JP2017058406A - Computer system and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a conversation specific to a specific use required by a user, with the use of a software agent.SOLUTION: Customization information including utterance contents and the like used for a specific use, such as reception in an office, is stored in a database (conversation DB) of a storage device by a user, voice of a visitor is analyzed by voice recognition processing (S3) and utterance contents matching the analysis result are retrieved from the DB of the storage device to utter by utterance processing (S5).SELECTED DRAWING: Figure 3

Description

  The present invention relates to a computer system and a program for responding to a person by voice using a software agent, for example.

  What is generally known about this type of computer system and program, for example, when a user speaks a question to a smartphone equipped with an interactive system, the smartphone recognizes the user's voice and means Some analyzes understand and respond to the intention of the question, and automatically start an optimal service function (for example, mail, scheduler, alarm, etc.) (for example, see Non-Patent Document 1).

NTT DoCoMo, “Talking Concierge”, [online], [Search March 4, 2015], Internet <URL: http://www.nttdocomo.co.jp/service/information/shabette_concier/>

  However, such a dialogue system is a general-purpose system developed for general users, such as accepting visitors at the company, accepting visit reservations, or interacting with care recipients, etc. There is an inconvenience that it is not suitable for a dialog specialized for a specific application.

  The present invention has been made in view of the above circumstances, and an object thereof is to provide a computer system and a program that enable a software agent to realize a dialog specialized for a specific use requested by a user.

  In the following, an example of an embodiment corresponding to each means for solving the problem is shown in parentheses.

The present invention is a computer system that responds to a person by voice using a software agent,
Voice recognition means for recognizing human voice (for example, voice input units 31, 44, S3, S25 to S30, S51, S111, S214, S290);
Speaking means (for example, voice output units 31, 44, S5, S40 to S44, S53, S90 to S97, S113, S145 to S152, S216, S255 to S291) for performing speech corresponding to the recognition result by the voice recognition means. , S298 to S302),
Storage means for storing information for generating utterance contents as a database (for example, storage device 24, EEPROM 36, S1, S10 to S12, S50, S70 to 74, S110, S210, S212, S225 to S230, S231 to S236, S292, S310 to S311),
Utterance content generation means for generating utterance content using information stored in the storage means (for example, CPU 21, 33, S4, S35 to S37, S52, S80 to S86, S112, S215, S245 to S249, S280 to S282, S293, S305 to S308),
The utterance means performs an utterance according to the utterance content generated by the utterance content generation means (for example, S40 to S44, S145 to S152, S255 to S291, S298 to S302),
The storage means is capable of inputting and storing information customized for a specific application using the software agent (for example, S10 to S12, S70 to 74, S225 to S230, S231 to S236, S310 to S311). ).
Further, the voice recognition means includes a discrimination means (for example, a microphone array) for distinguishing the user's voice from other voices, performs voice recognition based on the voice distinguished by the discrimination means, and the discrimination means Has a microphone array for performing sound source separation based on sound signals collected by a plurality of microphones (for example, two-point sound source separation by simultaneous equation solution between frames) and removing sound signals from a predetermined distance May be. The distinction unit may further include a sound source separation unit using independent component analysis.
The speech recognition means is a means for modifying a general recognition standard including acoustic patterns of vowels and consonants stored in advance, and based on acoustic properties included in speech collected as a sample from a user. The general recognition standard may be modified to include speaker adaptation means for performing speaker adaptation.
The speaker adaptation may be speaker adaptation in speech recognition using a Gaussian mixture distribution model.
The speaker adaptation may be performed using a deep neural network.
The storage means may include an input operation means for the user to input information customized for a specific application using the software agent on a management screen (for example, the management screen shown in FIG. 20).
This input operation means displays an utterance content area (for example, a balloon in FIG. 20) on the management screen and inputs the utterance content to the user in the utterance content area, and also utterance content area of the question to the user Branching according to the response contents of the user (for example, branching between YES arrow and NO arrow in FIG. 20), and further utterance content areas are displayed at each branch destination, and the response contents are included in the utterance content area. A flow diagram type input means for inputting the corresponding utterance content, wherein the utterance content generation means includes means for sequentially generating the utterance content in the utterance content area along the flow diagram input by the flow diagram type input means. Good.
The voice recognition means may include synonym recognition means for recognizing synonyms by equivalence. The synonym recognition means may recognize and synonymize synonyms based on a synonym database (for example, synonym list data stored in a buffer) input in advance.

Further, the storage means stores the name of the employee and the telephone number for calling the employee in association with each other in order to use the software agent for accepting the visitor (for example, S11, FIG. ) "Call DB")
When the name of the interviewee spoken by the visitor is recognized by the voice recognition means, an automatic for indexing the telephone number corresponding to the name from the storage means and calling the employee of the telephone number by telephone Calling means (for example, S6) may be further provided.

Further, the storage means stores a reservation frame for accepting a reservation of a visitor (for example, S63 to S65, S72, S73),
The utterance content generation means calculates the reservation frame from the stored information of the storage unit by recognizing the voice of the visitor applicant that has been transmitted online and that the user wants to make a reservation. The utterance content may be used (for example, S80 to S83).

Data storage means (for example, S120, S183, learning DB 14) for collecting and storing dialogue history data consisting of recognition results by the voice recognition means and utterance contents by the utterance means from a plurality of software agents via a network; ,
Machine learning means (for example, S186, S187, S200) for machine learning using the accumulated data by the data accumulation means as machine learning data;
Providing means (for example, S201) for providing a machine learning result by the machine learning means to the software agent;
The software agent performs control for reflecting the provided machine learning result on generation of utterance content by the utterance content generation means (for example, S205 to S07).

Further, the storage means stores dialogue information used for dialogue with a care recipient in order to use the software agent for nursing care (for example, S1, FIG. 9A),
The utterance content generation unit may calculate the utterance content for the care recipient in accordance with the speech recognition result of the care recipient by the voice recognition unit from the storage information of the storage unit (for example, S112).

In addition, the storage unit stores information for dialogue related to information in the website (for example, S227, S228, S223, S234),
The utterance content generation means determines the answer to the question from the stored information of the storage means by the voice recognition means recognizing the voice of the question transmitted online from the website viewer, and the utterance It may be content (for example, S246, S247, S249).

  The software agent (for example, site agent 47) matches the viewer in cooperation with an agent (for example, user agent 48) that stores the profile information of the website viewer as knowledge. The answer may be indexed from the stored information of the storage means to make the utterance content (for example, S281, S282).

The storage means stores a plurality of answer data for the user (for example, the contents of the answer in FIG. 17B), and a plurality of types of attributes for narrowing down a suitable one from the plurality of answer data. (For example, weighting data (for example, weighting in FIG. 17 (b)) in FIG. 17 (b) is stored, and each answer for the plurality of types of attributes is stored. Weighting data corresponding to the contents of the data for use is stored in association with the corresponding answer data (for example, A: 5, 5, 1, 1, 1..., B: 1, 1 in FIG. 17B). , 5,5,1 ..., C: 1,1,5,1,5 ...),
The utterance content generation means, when the utterance means sequentially asks the user about the plurality of types of attributes (for example, S299), reply data close to weighting for each attribute obtained by the user's reply to the utterance May include narrowing means (for example, S305 to S307) for narrowing down from the plurality of answer data to the utterance content.

  In addition, the storage means includes artificial storage means (for example, S310, S311) for artificially inputting the answer data and storing it in the database. When the utterance means sequentially asks about the attribute of the type (for example, S299), and the weight for each attribute is acquired by the answer to the utterance, the artificially input answer data is obtained. May be stored in association with each other (for example, S311).

  Moreover, you may memorize | store the data for the reply with respect to the consultation from the employee in a company as said some data for a reply (for example, FIG.17 (b)).

Further, it further comprises weight correction means (for example, S294) for correcting the weight stored in the storage means by the artificial storage means,
The weight correction means, when the answer data desired by the user is narrowed down by the narrowing means (for example, when YES is determined in S315), each attribute used for narrowing down by the narrowing means It may be configured to correct the weight for the answer data desired by the user based on the weight for (for example, S316, S317).

Further, the present invention is a program for responding by voice to a person using a software agent,
On the computer,
Voice recognition means (eg, S3, S25 to S30, S51, S111, S214, S290) for recognizing human voice;
Speaking means (for example, S40 to S44, S53, S90 to S97, S113, S145 to S152, S216, S255 to S291, S298 to S302) for speaking to a person,
Storage means (for example, S1, S10 to S12, S50, S70 to 74, S110, S210, S212, S225 to S230, storing information for generating utterance contents corresponding to the recognition result by the voice recognition means as a database. S231 to S236, S292, S310 to S311),
Utterance content generation means (for example, S4, S35 to S37, S52, S80 to S86, S112, etc.) for generating the utterance content corresponding to the recognition result by the voice recognition means using the information stored in the storage means. S215, S245 to S249, S280 to S282, S293, S305 to S308),
The utterance means performs an utterance according to the utterance content generated by the utterance content generation means (for example, S40 to S44, S145 to S152, S255 to S291, S298 to S302),
The storage means is capable of inputting and storing information customized for a specific application using the software agent (for example, S10 to S12, S70 to 74, S225 to S230, S231 to S236, S310 to S311). ).

  ADVANTAGE OF THE INVENTION According to this invention, when the user inputs the information customized for a specific use, the dialog specialized for the specific use which a user requests | requires is attained.

It is a system diagram which shows the whole structure of an artificial intelligence service. (A) is a block diagram which shows the control circuit of various servers and a personal computer (PC), (b) is a block diagram which shows the control circuit of a care terminal and a user terminal. (A) is a flowchart showing a main program of the reception PC, (b) is a diagram showing data stored in the database of the reception PC, and (c) is a flowchart showing a subroutine program for customizing information input processing. is there. (A) is a flowchart showing a subroutine program for face recognition processing, (b) is a flowchart showing a subroutine program for speech recognition processing, (c) is a flowchart showing a subroutine program for DB detection processing, (d ) Is a flowchart showing a subroutine program of speech processing. (A) is a flowchart showing the main process of the visitor reservation web server, and (b) is a flowchart showing the control operation of the reservation frame setting PC and the subroutine program of the customization information receiving process in the customization reservation web server. Yes, (c) is a diagram showing the storage data of the reservation frame in the DB of the customer reservation web server. (A) is a flowchart which shows the subroutine program of DB detection processing, (b) is a figure which shows the memory | storage data of DB (utterance) of the web server for customer reservations. (A) is a flowchart showing a subroutine program for speech processing, (b) is a flowchart showing a subroutine program for screen control processing, and (c) is a flowchart showing a subroutine program for calendar synchronization processing. It is a flowchart which shows control operation | movement with the terminal for care recipients, a centralized management server, a wireless network, and a user terminal. (A) is a figure which shows the memory content of DB (utterance) of a centralized management server, (b) is a flowchart which shows the subroutine program of speech processing. (A) is a flowchart which shows the subroutine program of a video chat process, (b) is a flowchart which shows the subroutine program of a website browsing process. It is a flowchart which shows the control operation | movement of PC and an artificial intelligence server (care learning process) of a special trader (care provider). It is a flowchart which shows the control operation | movement of an artificial intelligence server (other process) and a terminal for a care recipient (other process). (A) is a flowchart showing control operations of the company PC, site agent web server, and user terminal, and (b) is a flowchart showing a subroutine program for customization information input processing and a subroutine program for customization information storage processing. is there. It is a figure which shows the memory | storage data of customization DB in a site agent server. (A) It is a flowchart which shows the subroutine program of web page display processing, (b) is a flowchart which shows the subroutine program of DB search processing, (c) is a flowchart which shows the subroutine program of speech processing. (A) is a flowchart showing the control operation of the company PC (other processing) and the site agent web server (other processing), (b) is a flowchart showing the control operation of the DB search processing (developed type), (C) is a diagram showing a multi-agent system including a site agent and a user agent. (A) is a flowchart which shows the main program of employee consultation PC, (b) is a figure which shows the memory | storage data of consultation DB of employee consultation PC. (A) is a flowchart showing a subroutine program for speech processing, (b) is a flowchart showing a subroutine program for DB detection processing, (c) is a flowchart showing a subroutine program for customization information input processing, ( d) is a flowchart showing a subroutine program of feedback learning processing. It is a flowchart which shows control operation of an up store server and a user terminal. It is a management screen figure which shows the modification of the information input process for customization.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. First, the overall configuration of the artificial intelligence service will be described with reference to FIG. With respect to the Internet 1, a personal computer for company reception (hereinafter “PC”) 2, a reservation frame setting PC 3, a care recipient terminal 4, a company PC 7, an employee consultation PC 8, a user terminal 17, The wireless network 6 installed in the home requiring care 5, the various servers 10 to 13 of the artificial intelligence service company 9, the PC 16 of various specialists, and the upstore server 19 are connected so that they can communicate with each other. It is configured.

  The company reception PC is equipped with a reception agent as a software agent, and is installed at the company reception and has a function of acting as a receptionist. The reservation frame setting PC 3 is a PC that accepts patient reservations at a dental clinic or the like, or accepts visitor reservations at a restaurant.

  The care recipient terminal 4 is equipped with a care agent as a software agent, is composed of a tablet terminal, etc., is installed in the care recipient's home 5, and talks with the care recipient to care for the care recipient. belongs to. The terminal 4 for a care recipient is not limited to a care recipient, and may be a device for talking with an elderly person. In addition, the wireless network installed in the home requiring care 5 is a wireless network in which a plurality of wireless terminals with sensors are scattered in the space and the environment and physical conditions can be collected in cooperation with each other. It is. For example, the sensor device is made of energy harvesting, M2M or a battery. A wireless sensor network typically has an ad hoc function and a routing algorithm for sending data from each node to a central node. That is, there is a function of autonomously reconstructing another communication path when a failure occurs in communication between nodes. There are also elements of distributed processing because the nodes cooperate as a group. In addition, there is also a function that operates for a long time without receiving power supply from the outside. Examples of the sensor include a door opening / closing sensor that detects opening / closing of a door, a temperature sensor that detects indoor temperature, a humidity sensor that detects indoor humidity, a human sensor that detects humans, and a sleep that detects human sleep state. A sensor or the like is conceivable. Based on detection signals from these sensors, it is possible to confirm (monitor) the living situation such as the behavior of the care recipient.

  The company PC 7 is used when creating the website of the company. The employee consultation PC 8 is equipped with artificial intelligence, and is used to find and answer an answer when an employee asks a question about the company. The user terminal 17 is a smartphone, a tablet, or a personal computer possessed by the user.

  The artificial intelligence service company 9 is provided with a visitor reservation web server 10, a site agent web server 11, a centralized management server 12, and an artificial intelligence server 13. The visitor reservation web server 10 is equipped with a reservation agent as a software agent for displaying on the web a homepage of a vendor that has set a reservation frame based on the reservation frame set by the reservation frame setting PC 3. Is. Customers who browse the homepage can make reservations for the visit date and time by voice as will be described later.

  The site agent web server 11 is equipped with a site agent as a software agent, and displays the website of the company on the web based on company information input and set by the company PC 7. A character representing the site agent is displayed on the web, as will be described later, and the character answers the voice by a visitor talking to the character.

  The central management server 12 collects the conversations performed between the terminal 4 for the care recipient and the care recipient from each terminal 4 for the care recipient and stores them in the database, and also stores each care recipient home 5. The sensor detection data from the wireless network 6 installed in is collected and stored in a database. A care manager, a helper, a caregiver, or a family member of a care recipient operates the user terminal 17 to access the central management server 12, so that the conversation contents of the care recipient stored in the database of the central management server 12 are stored. Or sensor detection data from a wireless sensor network. The trajectory (history) of the behavior of the care recipient can be found from the sensor detection data from the wireless sensor network

  The artificial intelligence server 13 periodically sends the data recorded in the database of the central management server 12 and stores it in the learning database (hereinafter referred to as “DB”) 14 and has been transmitted from the PC 16 of the care provider. Data specifying the condition of the care recipient is stored in the learning DB 14. Then, the artificial intelligence server 13 performs machine learning using the both stored data as machine learning data, and stores the learning result in the artificial intelligence DB 15. For example, recent conversation data of the care recipient and sensor detection data from the wireless sensor network 6 are extracted based on data specifying that the condition of the care recipient deteriorated sent from the PC 16 of the care provider. Then, data mining is performed to find a pattern (NGPi) that is a precursor to the condition of the care recipient being deteriorated, and the NG pattern is stored in the artificial intelligence DB 15. Further, the utterance content for finding out whether or not the care recipient is applied to the NG pattern is generated and transmitted to each terminal 4 for the care recipient. The terminal 4 for the care recipient performs a conversation reflecting the received utterance content with the care recipient.

  Furthermore, the artificial intelligence server 13 transmits the learning result data with the above-mentioned NG pattern and the utterance content generated from the NG pattern to the up-store server 19 of the learning result up-store 18 and performs the process of registering in the DB 20.

  The up-store server 19 classifies the transmitted learning results by genre and registers them in the DB 20. For example, in the care classification, NGPi (i = 1, 2, 3...) Is used as the NG pattern, and “question of interest / interest” is used as customized data such as utterance contents derived from the NG pattern. . Other genres (classifications) include depression, manic depression, psychosis care, early childhood education, and the like. As described above, the learning result upstore 18 stores the learning results obtained by artificial intelligence classified by genre, and the user who wants to purchase the learning results upstores from the user terminal 17 or the PC 16 of various specialists. The server 19 is accessed, the learning result for each genre registered in the DB 20 is browsed, and if the learning result desired to be purchased is found, the learning result data is downloaded and used for its own artificial intelligence. As a result, the user can save the trouble of learning the artificial intelligence from the beginning, and there is an advantage that the ready-made machine learning result data can be downloaded and used for his / her artificial intelligence.

  Company reception PC2, reservation frame setting PC3, company PC7, employee consultation PC8, visitor reservation web server 10, site agent web server 11, centralized management server 12, artificial intelligence server 13, various professional PC16, and A hardware configuration of the upstore server 19 will be described with reference to FIG. Since these servers and PCs have the same hardware configuration, a representative example will be described with reference to FIG.

  These servers and PCs have a CPU (Central Processing Unit) 21 as a control center, a RAM (Random Access Memory) 22 that functions as a work area of the CPU 21, a ROM that stores a control program and control data. Read Only Memory) 23 and a storage device 24 comprising a hard disk or the like are provided. The CPU 21, RAM 22, ROM 23, and storage device 24 are connected to various external devices via a bus 25 and an interface unit 26.

  Specifically, a digital video camera input unit 27 that captures images of visitors and the like, a communication unit 28 for communicating with various servers via the Internet 1, a display unit 29 for displaying various information, and a user manually An input operation unit 30 such as a keyboard for inputting information, a voice input unit 31 for inputting user's voice, and a voice output unit 32 for speaking to the user by voice are provided.

  Next, the hardware configuration of the care giver terminal 4 and the user terminal 17 will be described with reference to FIG. Since these terminals have the same hardware configuration, a representative example will be described with reference to FIG.

  These terminals include a CPU (Central Processing Unit) 33 as a control center, a RAM (Random Access Memory) 35 functioning as a work area for the CPU 33, a ROM (Read Only) storing a control program and control data. An EEPROM (Electrically Erasable Programmable Read-Only Memory) 36 capable of reading and writing is provided. The CPU 33, RAM 35, ROM 34, and EEPROM 36 are connected to various external devices via a bus 37 and an interface unit 38.

