JP5073024B2 - Spoken dialogue device - Google Patents

Description

  Embodiments described herein relate generally to a voice interaction apparatus that performs a conversation by voice.

  In order for the voice interaction apparatus to interact with the user, it is necessary to utter in accordance with the content of the conversation. However, in order to perform a wide variety of utterances, advanced language processing and a huge database are required. Therefore, a method for generating the utterance content of the system based on the actual conversation history has been proposed. For example, one of the methods uses a dialogue history with a user in a voice dialogue, recognizes a user's utterance voice, and uses a recognition result as a dialogue history.

JP 2002-288155 A

  In the voice dialogue, the user's utterance has a large fluctuation. Semantically speaking, even if the utterance has the same content, the wording or ending changes, and the same sentence is not always uttered. Even in such an environment, it is necessary to generate a new utterance template using voice dialogue.

  Therefore, in view of the above problems, an embodiment of the present invention aims to provide a voice interaction device that generates an utterance template using a voice interaction with a user.

  An embodiment of the present invention provides a plurality of keywords related to a specific topic and information indicating a concept corresponding to each keyword in a voice interaction device that performs a voice conversation with a user using an utterance template stored in a template dictionary. When the keyword stored in the keyword dictionary is included in the stored keyword dictionary storage unit, the speech recognition unit that recognizes speech of the user's utterance, and the utterance that is recognized by the speech recognition unit, A voice dialog, comprising: a template generation unit that replaces the keyword with information indicating the concept corresponding to the keyword, sets the replaced speech sentence as a new speech template, and stores the template in the template dictionary Device.

1 is a block diagram of a voice interaction apparatus according to a first embodiment of the present invention. The flowchart of the speech sentence production | generation of Example 1. FIG. 5 is a flowchart for storing an utterance template according to the first embodiment. 10 is a flowchart of storing an utterance template according to the second embodiment. An example of generating a correct utterance sentence from the speech recognition result of the second embodiment. An example of generating a correct utterance sentence from the speech recognition result of the second embodiment. 10 is a flowchart of storing an utterance template according to the third embodiment. An example of interaction of Example 3. 10 is a first conversion example of an utterance template of Example 4. FIG. 10 is a second conversion example of the utterance template of the fourth embodiment.

BEST MODE FOR CARRYING OUT THE INVENTION

  DESCRIPTION OF EMBODIMENTS Hereinafter, an embodiment of a voice interaction device 1 according to the present invention will be described based on the drawings.

  A voice interaction apparatus 1 according to a first embodiment of the present invention will be described with reference to FIGS.

  As shown in FIG. 1A, the voice interaction device 1 according to the present embodiment is a device built in a robot 3 that interacts with the user 2 by voice, for example.

  First, the configuration of the voice interaction apparatus 1 will be described based on the block diagram of FIG.

  The voice interactive apparatus 1 includes a microphone 15 for inputting the voice of the user 2, a voice recognition unit 13 that recognizes the voice of the user 2 and converts it into a text utterance, and an utterance sentence according to the utterance content of the user 2. An utterance generation unit 11 to be generated, an audio synthesis unit 14 that converts an utterance sentence to a user into speech, a speaker 16 that outputs audio, a template generation unit 12 to generate an utterance plate, a keyword dictionary database 17, a template dictionary database 18, and a concept dictionary 19

  The keyword dictionary database 17 stores a plurality of keyword dictionaries 171. The keyword dictionary 171 is a file in which a plurality of keywords are registered as shown in FIG. 1, and each keyword dictionary 171 stores words that are highly relevant to a specific topic as keywords. For example, the keyword dictionary 171 in FIG. 1 is a dictionary related to “Kyoto trip”, and the keywords are “Kyoto”, “Shinkansen”, “Maiko”, “Yudofu”, and the like. Each keyword is given the concept of the keyword. For example, the concept of “Kyoto” is “place”, the concept of “Shinkansen” is “vehicle”, the concept of “maiko” is “people”, and the concept of “Yudofu” is “food”. The concept of each keyword in the keyword dictionary 171 is given by being pulled from the concept dictionary 19. In the concept dictionary 19, various words are classified into tens to hundreds of concepts. “Concept” means the essential characteristics of things.

  Each keyword dictionary 171 relates to a specific topic as described above, and is generated from text such as a dialog record and related information regarding each topic. The generation method is as follows.

  First, the keyword dictionary database 17 extracts only words obtained by removing particles / auxiliary verbs from each text.

Secondly, the keyword dictionary database 17 calculates the frequency (tf) of each word in the corresponding text and the number N of texts including the word other than the corresponding text. Further, the evaluation value idf = log (N _all / N) is calculated using the total number of texts N _all .

  Thirdly, the keyword dictionary database 17 calculates tf-idf, and extracts a word whose tf-idf is higher than the reference as a keyword.

  Although the keyword dictionary 171 is generated in advance for each topic, it can be generated in real time.

  A plurality of template dictionaries 181 are stored in the template dictionary database 18. The template dictionary 181 stores as many dialog types as “travel”, “TV”, “book”, “movie”, and the like. The template dictionary 181 is a file in which a plurality of utterance templates using each concept in the concept dictionary 19 are described as shown in FIG. For example, the utterance plate is “go to [location]”, “eating [food]”, etc., and the part of [] in which the concept in the utterance template is described is replaced with a keyword (for example, , “Kyoto”), the utterance can be generated. Note that the data structure of the utterance template may be in any format as long as the position where the keyword is to be inserted and the concept of the keyword to be inserted can be specified.

  A method in which the voice interaction apparatus 1 generates an utterance sentence for the user 2 will be described based on the flowchart of FIG.

  The voice interactive apparatus 1 determines in advance what topic to speak before starting a conversation with the user 2. The voice interactive apparatus 1 selects the keyword dictionary 171 and the template dictionary 181 according to the topic. As shown in FIG. 2, the voice interaction apparatus 1 utters a topic “Kyoto trip”, and therefore selects a keyword dictionary 171 storing a keyword related to “Kyoto trip” and a template dictionary 181 storing an utterance template related to “travel”. To do. The utterance generation unit 11 generates an utterance sentence using the selected keyword dictionary database 17 and template dictionary database 18.

  In step s1, the user 2 speaks.

  In step s2, the speech recognition unit 13 recognizes the speech of the user 2 by speech.

  In step s3, the utterance generation unit 11 selects the keyword “Kyoto” from the keyword dictionary 171 for the topic “Kyoto trip” that is currently interacting with the user 2. As a keyword selection method, for example, if the keyword stored in the keyword dictionary 171 is present in the utterance sentence recognized by the utterance generation unit 11, the keyword may be selected or selected randomly. Also good.

  In step s4, the utterance generation unit 11 selects an utterance template including the concept “location” of the selected keyword “Kyoto” in the template dictionary 181 corresponding to the type of topic “travel” that is currently interacting with the user 2. To do.

  In step s5, the utterance generation unit 11 generates an utterance sentence by combining the keyword in the selected keyword dictionary 171 and the utterance template in the template dictionary 181. For example, the utterance generating unit 11 generates an utterance sentence “going to Kyoto” by combining the keyword “Kyoto” and an utterance template “going to [location]”.

  In step s6, the speech synthesizer 14 utters the generated utterance sentence using the speaker 16.

  In this way, the voice interaction apparatus 1 utters speech by combining the keyword dictionary 171 and the template dictionary 181.

  Next, a method in which the voice interaction apparatus 1 stores a new utterance template in the template dictionary 181 using the conversation history with the user 2 will be described. For example, assume that the voice interaction device 1 and the user 2 have the following interaction.

Spoken Dialogue Device 1: “When should I go to Kyoto?”
User 2: “Kyoto is recommended for autumn”
A method in which the voice interaction apparatus 1 generates an utterance template from this dialogue example will be described based on the flowchart of FIG.

  In step s21, the user 2 utters "Kyoto is recommended for autumn" in response to the voice dialogue apparatus 1 uttering "When should I go to Kyoto?"

  In step s22, the speech recognition unit 13 performs speech recognition, thereby obtaining a speech recognition result “Kyoto is recommended for autumn”.

  In step s23, the template generation unit 12 performs a search to determine whether or not the keyword stored in the keyword dictionary 171 is included in the speech sentence that has been voice-recognized. If a keyword is included, the process proceeds to step s24 (in the case of yes), and if not included, the process is ended (in the case of no). Here, since the keyword “Kyoto” is included in the utterance of the user 2, the process proceeds to step s24.

  In step s24, since the keyword is included as a result of the search, the template generation unit 12 replaces that portion with the keyword concept. Here, the template generation unit 12 replaces the concept “location” of “Kyoto”.

  In step s 25, the template generation unit 12 sets the replaced sentence “[Location] is recommended for autumn” as a new utterance template and stores it in the template dictionary 181 “travel” of the template dictionary database 18.

  According to the present embodiment, the voice interaction apparatus 1 can utter in the future by using the new utterance template “[Location] is recommended in autumn”. Since this utterance template is stored in the template dictionary 181 “travel”, it can be used in combination with another keyword dictionary 171 in the case of “travel” dialogue. For example, the spoken dialogue apparatus 1 can generate an utterance sentence “Otaru recommends autumn” using the keyword “Otaru” in the keyword dictionary 171 of “Hokkaido Travel”.

  A voice interaction apparatus 1 according to a second embodiment of the present invention will be described with reference to FIGS.

  When a new utterance template is generated using the utterance sentence of the user 2, the utterance sentence of the user 2 cannot be accurately recognized as in the first embodiment. Therefore, in the present embodiment, one utterance template is generated using the utterance sentences that are voice-recognized by the utterances of the plurality of users 2.

  In the first embodiment, the utterance template is generated in real time immediately after the utterance of the user 2. However, in the voice interaction apparatus 1 of the present embodiment, the utterance template is generated after the conversation with the user 2 is completed.

  A first specific example in which the voice interaction apparatus 1 of the present embodiment generates an utterance template will be described with reference to the flowchart of FIG. 4 and FIG.

  In step s31, as in the first embodiment, the template generation unit 12 searches whether the keyword stored in the keyword dictionary 171 is included in the utterance sentence that is the voice recognition of the plurality of utterances of the user 2. . If a keyword is included, the process proceeds to step s32 (in the case of yes), and if not included, the process is terminated (in the case of no). Here, since each keyword of the user 2 includes the keyword “Kyoto”, the process proceeds to step s32.

  In step s32, the template generation unit 12 generates correct speech recognition utterances using the plurality of utterances including the same keyword in the following order, taking the description of FIG. 5 as an example. FIG. 5 shows a case where the template generation unit 12 has acquired a speech-recognized utterance sentence including five “Kyoto”, and the parentheses in the right side of FIG. It is an actual utterance sentence, and the underlined part in the sentence on the left side is the part that the voice recognition unit 13 has misrecognized.

  First, as illustrated in FIG. 5B, the template generation unit 12 converts other than “Kyoto” into hiragana for the five uttered sentences recognized for each voice.

  Secondly, as shown in FIG. 5B, the template generation unit 12 overlaps as much as possible five utterance sentences centering on the keyword “Kyoto” and unit characters (that is, hiragana units). Matching is performed so that there are more parts.

  Third, the template generation unit 12 selects hiragana having a large number of times of speech recognition among hiragana that has been associated. The template generation unit 12 may select a hiragana having a large number of absolute times, or may select a hiragana having a relatively large number of times. Whether or not the number is absolutely large is determined using a threshold value. Whether or not the number of times is relatively large is determined based on the ranking of the number of times of voice recognition (for example, whether or not the number of times is the largest).

  Fourth, as shown in FIG. 5C, the template generation unit 12 generates a character string by connecting the selected hiragana and the keyword “Kyoto”, and sets the sentence “I went to Kyoto” as a correct sentence. Generate.

  Fifthly, as shown in FIG. 5D, the template generation unit 12 generates a sentence “I went to Kyoto” by assigning kanji to the correct sentence from the original speech recognition result.

  In step s33, the template generation unit 12 replaces “Kyoto” with the concept [place] and sets a sentence “I went to [place]” as a new utterance template.

  In step s34, the template generation unit 12 stores the new utterance template “I went to [location]” in the template dictionary 181 “travel” of the template dictionary database 18.

  In the first specific example, five utterance sentences different from the user 2 are integrated to generate one utterance sentence. Therefore, the erroneously recognized part can be deleted. Also, “Kiyomizu-dera” that was spoken only once is deleted, and only the part that the user 2 often utters remains. Therefore, the habit of the user 2's utterance can also be acquired.

  A second specific example in which the speech dialogue apparatus 1 of the present embodiment generates an utterance template will be described with reference to FIG.

  FIG. 6 shows an example in which the voice interaction apparatus 1 generates an utterance template from the voice recognition result of the user 2 who has a habit of “~ jan”. Since the mustache is uttered frequently, the probability of appearing in the recognition result is high, and there is a high possibility that it remains in the utterance template as shown in FIG. As illustrated in FIG. 6A, the template generation unit 12 acquires a speech-recognized utterance that includes four “Shinkansen”. In FIG. 6A, the text in parentheses () is the actual utterance sentence of the user 2, and the underlined part of the text on the left side is the part that the speech recognition unit 13 has erroneously recognized.

  First, as illustrated in FIG. 6B, the template generation unit 12 converts hiragana other than the keyword “Shinkansen” with respect to each speech-recognized speech sentence.

  Secondly, as shown in FIG. 6B, the template generation unit 12 associates the four “sentences” as much as possible with the hiragana unit around the keyword “Shinkansen”. Do.

  Third, the template generation unit 12 selects hiragana having a large number of times of speech recognition among hiragana that has been associated. The template generation unit 12 may select a hiragana having a large number of absolute times, or may select a hiragana having a relatively large number of times. Whether or not the number is absolutely large is determined using a threshold value. Whether or not the number of times is relatively large is determined based on the ranking of the number of times of voice recognition (for example, whether or not the number of times is the largest).

  Fourth, as shown in FIG. 6C, the template generation unit 12 generates a character string by connecting the selected hiragana and the keyword “Shinkansen”, and the sentence “Shinkansen is fast” is the correct sentence. Generate.

  Fifth, as shown in FIG. 6D, the template generation unit 12 assigns kanji from the original speech recognition result and generates a sentence “Shinkansen is fast”.

  Sixth, the template generation unit 12 replaces the keyword “Shinkansen” with the concept [vehicle] and sets a sentence “[location] is early” as a new utterance template.

  Seventh, the template generation unit 12 stores the new utterance template “[vehicle] is fast” in the template dictionary 181 “travel” of the template dictionary database 18.

  According to the second specific example, since the voice interaction apparatus 1 does not perform linguistic analysis, it can also store a moustache such as “~ jan” that has no linguistic meaning as an utterance template.

  When a large amount of utterance templates are generated by this method and the voice dialogue apparatus 1 utters using the utterance template, the voice dialogue apparatus 1 that speaks in a manner similar to that of the user 2 can be constructed.

  According to the present embodiment, the robot 3 with the built-in voice interaction device 1 can make a utterance with a character, not a tasteless dry utterance, and can speak to the user 2 because the way of speaking is similar to that of the user 2. You can have a sense of affinity.

  A voice interaction apparatus 1 according to a third embodiment of the present invention will be described with reference to FIGS.

  In Example 2, an utterance template was generated using a plurality of speech-recognized utterance sentences. However, with this method, only one utterance template is generated using all speech-recognized utterance sentences including keywords stored in the same keyword dictionary 171. In addition, since the user 2 does not always utter the same sentence, it is difficult to obtain a correct utterance template even if the speech sentences whose voices are recognized by completely different utterances are associated with each other.

  Therefore, in the present embodiment, the speech recognition utterances including the keywords stored in the same keyword dictionary 171 are grouped only by speech sentences having a high degree of similarity, and only the utterance sentences within the same group are used. The correspondence described in 2 is performed and an utterance template is generated.

  A first specific example in which the speech dialogue apparatus 1 of the present embodiment generates an utterance template will be described based on the flowchart of FIG.

  In step s41, as in the first embodiment, the template generation unit 12 searches for whether or not the keyword stored in the keyword dictionary 171 is included in each speech sentence that has been voice-recognized. If a keyword is included, the process proceeds to step s32 (in the case of yes), and if not included, the process is terminated (in the case of no).

  In step s42, the utterance sentences including the same keyword are grouped for each utterance sentence having a high similarity with respect to the utterance sentences recognized by speech including the keywords stored in the keyword dictionary 171. This grouping method will be described in detail later.

  In step s43, after performing grouping, the template generation unit 12 generates an utterance template using an utterance sentence of the same group. Since the subsequent processing is the same as that of the second embodiment, a description thereof will be omitted.

  Next, the utterance sentence grouping method performed in step s42 by the template generation unit 12 will be described in detail.

  The first grouping method is a method that uses a dialogue history with the user 2. The utterance of the user 2 when the voice interactive apparatus 1 utters the same utterance is likely to be the same utterance. FIGS. 8A and 8B are examples of answers of the user 2 when the voice dialogue apparatus 1 makes the same question a plurality of times, which will be described below.

  In the example of FIG. 8 (a), when the voice dialogue apparatus 1 asks twice "Why should I go to Kyoto?", The user 2 said "Kyoto is recommended for autumn" and "Kyoto is so. “Hey, autumn is recommended”.

  In the example of FIG. 8B, when the voice interaction apparatus 1 asks twice "How do you go to Kyoto?", The user 2 may say "That's right, I'm going to get on the Shinkansen." Suppose you answer each.

  Answers to the same question have many overlapping character strings like the underlined portion of the sentence in FIG. As described above, when the dialogue is performed many times or for a long time, the answer of the user 2 when the voice dialogue apparatus 1 asks the same question is regarded as a sentence having a high similarity, and the template generation unit 12 has the same group. Divide into In the first grouping method, the same question performed by the voice dialogue apparatus 1 may not be continuous as a dialogue, and there may be another dialogue between the first question and the next question.

  The second grouping method is a method in which the voice interaction apparatus 1 listens again using a question sentence that causes the user 2 to repeat the same utterance sentence. After the user 2 utters something, the voice interaction apparatus 1 performs the utterance to listen back like “What did you say?”. Then, the user 2 is likely to utter content similar to the sentence spoken earlier. Since the user 2 himself / herself does not remember exactly what was spoken, the user 2 does not always utter exactly the same sentence, but is likely to utter a sentence with similar contents. Therefore, the utterances of the user 2 before and after the speech dialogue apparatus 1 utters such utterances are regarded as sentences having high similarity and are divided into the same group. In order to increase the sentence of the same group, the voice interactive apparatus 1 may utter an intentionally listening back. In the second grouping method, the re-listening performed by the voice dialogue apparatus 1 needs to be continuous as a dialogue.

  The third grouping method is a method that uses the degree of similarity of each speech sentence that has been voice-recognized, and uses the edit distance as the degree of similarity. The template generation unit 12 calculates the edit distance between speech sentences that have been speech-recognized including the keywords stored in the same keyword dictionary 171. The template generation unit 12 regards sentences having a close edit distance as having high similarity, and divides two utterance sentences having an edit distance equal to or less than a threshold into the same group. The “edit distance” is a numerical value indicating how different two character strings are. For example, one character string is transformed into another character string by inserting, deleting, or replacing characters. Means the minimum number of steps required.

  According to the present embodiment, by creating an utterance template by associating utterance sentences with high similarity, more character strings remain in the utterance template, and a longer and more complex utterance sentence can be generated.

  A voice interaction apparatus 1 according to a fourth embodiment of the present invention will be described with reference to FIGS.

  In the present embodiment, not only the generated utterance template is stored as it is, but also a sentence obtained by converting the utterance template based on a certain conversion rule is stored as the utterance template.

  Hereinafter, an example of the first conversion rule will be described.

  The first conversion rule changes the style of the original utterance template to a positive sentence, a negative sentence, or a question sentence. As an example, a rule for converting a positive sentence into a question sentence will be described. The conversion from an affirmative sentence to a question sentence may be performed mainly by converting a word at the end of the sentence. The following are conversion rules corresponding to one or two types of morphemes at the end of the sentence when the utterance template is analyzed. If a particle comes at the end of the sentence, the particle is ignored.

・ Noun + auxiliary verb → noun + "What?"
・ Noun → Noun + “What?”
・ Verb + auxiliary verb → basic verb + “no?”
・ Verb-> Verb basic form + "no?"
・ Adjective + auxiliary verb → adjective basic form + “no?”
-Adjective → Adjective basic form + "no?"
As illustrated in FIG. 9A, a method will be described in which the template generation unit 12 converts an utterance template “[Place] is recommended in autumn” into a question sentence as an example. FIG. 9 shows the conversion process.

  First, the template generation unit 12 performs morphological analysis on the utterance template as shown in FIG. Then, in this utterance template, if attention is paid only to morphemes other than particles, the end of the sentence is composed of “noun + auxiliary verb”.

  Next, as shown in FIG. 9C, the template generation unit 12 applies the above rule “noun + auxiliary verb → noun +“ no? ”” And “recommend” is “recommended?” Convert to

  Next, the template generation unit 12 generates a new utterance template “Would you recommend autumn for [location]?”.

  Hereinafter, an example of the second conversion rule will be described.

  The second conversion rule changes the tense of the utterance template. Below are the rules for converting from present tense to past tense.

