for the post-processing program, indicates that the recognition result is assigned to the global variable. When the likelihood of the first candidate is larger than the first threshold and the difference in likelihood is larger than the second threshold difference, or the likelihood of the first candidate is larger than the second threshold. And, when the difference in the likelihood is larger than the first threshold difference, a determination is made to confirm the first candidate, and in the cases other than (C) and (B), the re-input process is performed for the first candidate. In the cases other than (D) and (C), the flag string in the interactive cell mode increases the counter value of the global variable according to the recognition result in the post-processing program. When indicating that to perform, if the likelihood of the first candidate is larger than the first threshold and the difference in the likelihood is larger than the second threshold difference, or the likelihood of the first candidate is When it is larger than the second threshold and the difference in likelihood is larger than the first threshold difference, it is determined to use the first candidate as it is, and (E) and (D), the recognition result is Discard and determine to proceed to the next process. If (F) or (E) is not true, the recognition result is discarded, and the voice recording file is substituted as the recognition result and stored in the database unit. Spoken dialogue apparatus according to claim. 11. When performing a voice interaction process in a voice interaction device including a voice interaction processing unit connected to a user's telephone via a communication line, a plurality of voices for determining a voice input from the telephone are set. A branch decision is made by determining a candidate and the likelihood for each candidate as a recognition result, and comparing the likelihood with a preset reference value for use in making the decision. A method for voice interaction processing, characterized in that how to process the recognition result is determined by. 12. The voice interaction processing method according to claim 11, wherein the voice processing information having a first information group including first and second information and a second information group including third information is stored in advance in the storage unit. And storing the first information in the storage unit, the first and second information is read from the storage unit, the recognition result is determined by using the first and second information, and the third information is read from the storage unit. A voice interaction processing method, characterized in that the branch determination is performed using the third information. 13. The voice interaction processing method according to claim 12, wherein the first and second information are read from the storage unit to a memory, and the line processing unit is used by using the read first information. The second information read out to the memory by performing a voice dialogue between the basic processing unit and the telephone via the line processing unit and receiving a voice from the telephone via the line processing unit. Is used to determine the recognition result, read the third information from the storage unit, read the reference value stored in advance in the memory, compare the likelihood with the reference value, and A voice interaction processing method, characterized in that a previous branch decision is made based on the third information. 14. The voice interaction processing method according to claim 13, wherein a voice recording unit receives voice from the telephone,
A voice interaction processing method characterized by performing a voice recording process to create a voice recording file. 15. The speech dialogue processing method according to claim 14, wherein the speech processing information stored in advance in the storage unit includes a speech dialogue sequence having a plurality of dialogue cells, and the dialogue cell is the first dialogue cell. 1 information group and the 2nd information group containing the conversation cell mode as the 3rd information which determines the processing of the conversation cell, and the above-mentioned branch judgment of the above-mentioned likelihood and the above-mentioned reference value. A voice interaction processing method, characterized in that the comparison is performed based on the interaction cell mode. 16. The voice interaction processing method according to claim 15, wherein the voice processing information includes a voice message having a plurality of files in which voice messages are recorded, and the first information group includes: A voice file name as the first information that specifies a file used when performing the voice dialogue between the basic processing unit and the telephone, and when determining the recognition result. The second information group includes a recognition grammar as the second information to be used, the second information group includes a preprocessing program describing preprocessing required for the voice interaction processing, in addition to the interaction cell mode, and the voice interaction. A post-processing program that describes post-processing required for processing, and a next cell pointer that specifies a dialogue cell to proceed to after the processing in the dialogue cell is completed. The cell mode depends on the pre-processing program, the post-processing program, and the next cell pointer. 17. The voice dialogue processing method according to claim 16, wherein the voice dialogue sequence has a global variable set corresponding to the dialogue cell, and the dialogue cell mode is plural. And (b) whether or not the dialogue cell mode includes voice recognition processing, (b) whether or not there is a branch to the next dialogue cell in the dialogue cell, (C) Whether the post-processing program substitutes the recognition result into the global variable, or (D) Whether the post-processing program registers the recognition result in the database unit installed in the storage unit. Or (e) the post-processing program increases the counter value in the global variable or the database unit for the recognition result. Speech dialogue processing method characterized by whether or not to, shows the. 18. The voice interaction processing method according to claim 17, wherein the reference value is a first threshold value, a second threshold value, a first threshold value difference, or a second threshold value.
It is set as four reference values of a threshold difference, and the branching determination is performed on the likelihood of the first candidate among the plurality of candidates and the difference between the likelihoods of the first candidate and the second candidate. ) The likelihood of the first candidate is greater than the first threshold,
And when the difference in the likelihood is larger than the first threshold difference, a process of using the first candidate as it is is determined, and (B) in a case other than (A) and the flag string in the interactive cell mode is: For the dialogue cell, indicates that there is a branch to the next dialogue cell according to the voice recognition processing, or for the post-processing program, indicates that the recognition result is substituted into the global variable. When the likelihood of the first candidate is greater than the first threshold and the difference in likelihood is greater than the second threshold difference, or the likelihood of the first candidate is greater than the second threshold. If the difference is large and the difference in likelihood is larger than the first threshold difference, it is decided to perform confirmation processing for the first candidate, and in cases other than (C) and (B), re-input for the first candidate. place In the case other than (D) and (C), and the flag string in the interactive cell mode performs a process of increasing the counter value of the global variable according to the recognition result for the post-processing program. When the likelihood of the first candidate is larger than the first threshold and the difference in the likelihood is larger than the second threshold difference, or the likelihood of the first candidate is When the difference between the likelihoods is larger than the second threshold and the difference between the likelihoods is larger than the first threshold difference, a process of using the first candidate as it is is determined, and in a case other than (E) and (D), the recognition result is Discard and decide to proceed to the next process. In the case other than (F) and (E), the recognition result is discarded, the voice recording file is substituted as the recognition result, and the process of saving in the database unit is determined. This And a voice interaction processing method characterized by: