JPH1031497A - Voice conversation control method and voice conversation system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently conduct voice conversation between a user and a system by eliminating the system output which is for confirmation only and providing the recognition results including a neat guidance sentence. SOLUTION: A conversation control means 5 requests the guidance contents which urge the next action to a task control means 10. The response to the request is a question for a name. Then, the means 5 asks for a latest keyword to a keyword holding means and obtains the 'Material Section' which has been just POPed. Then, the means 5 transmits the next guidance contents and the latest keyword to an response generating means 7 and instructs it to generate a response sentence. Then, the means 7 generates a question sentence, such as 'What is your name in the Material Section?' and transmits the sentence to a voice output means 5. Then, the means 5 sends the recognition vocabulary to a recognition vocabulary supplementing means 3 in order to confirm the name.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a computer system used for performing information retrieval and the like, and more particularly, to a computer system provided with a microphone and a voice input / output interface such as a speaker or a telephone so that anyone can use it easily. The present invention relates to a voice interaction system.

[0002]

2. Description of the Related Art Generally, a speech dialogue system uses a speech recognition technique for data input to a user's system. Voice data input is very effective when used over a telephone or when used in a so-called hands-eye busy situation such as driving a car.

Normally, data is input for a plurality of items (for example, the affiliation and name of a partner in a telephone number search application, the date and time, adults or children, the number of sheets, etc. in a ticket reservation application). But on the other hand,
It is impossible to achieve a recognition rate of 100% in voice recognition, and it is essential to check the recognition result of the system for each input item for a user input. For example,
An example of dialogue in a voice dialogue system that can recognize sentences or phrases is as follows: System: "Please select the affiliation and name of the other party" User: "Mr. Sato of the Materials Division" System: "Sato of the Materials Division" User: Yes. As in this example, when recognition is correctly performed in a system capable of performing speech recognition in units of sentences or phrases, the input time is short and the use efficiency is improved. However, in the case of recognition by sentence or phrase, not only the number of types of recognition targets increases by the combination of a plurality of keywords (in this example, the affiliation and name of the partner), but also the variety of expressions, including particles, varies. , The recognition performance is lower than that of word recognition. Further, if the recognition is wrong, the following problem also occurs.

[0004] System: "Please choose the affiliation and name of the other party" User: "Mr. Sato of the Materials Division" System: "It's Kato of the Materials Division" User: "No" System: "Please choose the affiliation and name of the other party The first system question included two keywords: affiliation and name. In this case, the system alone can determine whether the user's rejection means that either the affiliation or the name has been misrecognized, or that both have been misrecognized. Can not. Thus, the same question is repeated until both the affiliation and name keywords are correctly recognized. In this case, there is a problem that it takes time. As another countermeasure, a method in which only the user who has been erroneously recognized may input again is also conceivable. In this case, you need to recognize utterances that do not know your affiliation or name,
In addition to the problem that the demand for the performance of speech recognition is higher than the level that can be currently dealt with, there is a very difficult problem of how to guide the utterance of the user.

On the other hand, in the case of word recognition, the variety of expressions to be recognized is suppressed, and almost satisfactory recognition performance can be obtained even at the current level. In addition, in order to ask and confirm items one by one, data can be input one item at a time. For example,

[0006] System: "Please add the other party's affiliation" User: "Materials Section" System: "Is the Materials Section?" User: "Yes" System: "Please enter the name of the other party" User: "Sato" System: "Is Sato?" User: "Yes" However, as shown here, there remains a major problem that the exchange becomes longer and the whole tends to take longer.

[0007]

In the above-mentioned conventional speech dialogue system, the current speech recognition performance and the temporal utilization efficiency are incompatible parameters.

[0008] It is an object of the present invention to provide a speech dialogue system capable of inputting data most efficiently within the range of currently available speech recognition performance and realizing a smooth dialogue between a user and the system. is there.

[0009]

According to the present invention, there is provided a dialogue control method using the following means and a voice dialogue system provided with the following means.

[0010] A request is issued to the task management means, and the response generation means, the recognized vocabulary supplementing means, the keyword determination means, and the keyword holding means are controlled in accordance with the response result, and the speech recognition means is controlled via the recognized vocabulary supplementing means. Further, under the control of the speech synthesizing means via the response generating means, and the dialog control means for proceeding the dialog between the system and the user, the keyword holding means, based on the request of the dialog control means, Holding, deleting, notifying of the latest keyword, the dialog control means requests the task management means for guidance content for prompting the next action in the progress of the dialogue, receives the guidance content, and sends the guidance content to the keyword holding means. Requesting the latest keyword and receiving the latest keyword, the guidance content and the latest keyword, and An instruction to generate a response sentence using them is sent to the response generation means, and the response generation means, in accordance with the instruction received from the interaction control means, includes a guidance sentence for prompting a next action in the progress of the interaction, At the same time, a response sentence including the latest keyword, which is also the recognition result of the previous stage received, is generated and output to the speech synthesis means. Then, the dialog control means sends the task management means the next recognized vocabulary in the progress of the dialog. , Receiving the recognition vocabulary, and sending the recognition vocabulary comprising task-dependent keywords to the recognition vocabulary supplementing means. The recognition vocabulary supplementing means adds “cancel”, “cancel” to the recognition vocabulary received from the dialog control means. Supplement words that represent commands that are independent of tasks, such as help, stop, repeat, don't know, or any Handover, and the keyword determination unit compares the supplemented recognized vocabulary obtained by the recognized vocabulary supplementing unit with the recognition result obtained by the voice recognition unit, and the recognition result is a command independent of the task. Or a keyword depending on the task, and sends the result of the determination to the dialogue control means. Next, the dialogue control means determines whether the keyword is a keyword based on the determination result of the keyword determination means. Sends the keyword to the keyword holding unit and sends the keyword to the task management unit, and if the determination result is a command, performs a process for each command.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of a voice dialogue system according to the present invention. The spoken dialogue system
It is a system in which the user and the system communicate using voice for the purpose of performing a certain task. The voice dialogue system according to the present invention operates under the control of the dialogue control means (5). The operation of the dialog control means (5) will be described later in detail.

The task management means (10) is in charge of all tasks dependent processing. It manages task-dependent information (task execution procedure, input waiting vocabulary for each scene, etc.) and responds to inquiries about the progress of tasks from the dialog control means.

The speech recognition means (2) recognizes the input speech (1) within the range of the given recognition vocabulary, and outputs the recognition result as one word or a sequence of a plurality of words. Various methods are conceivable as a method of realizing the voice recognition means (2), and the present invention does not limit the method. For example,
A method using a probabilistic model called HMM (Hidden Markov Model) is easy to handle. With this method, just giving any recognized vocabulary as text,
A configuration capable of voice recognition can be provided. The details are described in "Seiichi Nakagawa: Speech Recognition by Probabilistic Model, IEICE, 1988" and other documents, and are omitted in this specification.

The recognition vocabulary supplementing means (3) supplements the recognition vocabulary received from the dialog control means (5), especially words representing commands independent of tasks, and passes them to the speech recognition means. Here, the command is a word that means negation or cancellation, such as "no,""no,""no,""cancel," or "cancel," a word that means an advisory request, such as "help,""stop,""stop." There are words that mean a request to stop the processing of the system, such as "", "stop", and words, such as "again" and "repeat", that mean a request to re-output the immediately previous system response. Furthermore, words that can be supplemented by the recognized vocabulary supplementing means (3) include words that mean unknown, such as “I don't understand” or “I don't know”, and words that mean arbitrary, such as “I don't care”, “Anything is fine”, or “Any”. And FIG. 3 shows an example of the recognition vocabulary. The recognition vocabulary of the personal name is taken as an example, and the recognition vocabulary before (a) and after (b) the supplement is shown.

The keyword determining means (6) determines whether the recognition result obtained from the voice recognition means (2) is a command independent of the task or a keyword dependent on the task, and sends the result to the dialog control means. Here, the determination result is expressed, for example, as shown in FIG. In the example shown in FIG.
One determination result is represented by a combination of two values. The left side shows the type of the determination result, “CMD” represents a command, and “KW” represents a keyword. Also, "C
"CANCEL", "HE" combined with "MD"
"LP" indicates the type of command. Furthermore, "KW"
“Materials section” and “Sato” combined with are showing actual data values.

The keyword holding means (6) holds the passed keywords in a stack format or notifies the dialog control means of the keywords stored in the stack, according to the instruction of the dialog control means.

The response generation means (7) generates a response sentence which inquires of the contents (actual data values such as names) of the items necessary for performing the task in accordance with the instruction from the interaction control means (5). I do.

The voice synthesizing means (8) converts the response sentence obtained from the response generating means (7) into a voice waveform and converts the electric signal into a sound wave by a device such as a speaker. Output as

FIG. 2 is a diagram showing a processing flow of the dialog control means. For simplicity, the flow when only data input is performed in the voice interaction system according to the present invention is shown. In the actual task, presentation of only the result to the user is also included.

Next, the processing procedure will be described in accordance with the flow of FIG. 2 on the assumption that a telephone connection service or the like is a task and "materials section" is input as the affiliation name and "Sato" is input as the personal name.

The task management means is requested for guidance contents for prompting the next action in the progress of the dialog. First, when the service is started, the dialog control means (5)
It requests the task management means (10) for guidance contents for prompting the next action in the progress of the dialogue. The response to this request is obtained as the following guidance content is "affiliation question". Next, the dialog control means (5) tries to receive the latest keyword from the keyword holding means. Immediately after the start of use of the system, the keyword holding unit is empty, and the fact is notified to the dialog control unit. Next, the dialogue control means (5) sends the next guidance content ("Affiliation question") and the latest keyword ("None" at this stage) to the response generation means (7), and sends the response sentence. Instruct to generate. In the response generation means (7), since the guidance content is "question of affiliation", a question sentence such as "please belong" is generated and sent to the voice output means (8). Subsequently, the dialog control means (5) sends a recognized vocabulary for voice recognition of "belonging" to the recognized vocabulary supplementing means (3). The recognition vocabulary here is received together from the task management means (10) when the guidance content for prompting the next action is requested. Alternatively, an inquiry may be made to the task management means again at this stage to accept it. Thereafter, the recognized vocabulary supplemented by the recognized vocabulary supplementing means (3) is sent to the speech recognition means (2). Then, the recognition result of the user's utterance by the voice recognition means (2) is added with a determination as to whether it is a command or a keyword via the keyword determination means (4), and is performed in the form shown in FIG. 5)
Is returned to. According to the current assumption, the result of acceptance here is "K
W] + “Materials Section”. That is, the dialog control means (5) POPs the keyword on the stack of the keyword holding means (6) and further notifies the task management means (10) of the keyword. at this point,
The task management means (10) internally updates the task progress status.

Here, returning to the beginning of the flow of FIG. 2, the dialogue control means (5) requests the task management means (10) again for guidance contents for prompting the next action in the progress of the dialogue. The answer to this request is obtained as the next guidance content is "name question". Next, the dialogue control means (5) inquires the latest keyword from the keyword holding means, and obtains the "materials section" that has just been POP. Then, the dialogue management means (5) sends the next guidance content ("question of name") and the latest keyword ("materials section") to the response generation means (7).
And instruct it to generate a response sentence. In response generation means (7), include "material section" and "name"
, Such as "What is the name in the material section?" Or "Who is the material section?", And sends it to the voice output means (8). Subsequently, the interaction control means (5) sends a recognition vocabulary for voice recognition of the "name" to the recognition vocabulary supplementing means (3). Further, the result of recognition by the voice recognition means (2) is added with a judgment as to whether it is a command or a keyword through the keyword judgment means (4), and is returned to the dialogue control means (5) in a format as shown in FIG. It is. Under the current assumption, the result of acceptance here is “KW” + “Sato”. By repeating the above, the basic dialogue proceeds.

