JP2009198871A - Voice interaction apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice interaction apparatus capable of continuing interaction without causing failure when reliability of verb in the results of recognition of utterance related to user's action (verb) is low. <P>SOLUTION: This voice interaction apparatus is provided with an inputting part 11 for inputting utterance, a first utterance analyzing part 12 for analyzing the input utterance, a verb determining part 13 for determining whether a verb is included in the results of analysis or not, a verb reliability determining part 14 for determining whether the verb is reliable or not when it is determined that the verb is included in the results of analysis, a response generating part 15 for generating a response to the utterance by using a response template fixed in advance and the verb when it is determined that the verb is reliable, an emotion question generating part 16 for generating a question for asking about emotion corresponding to the contents of the utterance by using a question template fixed in advance, and an outputting part 22 for outputting the generated response and question. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a voice dialogue apparatus and a voice dialogue program that are effective when a large amount of user behavior is input as user utterances.

Technologies such as a chat dialogue system and a listening dialogue system have been proposed in which a user utterance is recognized by speech and a response corresponding to the recognition result is output to interact with the user. As such a technique, for example, a threshold is set for the reliability of the speech recognition result for the user utterance, and if the reliability is equal to or higher than the threshold, a response corresponding to the recognition result is output, and if the reliability is lower than the threshold, recognition is performed. A voice interaction device that rejects a result and outputs a response described in an autonomous response dictionary (see, for example, Patent Document 1) is known.
JP 2002-1967895 A

  However, in the spoken dialogue apparatus of Patent Document 1, when the recognition result of the user utterance is, for example, “I went to a hot spring on the weekend”, the autonomous response dictionary is searched if the reliability of the recognition result is less than the threshold value. And output a response. Specifically, as a response to “I went to a hot spring on the weekend”, “Speak something”, “Is there anyone?”, “Hey hey”, etc. are output, which is clearly inappropriate, and as a result There is a problem that the dialogue of the future will be broken.

  The present invention has been made to solve the above-described problems, and when the verb in the recognition result of the utterance regarding the user's action (verb) is low in reliability, the dialogue with the user is not broken. It is an object of the present invention to provide a continuous voice dialogue apparatus and a voice dialogue program.

  In order to achieve the above object, the voice interaction apparatus according to claim 1 includes an input means for inputting an utterance by a user, and a first utterance analysis means for analyzing an initial utterance by the user input by the input means. , A verb determination unit for determining whether or not a verb is included in the analysis result by the first utterance analysis unit, and a determination by the verb determination unit that a verb is included in the analysis result by the utterance analysis unit Sometimes a verb reliability determination means for determining whether or not the verb is reliable, and when the verb reliability determination means determines that the verb is reliable, a predetermined response template and the verb A response generating means for generating a response to the first utterance and a predetermined feeling when the verb reliability determining means determines that the verb is unreliable. An emotion question generating means for generating a question asking the emotion of the first utterance of the user by using a question template, and an output for outputting the response generated by the response generating means and the question generated by the emotion question generating means Means.

  According to the first aspect of the present invention, when the verb included in the user's utterance is unreliable, the question can be asked about the user's emotion and the dialogue can be guided in a specific direction.

  According to a second aspect of the present invention, there is provided a voice dialogue apparatus according to the first aspect, wherein the user's utterance input by the input means is analyzed with respect to the question output by the output means. Whether the speech analysis means, the emotion expression storage means storing a word expressing emotion, and the analysis result by the second utterance analysis means include a word expressing the emotion stored in the emotion expression storage means An emotion expression determination means for determining the emotion response determination template, and when the emotion expression determination means determines that a word representing the emotion is included in the analysis result by the emotion response analysis means, a predetermined emotion response template and the emotion An emotion response generation unit that generates a response sentence using a word that expresses the emotion with respect to an expression utterance, and an analysis by the emotion response analysis unit by the emotion expression determination unit When it is determined that a word expressing emotion is not included in the result, a conflict generating means for generating a conflict for the emotion expression utterance using a predetermined conflict is further provided, and the output means includes the emotion The response sentence generated by the response generation unit and the interaction generated by the interaction generation unit are output.

  According to the second aspect of the present invention, it is possible to continue the dialogue by outputting a response sentence and an answer to the response to the question asking the user's emotion.

  The spoken dialogue apparatus according to claim 3, wherein, in the spoken dialogue apparatus according to claim 2, when the verb generation unit determines that the verb determination unit does not include a verb in the analysis result by the utterance analysis unit. In addition, using a predetermined interaction, an interaction is generated.

  According to the third aspect of the present invention, when a verb is not included in the first utterance of the user, it is possible to output the autonomy and continue the conversation.

  The spoken dialogue program according to claim 4, wherein the computer includes an utterance analysis unit that analyzes an initial utterance by the input user, a verb determination unit that determines whether or not a verb is included in an analysis result by the utterance analysis unit, When the verb determining means determines that the verb is included in the analysis result by the utterance analyzing means, the verb reliability determining means for determining whether or not the verb is reliable, the verb reliability determining means When it is determined that the verb is reliable, a response generation unit that generates a response to the first utterance using a predetermined response template and the verb, and the verb reliability determination unit determines the verb Is determined to be unreliable, a question that asks the emotion corresponding to the content of the user's first utterance using a predetermined emotion question template Emotion question generating means to be formed, to function as.

  According to the fourth aspect of the present invention, when the verb included in the user's utterance is unreliable, the question can be asked about the user's emotion and the dialogue can be guided in a specific direction.

  As described above, according to the speech dialogue apparatus and the speech dialogue program of the present invention, the dialogue with the user fails when the verb in the recognition result of the utterance including the verb representing the user's behavior is low in reliability. The effect is that an appropriate response can be generated so as to continue without interruption.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In addition, this invention is not limited to the following embodiment, It is applicable also to what changed the design within the range described in the claim.

  FIG. 1 is a block diagram showing the main configuration of a voice interaction apparatus according to the present invention. As shown in the figure, the voice interaction apparatus according to the present embodiment includes an input unit 11, a first utterance analysis unit 12, a verb determination unit 13, a verb reliability determination unit 14, and a response generation unit 15. An emotion question generation unit 16, a second utterance analysis unit 17, an emotion expression dictionary 18, an emotion expression determination unit 19, an emotion expression response generation unit 20, a conflict generation unit 21, an output unit 22, It has.

  The input unit 11 is composed of a microphone, for example, and collects a user's utterance and generates an audio signal.

  The first utterance analysis unit 12 refers to a word registered in a recognition dictionary database (not shown) in which a plurality of words are registered, and recognizes a user utterance based on an audio signal generated by the input unit 11. The top predetermined number (for example, 10) candidates are selected from the most probable candidates, and morpheme analysis is performed on them using a general morpheme analyzer.

  The verb determination unit 13 determines whether or not a verb is included in the user utterance based on the analysis result by the first utterance analysis unit 12. The verb determination unit 13 determines that a verb is included when a predetermined condition is satisfied, and determines that a verb is not included when the predetermined condition is not satisfied. In the present embodiment, the condition that “8 or more candidates among the top 10 candidates of the analysis result include a verb” is set.

  The verb reliability determination unit 14 determines whether or not the verb is reliable when the verb determination unit 13 determines that the verb is included in the user utterance. The verb reliability determination unit 14 determines that the verb is reliable when a predetermined condition is satisfied, and determines that the verb is not reliable when the predetermined condition is not satisfied. In the present embodiment, a condition is set that “the verb is included in seven or more candidates among the top ten candidates for analysis results”.

  The response generation unit 15 determines that the analysis result by the first utterance analysis unit 12 includes the verb by the verb determination unit 13 and the verb reliability determination unit 14 determines that the verb is reliable. A response is generated using the analysis result by the first utterance analysis unit 12. For example, if the verb included in the analysis result is “go”, the template prepared in advance shown in FIG. 2, for example, “~ nada” and the verb “go” are used. "Da" is generated as a response.

  The emotion question generation unit 16 determines that the analysis result by the first utterance analysis unit 12 includes a verb by the verb determination unit 13 and the verb reliability determination unit 14 determines that the verb is not reliable. The question which asks the user's feeling with respect to the content which a user utterance shows is generated. That is, the emotion question generation unit 16 does not use the analysis result of the user utterance, but asks an emotion that is considered to be possessed when the user utters an action (verb). The emotion question generation unit 16 generates a question using a template prepared in advance, for example, “How did you feel?”

