JP6773074B2 - Response generation method, response generator and response generation program - Google Patents

Description

The present invention relates to a response generator, a response generation method, and a response generation program that respond to a user.

Based on the voice recognition means that recognizes the user's voice, the structural analysis means that analyzes the structure of the voice recognized by the voice recognition means, and the structure of the voice analyzed by the structural analysis means, a response sentence to the user's voice is generated. A response generation device including a response output means for generating and outputting the generated response sentence is known (see, for example, Patent Document 1).

Japanese Unexamined Patent Publication No. 2010-157081

However, in the response generation device as described above, it takes time to analyze the structure of the voice and generate the response sentence, and a response waiting occurs. For this reason, there is a risk that the dialogue may feel uncomfortable.

The present invention has been made to solve such a problem, and a main object of the present invention is to provide a response generation method, a response generation device, and a response generation program capable of alleviating the discomfort of dialogue due to waiting for a response. And.

One aspect of the present invention for achieving the above object is a step of recognizing a user's voice, a step of analyzing the structure of the recognized voice, and a step of analyzing the structure of the analyzed voice of the user. A response generation method including a step of generating a response sentence to a voice and outputting the generated arbitrary response sentence, the step of generating the recognized user's voice as a repeated response sentence, and the above-mentioned The response generation method is characterized by including a step of outputting the generated repetitive response sentence before outputting an arbitrary response sentence based on a voice structure.
In this aspect, the generated iteration further comprises a step of analyzing the phoneme of the user's voice and a step of generating an aizuchi response to the user's voice based on the result of the phoneme analysis. Before outputting the response statement, the response of the generated aizuchi may be output.
In this aspect, a plurality of response candidates for the user's voice are generated based on the analyzed voice structure, and the generated repetitive response sentence and the generation are generated from the generated plurality of response candidates. A response candidate that overlaps with the response of the aizuchi may be excluded, and the response candidate selected from the excluded response candidates may be used as the optional response statement.
In this aspect, the keyword and the part of speech of the keyword are extracted from the recognized user's voice, and the plurality of keywords, the part of speech of each keyword, and the additional ending are associated with each other based on the additional information. A repetitive response sentence may be generated by selecting the additional ending corresponding to the extracted keyword and part of speech and adding the selected additional ending to the extracted keyword.
One aspect of the present invention for achieving the above object is a voice recognition means for recognizing a user's voice, a structural analysis means for analyzing the structure of the voice recognized by the voice recognition means, and an analysis by the structural analysis means. A response generation device including a response output means for generating a response sentence to the user's voice based on the generated voice structure and outputting the generated arbitrary response sentence, which is recognized by the voice recognition means. A repeat generation means for generating the voice of the user as a repeated response sentence is provided, and the response output means is provided by the repeat generation means before outputting an arbitrary response sentence based on the structure of the voice. It may be a response generator characterized by outputting the generated repetitive response statement.
One aspect of the present invention for achieving the above object is a process of recognizing a user's voice, a process of analyzing the structure of the recognized voice, and a process of analyzing the user's voice based on the analyzed voice structure. A process of generating a response sentence to a voice and outputting the generated voluntary response sentence, a process of generating the recognized user's voice as a repeated response sentence, and a voluntary response based on the structure of the voice. A response generation program may be characterized in that a computer is made to execute a process of outputting the generated repeated response sentence before outputting the statement.

According to the present invention, it is possible to provide a response generation method, a response generation device, and a response generation program that can alleviate the discomfort of dialogue due to waiting for a response.

It is a block diagram which shows the schematic system configuration of the response generation apparatus which concerns on Embodiment 1 of this invention. It is a flowchart which shows the processing flow of the response generation method which concerns on Embodiment 1 of this invention. It is a block diagram which shows the schematic system configuration of the response generation apparatus which concerns on Embodiment 2 of this invention. It is a flowchart which shows the processing flow of the response generation method which concerns on Embodiment 2 of this invention. This is an example of additional information stored in the memory.

Embodiment 1
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a schematic system configuration of the response generator according to the first embodiment of the present invention. The response generation device 1 according to the first embodiment is a voice recognition unit 2 that recognizes the user's voice, a structural analysis unit that analyzes the structure of the voice, and 3, and a response output that generates and outputs a response sentence to the user's voice. It includes a unit 4 and a repeat generation unit 5 that generates a repetitive response statement.

The response generation device 1 is, for example, a CPU (Central Processing Unit) that performs arithmetic processing, a ROM (Read Only Memory) or a RAM (Random Access Memory) that stores an arithmetic program, a control program, or the like executed by the CPU. The hardware is configured around a microcomputer consisting of a memory composed of a memory, an interface unit (I / F) for inputting / outputting signals to and from the outside, and the like. The CPU, memory, and interface unit are connected to each other via a data bus or the like.

The voice recognition unit 2 is a specific example of the voice recognition means, performs voice recognition processing based on the user's voice information acquired by the microphone 6, converts the user's voice into text, and generates character string information. The voice recognition unit 2 detects the utterance section from the user's voice information output from the microphone 6, and performs pattern matching on the detected utterance section voice information by referring to, for example, a statistical language model. Recognize. Here, the statistical language model is a probability model for calculating the appearance probability of a linguistic expression such as the appearance distribution of words and the distribution of words appearing next to a certain word, and the connection probability is learned in morpheme units. It is a thing. The statistical language model is stored in advance in the above memory or the like. The voice recognition unit 2 outputs the recognized user's voice information to the structural analysis unit 3 and the repeat generation unit 5.

The structural analysis unit 3 is a specific example of the structural analysis means, and analyzes the structure of the voice information recognized by the voice recognition unit 2. The structural analysis unit 3 performs morphological analysis or the like on the character string information indicating the voice information of the user who has been voice-recognized by using a general morphological analyzer, and interprets the meaning of the character string information. The structural analysis unit 3 outputs the analysis result of the character string information to the response output unit 4.

The response output unit 4 is a specific example of the response output means, and generates a response sentence (hereinafter, referred to as a voluntary response sentence) for the user's voice information based on the structure of the voice information analyzed by the structural analysis unit 3. Then, the generated voluntary response statement is output. The response output unit 4 generates, for example, a voluntary response sentence to the user's voice information based on the analysis result of the character string information output from the structural analysis unit 3. Then, the response output unit 4 outputs the generated response sentence by using the speaker 7.

More specifically, the structural analysis unit 3 extracts the predicate argument structure in the character string information "eat pork cutlet" and specifies the predicate "eat" and the case particle "o". Then, the response output unit 4 extracts the types of case particles that can be related to the predicate "eat" specified by the structural analysis unit 3 from the deficient case dictionary database 8 in which the correspondence between the predicate and the case particles is stored. To do. The deficient dictionary database 8 is built in the above memory, for example.

The response output unit 4 generates, for example, a predicate argument structure such as “what to eat”, “where to eat”, “when to eat”, and “who to eat” as a voluntary response sentence. Further, the response output unit 4 randomly selects from other predicate argument structures excluding the surface case “o” that does not match the user's voice in the generated predicate argument structure, and selects the predicate argument. The structure is a voluntary response statement. In this way, the response output unit 4 interprets the meaning of the voice information based on the structure of the voice information analyzed by the structure analysis unit 3 and generates a plurality of voluntary response candidates. Then, the response output unit 4 selects the optimum candidate from the plurality of generated voluntary response candidates and sets it as a voluntary response statement. The response output unit 4 selects, for example, the predicate argument structure "who did you eat with?" And outputs it as a voluntary response sentence.

By the way, as described above, structural analysis of voice information and generation of a response sentence thereof require time (for example, about 3 seconds), and processing cost is high. For this reason, there is a risk that a response wait will occur and the dialogue will feel uncomfortable.

On the other hand, in the response generation device 1 according to the first embodiment, the repeat generation unit 5 refers to the user's voice recognized by the voice recognition unit 2 as a repeat response sentence (hereinafter, referred to as a repeat response sentence). ). Then, the response output unit 4 outputs the repeat response sentence generated by the repeat generation unit 5 before outputting the voluntary response sentence based on the voice structure.

As a result, since the repeated response sentence simply repeats the recognized user's voice as it is, it does not require a generation time (for example, about 1 second), and the processing cost is low. Therefore, it is possible to output a repeated response statement with a low processing cost while waiting for a response until the optional response statement generated based on the voice structure having a high processing cost is output. As a result, it is possible to alleviate the discomfort of the dialogue caused by the large interval between the dialogues caused by waiting for a response.

The repeat generation unit 5 generates the voice information recognized by the voice recognition unit 2 as a repeat response sentence for performing so-called parrot return. The repeat generation unit 5 outputs the generated repeat response statement to the response output unit 4. Then, the response output unit 4 sends the iterative response sentence output from the iterative generation unit 5 to the speaker 7 before the voluntary response sentence generated based on the analysis result of the character string information output from the structural analysis unit 3. Output from. In this way, a plurality of response statements having different processing costs are generated in parallel, and the response statements are output in the order of generation. As a result, it is possible to realize a dialogue that maintains the continuity of the dialogue and does not impair the tempo.

FIG. 2 is a flowchart showing a processing flow of the response generation method according to the first embodiment.
The voice recognition unit 2 performs voice recognition of the user's voice information acquired by the microphone 6 (step S101), and outputs the recognized user's voice information to the structure analysis unit 3 and the repeat generation unit 5.

The repeat generation unit 5 generates the voice information recognized by the voice recognition unit 2 as a repeat response sentence (step S102), and outputs the generated repeat response sentence to the response output unit 4.
The response output unit 4 outputs the repeat response statement output from the repeat generation unit 5 from the speaker 7 (step S103).

In parallel with the above (step 102) and (step 103), the structural analysis unit 3 analyzes the structure of the voice information recognized by the voice recognition unit 2 (step S104), and responds with the analysis result of the character string information. Output to the output unit 4.

The response output unit 4 generates a voluntary response sentence based on the analysis result of the character string information output from the structural analysis unit 3 (step S105), and outputs the generated voluntary response sentence from the speaker 7 (step S106).

As described above, in the first embodiment, the recognized user's voice is generated as the repeated response sentence, and the repeated response sentence is output before the optional response sentence based on the structure of the voice is output. As a result, it is possible to output a repeated response statement with a low processing cost while waiting for a response until the optional response statement generated based on the structure of the voice having a high processing cost is output. Therefore, it is possible to alleviate the discomfort of the dialogue caused by the large interval between the dialogues caused by waiting for a response.

Embodiment 2.
FIG. 3 is a block diagram showing a schematic system configuration of the response generator according to the second embodiment of the present invention. In addition to the configuration of the response generation device 1 according to the first embodiment, the response generation device 20 according to the second embodiment includes a phoneme analysis unit 21 that analyzes the phoneme of the user's voice information, and an aizuchi for the user's voice information. It is characterized in that it further includes an aizuchi generation unit 22 that generates a response.

The phoneme analysis unit 21 is a specific example of the phoneme analysis means, and analyzes the phoneme of the user's voice information based on the user's voice information acquired by the microphone 6. For example, the phoneme analysis unit 21 estimates a break in the user's voice by detecting a change in the volume level or a frequency change (fundamental frequency, etc.) of the voice information. The phoneme analysis unit 21 outputs the phoneme analysis result to the aizuchi generation unit 22.

The aizuchi generation unit 22 is a specific example of the aizuchi generation means, and generates an aizuchi response (hereinafter referred to as an aizuchi response) to the user's voice based on the phoneme analysis result output from the phoneme analysis unit 21. .. For example, the aizuchi generation unit 22 searches the standard response database 23 in which the pattern of the aizuchi is stored when the volume level of the voice information becomes equal to or less than the threshold value. Then, the aizuchi generation unit 22 randomly selects an aizuchi response from the standard response database 23. The standard response database 23 stores a plurality of patterns used for aizuchi such as "Yeah. Yeah.", "I see.", And "Hmm." The standard response database 23 is built in the above memory or the like. The aizuchi generation unit 22 outputs the generated aizuchi response to the response output unit 4.

The response output unit 4 outputs the reciprocal response generated by the reciprocal generation unit 22 from the speaker 7 before the repetitive response sentence generated by the repetitive generation unit 5.

The phoneme analysis unit 21 can detect a change in volume level in real time. Further, the amount of frequency calculation when the phoneme analysis unit 21 detects the frequency change is smaller than that of pattern matching, and the processing delay is small. In this way, the phoneme analysis unit 21 performs phoneme analysis using features with low processing costs. Therefore, the generation time of the aizuchi response is shorter than the generation time of the repeated response statement (for example, about 300 msec), and the processing cost is lower.

Therefore, it is possible to output an aizuchi response with a lower processing cost before the repeated response statement is output. As a result, the connection between the dialogues becomes smoother, and the discomfort of the dialogue can be further alleviated. Furthermore, more responses and response statements with different processing costs are generated in parallel and output in the order of generation. As a result, it is possible to realize a more natural dialogue that maintains the continuity of the dialogue more smoothly and does not impair the sense of tempo.

The aizuchi generation unit 22 routinely generates an aizuchi response, and the repeat generation unit 5 only superficially interprets the speech recognition result to generate a repeat response sentence. Therefore, it is assumed that the response output unit 4 generates a voluntary response candidate similar to the aizuchi response generated by the aizuchi generation unit 22 and the repeat response generated by the repeat generation unit 5.

On the other hand, the response output unit 4 excludes the aizuchi response generated by the aizuchi generation unit 22 and the voluntary response candidate overlapping with the repeat response generated by the repeat generation unit 5 from the voluntary response candidates. Then, the response output unit 4 selects the optimum candidate from the excluded voluntary response candidates and sets it as a voluntary response statement. As a result, duplicate unnecessary words can be eliminated and a more natural dialogue can be realized.

For example, in response to the user's utterance "It's hot today", the Aizuchi generation unit 22 generates an Aizuchi response "Yeah". Subsequently, the repeat generation unit 5 generates a repeat response sentence "It's hot". On the other hand, the response output unit 4 generates voluntary response candidates "I don't like it", "How long is it hot?", "It's hot", "That's right" and the like. The response output unit 4 eliminates "hot" that overlaps with the iterative response statement generated by the iterative generation unit 5 from the generated voluntary response candidates. Then, the response output unit 4 selects, for example, "how hot is it?" From the excluded voluntary response candidates, and sets it as a voluntary response sentence.

An example of the dialogue generated as described above is shown below. In the following example, M is a response sentence and a response of the response generator 20, and U is a user's utterance.
U: "It's hot today."
M (Aizuchi response): "Yeah."
M (repeated response sentence): "It's hot."
M (voluntary response): "How long will it be hot?"
In this way, it is possible to eliminate duplicate unnecessary words and realize a more natural dialogue while maintaining the continuity of the dialogue more smoothly.

In the response generation device 20 according to the second embodiment, the same parts as those of the response generation device 1 according to the first embodiment are designated by the same reference numerals, and detailed description thereof will be omitted.

FIG. 4 is a flowchart showing a processing flow of the response generation method according to the second embodiment.
The phoneme analysis unit 21 analyzes the phoneme of the user's voice information based on the user's voice information acquired by the microphone 6 (step S201), and outputs the phoneme analysis result to the aizuchi generation unit 22.

The aizuchi generation unit 22 generates an aizuchi response to the user's voice based on the phoneme analysis result output from the phoneme analysis unit 21 (step S202), and outputs the generated aizuchi response to the response output unit 4.
The response output unit 4 outputs the aizuchi response output from the aizuchi generation unit 22 from the speaker 7 (step S203).

In parallel with the above processes (step S201) to (step 203), the voice recognition unit 2 performs voice recognition of the user's voice information acquired by the microphone 6 (step S204), and obtains the recognized user's voice information. Output to the structural analysis unit 3 and the repeat generation unit 5.

The repeat generation unit 5 generates the voice information recognized by the voice recognition unit 2 as a repeat response sentence (step S205), and outputs the generated repeat response sentence to the response output unit 4.
The response output unit 4 outputs the repeat response statement output from the repeat generation unit 5 from the speaker 7 (step S206).

In parallel with the above processes (step 205) and (step S206), the structural analysis unit 3 analyzes the structure of the voice information recognized by the voice recognition unit 2 (step S207), and the analysis result of the character string information. Is output to the response output unit 4.

The response output unit 4 generates a plurality of optional response candidates based on the analysis result of the character string information output from the structural analysis unit 3 (step S208). The response output unit 4 excludes the aizuchi response generated by the aizuchi generation unit 22 and the voluntary response candidate overlapping with the repeat response generated by the repeat generation unit 5 from the voluntary response candidates. Then, the response output unit 4 selects the optimum candidate from the excluded voluntary response candidates and sets it as a voluntary response statement (step S209). The response output unit 4 outputs the generated voluntary response statement from the speaker 7 (step S210).

The following is an example of a dialogue between the response generator 20 and the user.
M (topic provided): "What did you eat for lunch?"
U: "I ate pork cutlet."
M (Aizuchi response): "Yeah. Yeah."
M (repeated response sentence): "I ate pork cutlet."
M (voluntary response): "Who did you eat with?"
U: "I ate with my friends."
M (Aizuchi response): "That's right."
M (repeated response): "I ate with a friend."
M (voluntary response): "Where did you eat?"
U: "I ate at Yabaton."
M (Aizuchi response): "I see."
M (repeated response sentence): "I ate at Yabaton."
M (voluntary response sentence): "I ate it."
U: "It was delicious."
M (Aizuchi response): "Hmm."
M (repeated response sentence): "It was delicious."
M (voluntary response sentence): "I like it. Mr. XX."

As an example of the above dialogue shows, when the user utters, the aizuchi response, the repeated response sentence, and the voluntary response sentence of the response generator 20 are continuously connected at a good tempo to the utterance, and the connection between the dialogues is smoother. It turns out that

As described above, in the second embodiment, the phonology of the user's voice information is analyzed, the aizuchi response to the user's voice information is generated based on the analysis result, and the aizuchi response generated before the repeated response sentence is output is generated. Output. As a result, it is possible to output an aizuchi response with a lower processing cost before the repeated response statement is output. As a result, the connection between the dialogues becomes smoother, and the discomfort of the dialogue can be further alleviated.

Embodiment 3.
The repeat generation unit 5 according to the third embodiment extracts a keyword from the user's voice information recognized by the voice recognition unit 2, and adds a specific additional ending to the extracted keyword to perform a repeat response. It is characterized by generating a statement.

The repeat generation unit 5 generates the voice information recognized by the voice recognition unit 2 as a repeat response sentence for returning the parrot. Here, the naturalness of the dialogue is further improved by adding a specific additional ending to the user's voice information and returning the parrot, rather than returning the parrot as it is without changing the user's voice at all. For example, the naturalness of the dialogue is improved when the response generator 1 responds to the user's utterance "I went to the sea" by responding "I went to the sea" rather than simply responding "I went to the sea". To do.

For example, additional information in which a plurality of keywords, a part of speech of each keyword, and an additional ending are associated with each other is stored in the memory. The repeat generation unit 5 extracts a keyword and a part of speech of the keyword from the user's voice information recognized by the voice recognition unit. The repeat generation unit 5 selects an additional ending corresponding to the extracted keyword and part of speech based on the additional information stored in the memory. The repeat generation unit 5 generates a repeat response sentence by adding a selected additional ending to the extracted keyword.