・ Noun → Noun + “It was”
・ Noun + auxiliary verb → noun + "was"
・ Verb + auxiliary verb → verb combination + “ta”
・ Verb → Verb + + “ta”
・ Adjective + auxiliary verb → adjective combination + “ta”
・ Adjective → Adjective + + “ta”
The template generation unit 12 applies a rule of “noun + auxiliary verb → noun +“ was ”as shown in FIG. 9D to the utterance template“ [Place] is recommended for autumn ”. Then, “[Location] recommended autumn”.

  Hereinafter, an example of the third conversion rule will be described.

  The third conversion rule partially changes the word / phrase in the utterance template.

  In the spoken dialogue apparatus 1, there is a concept dictionary 19, in which various words are classified into the same concept, and basically words of the same concept are used in the same way in the sentence.

  For example, taking “sometimes” and “frequently” with the concept of time frequency as an example, even if only that part is replaced in the same sentence, such as “frequently go to the library” and “frequently go to the library”, the Japanese sentence Can be used as

  Therefore, first, the template generation unit 12 performs morphological analysis on the utterance template, searches each morpheme in the concept dictionary 19, and finds the concept of each morpheme. Next, the template generation part 12 uses what replaced the word in the same concept as a new speech template.

  As shown in FIG. 10A, a case where an utterance template “[Location] is recommended for autumn” will be described as an example.

  As shown in FIG. 10B, the template generation unit 12 performs morphological analysis, and when each morpheme is searched in the concept dictionary 19, it can be seen that the word “autumn” has the concept of “season”.

  Therefore, the template generation unit 12 extracts other words having the same concept as “autumn” as shown in FIG. 10 (b), and replaces them with the extracted words as shown in FIG. 10 (d). The utterance templates “[Location] is recommended for spring”, “[Location] is recommended for early spring”, “[Location] is recommended for the rainy season”, etc. are generated.

  Conversions combining some of the above examples are also possible.

Example of change

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 ... Voice dialogue apparatus, 11 ... Speech production | generation part, 12 ... Template production | generation part, 13 ... Speech recognition part, 14 ... Speech synthesis part, 171 ... Keyword dictionary, 181 ...・ Template dictionary

Claims (8)

In a voice interaction device for performing a voice interaction with a user using an utterance template stored in a template dictionary,
A keyword dictionary storage unit storing a plurality of keywords related to a specific topic, and information indicating a concept corresponding to each keyword,
A voice recognition unit that recognizes a user's speech,
When the keyword stored in the keyword dictionary is included in the utterance sentence recognized by the voice recognition unit, the keyword is replaced with information indicating the concept corresponding to the keyword, and the utterance is replaced. A template generation unit that sets a sentence as a new utterance template and stores it in the template dictionary;
A voice interactive apparatus characterized by comprising: The template generation unit
With respect to the plurality of uttered sentences that have been voice-recognized, parts other than the keywords are divided by unit characters,
With respect to each unit character of each utterance sentence, making an association based on the same keyword between the utterance sentences,
Generating the utterance template from a character string connecting the unit characters that are absolute or relatively large in the number of times of voice recognition among the unit characters associated with each other;
The spoken dialogue apparatus according to claim 1. An utterance generator for outputting utterances to the user;
The utterance generation unit outputs the same utterance multiple times,
The template generation unit generates the utterance template using each utterance sentence answered by the user with respect to the plurality of the same utterances.
The spoken dialogue apparatus according to claim 2. An utterance generator for outputting utterances to the user;
The utterance generation unit outputs a question sentence such that the user repeats the same utterance sentence as an answer multiple times,
The template generation unit generates the utterance template using each utterance sentence answered by the user with respect to the question sentence a plurality of times.
The spoken dialogue apparatus according to claim 3. The template generation unit
Among the utterances recognized by the voice recognition unit, the utterance template is generated using only utterances whose similarity is higher than a threshold.
The spoken dialogue apparatus according to claim 1. The template generation unit
Perform morphological analysis on the utterance template,
Based on the result of the morphological analysis, the original utterance template is changed to generate a new utterance template.
The spoken dialogue apparatus according to claim 2. The template generation unit
Based on the result of the morphological analysis, change the style of the utterance template to a positive sentence, a negative sentence, or a question sentence, or change the tense of the utterance template to generate a new utterance template.
The voice interactive apparatus according to claim 6. The template generation unit
Based on the result of the morphological analysis, a word other than the keyword in the original utterance template is changed to a sentence replaced with a word having the same concept as the word, and a new utterance template is generated.
The voice interactive apparatus according to claim 6.

Images