Next, an example will be described in which it is assumed that the user's utterance of the "material section" is erroneously recognized and the recognition result is "the facility section". It is assumed that the system output is "Who is in the facility section" instead of "Who is in the material section", and the user answers "No" to the system output. The result received from the keyword determination means (4) is not "KW" + "Sato" but "CMD" + "CA
NCEL ". In this case, the dialogue control means (5) instructs the keyword holding means (6) to POP one keyword (in this case, “materials section”). Furthermore, the task management means (10) is notified to cancel the POP "material section". Here, the processing of the dialogue control means (5) returns to the beginning of the flow in FIG. 2, and inquires of the task management means (10) and the keyword holding means (6), and the guidance content for prompting the next action is "belonging Is sent to the response generating means (7), and the response generating means (7) again generates a question sentence such as "Please belong."

Finally, an example of a dialogue when the voice dialogue system according to the present invention is used will be described. This is an example where the task is a telephone connection service.

System: “Please add the other party's affiliation” User: “Materials Section” System: “Who is in the Materials Section?” User: “Sato” System: “Call Sato-san” Here is an example of a dialogue in the case of misrecognition in the second item.

[0027] System: "Please go to the other party's affiliation" User: "Materials Section" System: "Who is the facility section?" User: "No," System: "Please try again to belong to the other party" User: "Materials Section" System: "Who is the Materials Section?": (Omitted hereafter) Next, an example of dialogue when misrecognition is made in the second and subsequent items is shown.

: (Some part omitted) System: "Who is in the materials section?" User: "Sato" System: "Connects the phone to Mr. Kato" User: "Is different" System: "Again, Materials section User: "Sato" System: "Connect Sato-san" In Figure 1, only audio is drawn as media between the user and the system, but characters, images, etc. A dialog system including other media may be used. Alternatively, a button input may be prepared, and a button such as a cancel button may be input. When using from a remote telephone, a touch tone signal may be used for cancellation or the like.

The task management means in FIG. 1 also includes processing specific to a task. For example, in the case of using the system from a remote telephone, control of a telephone line may be considered.

[0030]

According to the present invention, in a voice dialogue system for inputting data of a plurality of items by voice, a system output of only verification is omitted for a required confirmation when using voice recognition, and the recognition result is set as follows. By reducing the number of conversations by presenting it in a guidance sentence, the user and the system can proceed efficiently with a dialogue even with a spoken dialogue system that has only a speech recognition capability equivalent to word speech recognition. The effect is obtained.

[Brief description of the drawings]

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

FIG. 2 is a flowchart illustrating a processing procedure of a dialog control unit.

FIG. 3 is a diagram showing an example of a recognized vocabulary.

FIG. 4 is a diagram showing an example of a data structure passed from a keyword determination unit to a dialogue control unit.

FIG. 5 is a diagram illustrating an example of data stored in a keyword storage unit.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... User's voice, 2 ... Voice recognition means, 3 ... Recognition vocabulary supplement means, 4 ... Keyword judgment means, 5 ... Dialogue control means,
6 ... Keyword holding means, 7 ... Response generation means, 8 ... Speech output means, 9 ... System output speech, 10 ... Task management means.

Continued on the front page (72) Inventor Nobuo Hataoka 1-280 Higashi Koigakubo, Kokubunji-shi, Tokyo Inside Central Research Laboratory, Hitachi, Ltd.