More specifically, the repeat generation unit 5 uses the keyword "tonkatsu" and the part of speech "noun" and the keyword "" from the voice character string information "I ate tonkatsu" recognized by the voice recognition unit 2. , The part of speech "particle" of the keyword, the keyword "eaten", the part of speech "verb" of the keyword, the keyword "yo", and the part of speech "particle" of the keyword. Further, the repeat generation unit 5 selects the keyword "tonkatsu" and the part of speech "noun" from these extracted keywords and part of speech, and the extracted keyword "tonkatsu" and part of speech "noun" based on the additional information in the memory. Select the additional ending "ka" corresponding to "". Here, the repeat generation unit 5 arbitrarily extracts a keyword of a noun or an adjective from the character string information of the voice recognized by the voice recognition unit 2, and selects a corresponding additional ending.

When there are a plurality of additional endings corresponding to the extracted keywords and part of speech, the repeat generation unit 5 may select them according to a preset priority order. Similarly, when the repeat generation unit 5 selects, for example, the keyword "done" and the part of speech "interjection" from the character string information of the voice recognized by the voice recognition unit 2, the keyword is based on the additional information. Select the additional ending "ne" corresponding to "done" and the part of speech "interjection". By extracting not only the keywords but also the part of speech at the same time, it is possible to distinguish the above-mentioned "done" by the verb "done" and the verb "done".

For example, the voice recognition unit 2 may add the prototype "delicious" before utilization to the character string information "delicious" of the recognized user's voice and output it to the repeat generation unit 5. In this case, the repeat generation unit 5 extracts the keyword “delicious” and the part of speech “adjective”. The iterative generation unit 5 selects the additional ending "noka" corresponding to the keyword "delicious" and the part of speech "adjective" based on the additional information.

FIG. 5 is an example of additional information stored in the memory. In the additional information shown in FIG. 2, when the keyword is a wildcard "*", all the keywords are targeted. Therefore, when the keyword "pork cutlet" and the part of speech "noun" are extracted, the repeat generation unit 5 randomly selects one of the additional endings "ka" and "nanda" with reference to the additional information.

As described above, the repeat generation unit 5 extracts the keyword “pork cutlet” and the part of speech “noun”. Then, the iterative generation unit 5 randomly selects the additional ending “ka” corresponding to the keyword “tonkatsu” and the part of speech “noun” with reference to the additional information. Finally, the repeat generation unit 5 generates the repeat response sentence "tonkatsu ka" by adding the selected additional ending "ka" to the extracted keyword "tonkatsu". Here, the repeat generation unit 5 generates, for example, a repeat response sentence “tonkatsu, pork cutlet” by adding an additional ending “ka” to “tonkatsu, pork cutlet” which is obtained by repeating the extracted keyword twice. May be good. As a result, the tempo of the dialogue is enhanced, and the naturalness of the dialogue is further improved.

According to the third embodiment, the repeat generation unit 5 extracts the keyword and the part of speech from the voice information of the user, selects the additional ending corresponding to the keyword and the part of speech based on the additional information, and adds the additional ending to the keyword. Generate a repeat response statement by just doing the process. Therefore, since the iterative response statement can be generated by simple processing, the processing cost can be kept low. Further, not only the user's voice is simply returned to the parrot, but also various parrots can be returned by appropriately adding additional endings according to the user's voice information, so that the naturalness of the dialogue can be further improved.

The present invention is not limited to the above embodiment, and can be appropriately modified without departing from the spirit.

In the above embodiment, the response output unit 4 outputs the aizuchi response generated by the aizuchi generation unit 22 from the speaker 7, but the present invention is not limited to this. The response output unit 4 may perform an arbitrary response with a low processing load based on the aizuchi response generated by the aizuchi generation unit 22. For example, the response output unit 4 may perform vibration of the vibration device, lighting / blinking of the light device, display of the display device, operation of each part such as the limbs, head, and body of the robot, and any combination of these may be performed. You may go.