  Unlike the first utterance analysis unit 12, the second utterance analysis unit 17 includes a plurality of words that express emotions, such as “feeling good” and “fun”, and a plurality of words that modify these words, for example, The user's utterance is recognized based on the voice signal generated by the input unit 11 with reference to a word registered in a recognition dictionary database (not shown) in which “very”, “very”, etc. are registered. A new candidate is selected, and a morphological analysis is performed using a general morphological analyzer.

  In the emotion expression dictionary 18, a plurality of words expressing emotions are registered as shown in FIG.

  The emotional expression determination unit 19 uses the emotional expression dictionary 18 to determine whether or not the emotional expression is included in the analysis result by the second utterance analysis unit 17. Specifically, the emotional expression determination unit 19 determines that the emotional expression is included when the analysis result by the second utterance analysis unit 17 includes a word registered in the emotional expression dictionary 18, and the others. It is determined that it does not shoot because emotional expressions are included.

  The emotion expression response generation unit 20 generates a response when the emotion expression determination unit 19 determines that the analysis result of the second utterance analysis unit 17 includes the emotion expression. The emotional expression response generation unit 20 holds an emotional expression response generation database that stores response generation rules including a predetermined response generation template. FIG. 4 shows an emotion expression response generation database. As shown in the figure, in the emotion expression response generation database, “how much?” Is attribute information such as “degree” and “time” that the emotion represented by the emotion expression can have and attribute information “degree”. , And a response generation rule for generating a response such as “when?” With respect to the attribute information “time”. The emotion expression response generation unit 20 determines attribute information that the emotion represented by the emotion expression included in the analysis result described above may have, and generates a response asking for any attribute information regarding the emotion expression.

  When the analysis result by the first utterance analysis unit 12 does not include a verb and when the analysis result by the second utterance analysis unit 17 determines that the emotion expression is not included, Generate a response as a response. The reference generation unit 21 refers to the reference database (not shown) in which the references such as “hee”, “hmm”, “sore”, “so then”, “fumfum” are registered in advance. Generate.

  The output unit 22 is configured by, for example, a speaker, and outputs a response utterance to the user. The output unit 22 is not limited to voice output, and may display a response sentence on a screen or print it on paper.

  Next, the operation of the voice interaction apparatus in the present embodiment will be described along the flowchart shown in FIG. In this embodiment, as an example, “I went to a hot spring on the weekend” is considered as the first user utterance.

  In step 100, the input unit 11 determines whether a user utterance has been input. When a user utterance is input and an audio signal corresponding to the utterance is generated, the process proceeds to step 102. When a user utterance is not input, this determination is repeated until a user utterance is input. In the present embodiment, “I went to a hot spring on the weekend” is input as a user utterance, an audio signal corresponding to this is generated, and the process proceeds to step 102.

  In step 102, the first utterance analysis unit 12 recognizes the user utterance by referring to the words registered in the recognition dictionary database based on the voice signal generated by the input unit 11, and is the most preferred in this embodiment. 10 candidates are selected from the probable candidates and their morphological analysis is performed. FIG. 6 shows the top 10 candidates obtained as an analysis result.

  In step 104, the verb determination unit 13 determines whether or not a verb is included in the user utterance based on the analysis result of FIG. 6. If the verb is included, the process proceeds to step 106, and no verb is included. If so, go to Step 122. In the present embodiment, the verb determination unit 13 determines that the user utterance includes a verb when there are eight or more of the analysis results shown in FIG. 6 including the verb, and includes the verb. Is less than seven, it is determined that the user utterance does not include a verb. In the analysis result shown in FIG. 6, since all ten candidates include verbs, it is determined that the user utterance includes verbs, and the process proceeds to step 106.

  In step 106, the verb reliability determination unit 14 determines whether or not the verb included in the user utterance is reliable. If it is reliable, the process proceeds to step 126, and if not reliable, the process proceeds to step 108. In the present embodiment, the verb reliability determination unit 14 determines that the ten candidates of the analysis result shown in FIG. 6 include the verb when it is seven or more, and includes the verb. Is determined to be unreliable if the number is 6 or less.