Claims (18)

[Claims] A dialogue is provided by using guidance contents for prompting a next action in the progress of a dialogue managed by a task management means and the latest keyword which is a recognition result of a previous stage held by a keyword holding means. A response sentence including the latest keyword is generated and output by the response generating unit in the guidance sentence for prompting the next action in the progress of the process, and the task is managed by the task managing unit in the recognition vocabulary supplementing unit. A word representing a command independent from the task is supplemented to the recognition vocabulary in the next scene in the progress of the dialogue, and within the range of the recognition vocabulary after the supplementation, the speech recognition means recognizes the speech uttered by the user. The keyword determination means determines whether the recognition result is a command independent of the task or a keyword dependent on other tasks. If the judgment result is a command,
The process for each command is performed, and when the result of the determination is the keyword, the keyword that is the result of the determination is newly held in the keyword holding unit as the latest keyword. A voice dialogue control method, characterized in that the voice is advanced by notifying the management means and repeating the above. 2. The method according to claim 1, wherein the command includes at least a word meaning negation or cancellation, such as "no,""no,""no,""cancel," or "cancel." If the result of the determination is a word meaning the negation or cancellation, the keyword holding means is instructed to delete one of the latest keywords, and at the same time, the content of the immediately preceding request item is instructed to the task management means. A voice dialogue control method, wherein a notification of cancellation is provided. 3. The voice dialogue control method according to claim 1, wherein the command includes at least a word meaning an advice request such as "help", and the determination result is a word meaning the advice request. And a response generation means for outputting a message about the usage. 4. The voice interaction control method according to claim 1, wherein the command includes at least a word indicating a request to stop processing of the system such as “stop”, “stop”, “stop”, “stop”, and the like. A speech dialogue control method, wherein the dialogue progress is stopped when the determination result is a word meaning the stop request. 5. The voice dialogue control method according to claim 1, wherein said response generation means has means for storing one or more response sentences, which can be referred to later, and wherein said command is at least “again” or “repeat”. If the determination result is a word meaning the re-output request including a word meaning a re-output request of the immediately preceding system response, the response generation means may re-output the response sent immediately before. A voice interaction control method characterized by the following. 6. The method according to claim 1, wherein the command is at least "I do not know."
Including a word meaning unknown such as "I don't know", and when the determination result is a word meaning unknown, notifies the task management means that the content of the immediately preceding request item is unknown, At the same time, request the task management means for guidance that prompts the next action in the progress of the dialogue,
A spoken dialogue control method characterized in that the dialogue proceeds according to the contents of the response. 7. The voice interaction control method according to claim 1, wherein said command includes at least a word meaning an arbitrary word such as "don't care", "anything", "any", and said judgment result indicates that said arbitrary word is arbitrary. If it is a meaningful word, notify the task management means that the content of the request item is optional, and at the same time, request the task management means for guidance content that prompts the next action in the progress of the dialogue, and the response content A dialogue control method characterized in that the dialogue proceeds according to the following. 8. A request for providing information related to the progress of a task-dependent dialogue, such as guidance content for managing the progress of a task-dependent dialogue and prompting the next action in the progress of the dialogue. A task management means for responding information to the request, a recognition vocabulary supplementing means, a keyword determination means, a keyword holding means, and a response generation means in response to a request issued to the task management means and the response result. Dialogue control means for promoting a dialogue between a user and a user; a recognition vocabulary supplementary means for supplementing a new word to the recognition vocabulary received from the dialogue control means and passing the supplemented result to a speech recognition means; A speech recognition method for recognizing a speech uttered by a user and outputting one or more word sequences within the range of the recognition vocabulary specified by the output of the vocabulary supplementing means. And a keyword determining unit for determining the one or more word sequences; holding a keyword passed from the dialog control unit; deleting a latest keyword from the held keywords; or a dialog control unit A response generating means for generating a response sentence using data from the dialog control means in accordance with an instruction from the dialog control means, and in some cases, obtained from the response generating means. And speech synthesis means for converting the response sentence into a speech waveform and outputting the speech. In particular, the dialogue control means requests the task management means for guidance content for prompting the next action in the progress of the dialogue. Requesting the latest keyword from the keyword holding means and receiving the latest keyword, Wherein the volume latest keywords, and notifies the response generation means an instruction to generate a response sentence by using them, the response generation means,