In the above embodiment, the response output unit 4 outputs the repeat response sentence generated by the repeat generation unit 5 from the speaker 7, but the present invention is not limited to this. The response output unit 4 may output an arbitrary repeated response statement having a low processing load based on the repeated response statement generated by the repeat generation unit 5. For example, the response output unit 4 may output a repeated response sentence by using a display of a display device or the like, or may output by arbitrarily combining means.

Further, the present invention can also be realized, for example, by causing the CPU to execute a computer program for the processes shown in FIGS. 2 and 4.

Programs can be stored and supplied to a computer using various types of non-transitory computer readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-temporary computer-readable media include magnetic recording media (eg, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (eg, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, It includes a CD-R / W and a semiconductor memory (for example, a mask ROM, a PROM (Programmable ROM), an EPROM (Erasable PROM), a flash ROM, and a RAM (random access memory)).

The program may also be supplied to the computer by various types of transient computer readable media. Examples of temporary computer-readable media include electrical, optical, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

1 Response generator, 2 Speech recognition unit, 3 Structural analysis unit, 4 Response output unit, 5 Repeat generation unit, 6 Microphone, 7 Speaker, 8 Insufficient dictionary database, 21 Phonology analysis unit, 22 Aizuchi generation unit, 23 Standard Response database

Claims (6)

Steps to recognize the user's voice and
The step of analyzing the structure of the recognized voice and
A step of generating a response sentence to the user's voice based on the analyzed voice structure and outputting the generated arbitrary response sentence.
It is a response generation method including
From the recognized user's voice , a step of generating a plurality of repetitive response statements with different processing costs in parallel, and
Before outputting the response sentence optional based on the structure of the audio, including the steps of outputting the response sentence of repetitive said generated by the generated order response generating method, characterized in that .. The response generation method according to claim 1.
The step of analyzing the phonology of the user's voice,
A step of generating an aizuchi response to the user's voice based on the phonological analysis result is further included.
A response generation method, characterized in that the response of the generated aizuchi is output before the output of the generated repetitive response statement. The response generation method according to claim 2.
Based on the structure of the analyzed voice, a plurality of response candidates for the user's voice are generated, and from the generated plurality of response candidates, the generated repetitive response sentence and the generated response of the aizuchi are generated. A response generation method characterized in that a response candidate overlapping with the above is excluded, and a response candidate selected from the excluded response candidates is used as the optional response statement. The response generation method according to claim 3.
A keyword and a part of speech of the keyword are extracted from the recognized user's voice, and the extracted keyword and part of speech are extracted based on additional information in which a plurality of keywords, a part of speech of each keyword, and an additional ending are associated with each other. A response generation method, characterized in that a repetitive response sentence is generated by selecting the additional ending corresponding to the above and adding the selected additional ending to the extracted keyword. A voice recognition means that recognizes the user's voice,
A structural analysis means for analyzing the structure of the voice recognized by the voice recognition means, and
A response generator including a response output means that generates a response sentence to the user's voice based on the structure of the voice analyzed by the structural analysis means and outputs the generated arbitrary response sentence.
A repeat generation means for generating a plurality of repetitive response sentences having different processing costs in parallel from the user's voice recognized by the voice recognition means is provided.
The response output means outputs a plurality of repetitive response sentences generated by the repeat generation means in the order in which they are generated, before outputting an arbitrary response sentence based on the voice structure. A featured response generator. The process of recognizing the user's voice and
The process of analyzing the structure of the recognized voice and
A process of generating a response sentence to the user's voice based on the analyzed voice structure and outputting the generated arbitrary response sentence.
A process of generating a plurality of repetitive response statements having different processing costs in parallel from the recognized user's voice,
Before outputting the response sentence optional based on the structure of the voice response, characterized in that to execute a process of outputting the response sentence of repetitive said generated in the order the generated, to a computer Generation program.

Images