  FIG. 7 shows the verbs included in the analysis result candidates of FIG. 6 and the number of analysis result candidates including the verbs. As shown in FIG. 7, the verbs “go”, “similar”, “say”, “is”, and “simmer” included in the analysis result candidates are all included in the seven or more analysis result candidates. Therefore, it is determined that the verb included in the user utterance is unreliable, and the process proceeds to step 108.

  In step 108, the emotion question generation unit 16 generates a question asking an emotion that the user thinks to have when speaking about a verb included in the utterance. Specifically, the emotion question generation unit 16 generates, for example, “Hey, how did you feel?” As a question based on a template prepared in advance, regardless of what verb is included in the utterance. .

  In step 110, the output unit 22 outputs the generated question.

  In step 112, the input unit 11 determines whether or not the user's utterance for the question output by the output unit 22 has been input. When a user utterance is input and an audio signal corresponding to the utterance is generated, the process proceeds to step 114. When a user utterance is not input, this determination is repeated until a user utterance is input. In this embodiment, it is assumed that “good feeling” is input as the user utterance, and an audio signal corresponding to this is generated, and the process proceeds to step 114.

  In step 114, the second utterance analysis unit 17 selects words registered in the recognition dictionary database in which words expressing emotions and words that modify them are registered based on the audio signal generated by the input unit 11. The user utterance is recognized and the morphological analysis is performed. In this case, since the user is asked a question about emotions, the content of the user's utterance can be expected to be narrowed down to those showing emotions. It is possible to recognize using a recognition dictionary database specialized only for a plurality of words. In the present embodiment, it is assumed that “good feeling” was obtained as the most probable candidate of the analysis result.

  In step 116, the emotional expression determination unit 19 determines whether or not the emotional expression is included in the analysis result by the second utterance analysis unit 17 by using the emotional expression dictionary 18, and it is determined that the emotional expression is included. If YES in step 118, the flow advances to step 118. If it is determined that no emotional expression is included, the flow advances to step 122. In the present embodiment, since “feeling good” is registered as an emotion expression in FIG. 3, it is determined that an emotional expression “feeling good” is included in the analysis result, and the process proceeds to step 118.

  In step 118, the emotional expression response generation unit 20 generates an emotional expression response based on the emotional expression and the response generation rule stored in the emotional expression response generation database shown in FIG. In the present embodiment, the emotion expression response generation unit 20 uses a response generation template as a response that asks "degree" as attribute information that the emotion represented by the emotion expression "I felt good" stored in the emotion expression response generation database. Using “How much is it?”, “How pleasant was it?” Is generated.

  In step 120, the output unit 22 outputs the generated emotion expression response.

  On the other hand, it is assumed that “feeling bought” is obtained as the most probable candidate of the analysis result by the second utterance analysis unit 17 in step 114.

  In this case, in step 116, the emotional expression determination unit 19 determines whether or not an emotional expression is included in the analysis result by using the emotional expression dictionary 18. As a result, “I bought a feeling” is not registered in the emotion expression dictionary 18 shown in FIG. 3, so it is determined that the analysis result does not include an emotion expression, and the process proceeds to step 122.

  In step 122, the interest generation unit 21 randomly selects from the considerations registered in the consideration database, and generates, for example, “Yes”, and in step 124, the output unit 22 outputs the consideration.

  As another example of the present embodiment, it is assumed that the first utterance analysis unit 12 selects ten candidates shown in FIG. 8 in step 102 for a user utterance “I went to a hot spring on the weekend”. .

  In this case, in step 104, the verb determination unit 13 determines whether or not a verb is included in the user utterance based on the analysis result of FIG. Since all of the ten candidates in FIG. 8 include a verb, the user utterance is determined to include a verb according to the above-described conditions.

  In step 106, the verb reliability determination unit 14 determines whether or not the verb included in the user utterance is reliable based on the above-described condition. In the case of this example, as shown in FIG. 9, the verb “go” is included in the eight candidates, so that it is determined to be reliable, and the process proceeds to step 126.