According to the instruction received from the interaction control means, a response sentence including the latest keyword, which is also a recognition result of the previous stage received at the same time, is generated in the guidance sentence for prompting the next action in the progress of the dialogue, and speech synthesis is performed. The dialogue control means requests the task management means for the next recognition vocabulary in the course of the dialogue, receives the recognition vocabulary, and supplements the recognition vocabulary consisting of task-dependent keywords with the recognition vocabulary supplement. Means, the recognition vocabulary supplementing means supplements the recognition vocabulary received from the dialog control means with a word representing a command independent of a task, passes the word to the speech recognition means and the keyword determination means, and passes the word to the keyword recognition means. Compares the supplemented recognition vocabulary obtained by the recognition vocabulary supplementing means with the recognition result obtained by the speech recognition means, and the recognition result is A command independent of the task,
It is determined whether the keyword depends on the task, the determination result is sent to the dialog control means, and further, the dialog control means, based on the determination result of the keyword determination means, if the determination result is a command, A voice dialogue system which performs a process for each command and, when the determination result is a keyword, sends the keyword as the determination result to the keyword holding means and simultaneously sends the keyword to the task management means. 9. The speech dialogue system according to claim 2, wherein the command includes at least a word meaning negation or cancellation such as "no,""no,""no,""cancel," or "cancel." If the determination result is a word meaning the negation or cancellation, the dialogue control means instructs the keyword holding means to delete one newest keyword, and at the same time, issues a request to the task management means. A voice dialogue system for notifying that the content of an item has been canceled. 10. The voice interaction system according to claim 2, wherein said command includes at least a word meaning an advice request such as "help", and the result of judgment by said keyword judgment means is a word meaning the advice request. Wherein the dialogue control means instructs the response generation means to provide a help response, and the response generation means outputs a message such as usage in accordance with the instruction. 11. The voice interaction system according to claim 2, wherein the command includes at least a word indicating a request to stop processing of the system, such as "stop", "stop", "stop", "stop", and the keyword. If the result of the determination by the determination means is a word meaning the stop request, the dialogue control means stops the dialogue progress. 12. The speech dialogue system according to claim 2, wherein said response generating means has means for storing one or more response sentences, which can be referred to later, and wherein said command is at least "again" or "repeat". When the result of the keyword determination means is a word meaning the re-output request, the dialog control means outputs the word immediately before to the response generation means. A spoken dialogue system that issues an instruction to re-output a response sentence. 13. The voice interaction system according to claim 2, wherein said command is at least "I do not understand".
If the keyword includes a word meaning unknown, such as "I don't know," and the result of the determination by the keyword determination means is a word meaning the unknown, the dialog control means sends the content of the immediately preceding request item to the task management means. Notify that is unknown,
At the same time, a voice dialogue system which requests the task management means for guidance content for prompting the next action in the progress of the dialogue, and proceeds with the dialogue according to the response content. 14. The voice dialogue system according to claim 2, wherein the command includes at least a word meaning an arbitrary word such as “don't care”, “anything”, “any”,
When the result of the determination by the keyword determination means is a word meaning the arbitraryness, the dialogue control means notifies the task management means that the content of the request item is optional, and at the same time, notifies the task management means of the dialogue. A spoken dialogue system that requests guidance content for prompting the next action in the process, and performs dialogue according to the response content. 15. The method of claim 8, 9, 10, 11, 12, 1.
15. A voice interaction system in which the contents of the tasks described in 3 and 14 are a telephone number search service or a telephone line connection service. 16. The method of claim 8, 9, 10, 11, 12, or 1.
15. A spoken dialogue system in which the contents of the tasks described in 3 and 14 are a transportation service, an entertainment or facility search service, or a reservation service. 17. The method of claim 8, 9, 10, 11, 12, 1.
15. The voice interaction system according to items 3 and 14, wherein the content of the task is a product search service, a sales service, or a support service. 18. The method of claim 8, 9, 10, 11, 12, 1.
15. A voice interaction system in which the contents of the tasks described in 3 and 14 are a route information providing service or a route guidance service.