  In step 126, the response generation unit 15 generates an input utterance response to the user utterance. Any known method can be used as a method for generating a response. For example, a response “I went” is generated by using a verb “go” and a template “-daa” prepared in advance. To do. Alternatively, a response may be generated by applying the method described in Japanese Patent Application Laid-Open No. 2007-206888. In this method, the response generation unit 15 performs a case analysis on the most probable candidate of the user utterance analysis result “I went to a hot spring on weekends”, extracts case elements and predicates, and confirms the case elements "Who did you?" Or "Who did you?"

  In step 128, the output unit 22 outputs a response.

  As described above, the voice interaction apparatus according to the present embodiment does not generate an erroneous response even when the verb included in the user utterance is unreliable, and thus continues the conversation with the user without failing. Can do. In addition, unlike simply returning a conflict, by explicitly asking the user's emotions, it is possible to prompt the user's next utterance and improve the naturalness of the conversation.

It is a block diagram which shows the main structures of the voice interactive apparatus which concerns on this invention. It is a figure which shows a response template. It is a figure which shows the structure of an emotion expression dictionary. It is a figure which shows the structure of an emotion expression response production | generation database. It is a flowchart which shows the flow of an effect | action of the voice interactive apparatus which concerns on this invention. It is a table | surface which shows the example (the 1) of the candidate of the analysis result of a user utterance. It is a table | surface which shows the example (the 1) of the number of candidates which contain the verb contained in the candidate of the analysis result of a user utterance, and each verb. It is a table | surface which shows the example (the 2) of the candidate of the analysis result of a user utterance. It is a table | surface which shows the example (the 2) of the number of the candidates included in the candidate of the analysis result of a user utterance, and each verb.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Input part 12 1st utterance analysis part 13 Verb determination part 14 Verb reliability determination part 15 Response generation part 16 Emotion question generation part 17 Second utterance analysis part 18 Emotion expression dictionary 19 Emotion expression determination part 20 Emotion expression response generation Unit 21 phase generator 22 output unit

Claims (4)

An input means for inputting a user's utterance;
First utterance analysis means for analyzing a first utterance by a user input by the input means;
Verb determining means for determining whether or not a verb is included in the analysis result by the first utterance analyzing means;
A verb reliability determination unit that determines whether or not the verb is reliable when the verb determination unit determines that the verb is included in the analysis result by the utterance analysis unit;
A response generation unit that generates a response to the first utterance using a predetermined response template and the verb when the verb reliability determination unit determines that the verb is reliable;
An emotional question generating means for generating a question asking the emotion of the user's first utterance using a predetermined emotional question template when the verb reliability determining means determines that the verb is unreliable;
An output means for outputting the response generated by the response generation means and the question generated by the emotion question generation means;
Spoken dialogue device with Second utterance analysis means for analyzing the user's utterance input by the input means for the question output by the output means;
An emotional expression storage means for storing words expressing emotions;
Emotion expression determination means for determining whether or not a word expressing the emotion stored in the emotion expression storage means is included in the analysis result by the second utterance analysis means;
When it is determined by the emotion expression determination means that the analysis result by the emotion response analysis means includes a word expressing emotion, using a predetermined emotion response template and the word expressing the emotion, An emotion response generating means for generating a response sentence for an emotional expression utterance;
A conflict generating means for generating a conflict for the emotional expression utterance using a predetermined conflict when it is determined by the emotional expression determining unit that the analysis result by the emotion response analyzing unit does not include a word expressing the emotion And further comprising
The spoken dialogue apparatus according to claim 1, wherein the output unit outputs the response sentence generated by the emotion response generation unit and the interaction generated by the interaction generation unit.   3. The spoken dialogue according to claim 2, wherein when the verb determining unit determines that the verb is not included in the analysis result of the utterance analyzing unit, the verb generating unit generates a conflict using a predetermined conflict. apparatus. Computer
Utterance analysis means for analyzing the first utterance by the input user,
Verb determining means for determining whether or not a verb is included in the analysis result by the utterance analyzing means,
Verb reliability determination means for determining whether or not the verb is reliable when the verb determination means determines that the verb is included in the analysis result by the utterance analysis means,
A response generation unit that generates a response to the first utterance using a predetermined response template and the verb when the verb reliability determination unit determines that the verb is reliable;
Emotion question generation for generating a question asking an emotion corresponding to the content of the user's first utterance using a predetermined emotion question template when the verb reliability determination means determines that the verb is unreliable means,
Spoken dialogue program to function as.

Images