  Specifically, a digital video camera input unit 39 that captures images of care recipients, a wireless communication processing unit 41 for communicating with various servers via the Internet 1, a display unit 44 for displaying various information, and a user Input operation unit 42 such as a touch panel for manually inputting information, voice input unit 43 for inputting user's voice, voice output unit 43 for speaking to the user by voice, GPS information from satellite and base station A position information acquisition unit 45 is provided for acquiring the current position based on the radio wave from the wireless LAN access point and the radio wave from the wireless LAN access point.

  Although the server, PC, and terminal described above use a Neumann type general computer, a neural network processor (NNP) may be used. A large number of “artificial neurons” modeled on real neurons are mounted on the NNP chip, and each neuron cooperates in a network.

  Further, a quantum computer employing a “quantum annealing method” may be used. Thereby, the time required for the optimization calculation in machine learning can be greatly shortened. “Artificial intelligence” is a broad concept that includes software agents.

  Next, the main program of the company reception PC 2 will be described with reference to FIG. The CPU 21 of the receiving PC performs customization information input processing in step S (hereinafter simply referred to as “S”) 1. This customization information input process is a process for inputting information such as utterance contents customized so as to be suitable for the company that accepts the information. This process enables a dialog customized for the company to visitors who have visited the company. Also, by this customization information input process S1, the employee's mobile number is input and set, and the visitor automatically calls the employee by calling the mobile number of the employee who wants to have an interview. An example of the customization information input by the customization information input process (S1) is shown in FIG.

  Next, the reception PC 2 performs a face authentication process S2. This face authentication process S2 is for recognizing the face of the visitor and activating the reception PC 2 from the sleep state. Next, voice recognition processing S3 is performed. This voice recognition process is a process for recognizing the voice of the visitor and finding out the name of the person in charge of the interviewer. Next, DB search processing S4 is performed. This DB search process S4 is a process for determining the utterance content and the employee's mobile number stored in the DB of the storage device 24. Next, an utterance process S5 is performed. This utterance process is for retrieving the utterance content stored in the DB of the storage device 24 and responding to the visitor by voice based on the retrieval result. Next, employee call processing S6 is performed. This employee calling process S6 is a process for automatically calling an employee based on the mobile phone number corresponding to the visitor's request name.

  Next, the control operation of the subroutine information input processing for customization in S1 will be described with reference to FIG. First, in S10, it is determined whether or not an input operation for customization information has been performed. If not, the customization information input process ends. If the operator of the company operates the input operation unit 30 in advance and inputs the customization information, the control advances to S11, and the input utterance number and the utterance content are associated with each other in the DB of the storage device 24 (FIG. 3). The process of storing in (b) is performed. Next, in S12, it is determined whether or not an end operation has been performed, and the storage process in S11 is repeatedly executed until the end operation is performed. Then, when the operator finishes the operation, a determination of YES is made from S12, and the customization information input process is completed.

  The result of the customization information input process is shown in FIG. The utterance number input by the operator and the utterance contents are associated with each other and stored in the utterance DB. For example, in the utterance number 1 "Hello. Today Thank you for coming to our office." In the speech number 2 "Please tell us the name of the person in charge." In the speech number 3 was "I know. I connect your phone Please wait for a while. "

  Also, as the mobile number input by the operator, in the calling DB, for example, Tanaka: 090- ○ △ ×-× ○○, Wada: 090- △ × ○-△△ ×, Nakajima: 090- ○ × △-○ × △ And so on.

  Next, the operation of the subroutine program of the face authentication process S2 will be described with reference to FIG. In S15, it is determined whether or not the utterance number is “1” or more. If the utterance number is “1” or more, the face authentication process is ended. This utterance number is designated as “1” by S19. At a stage where the utterance number is not yet designated as “1”, the control proceeds to S16, and a process of analyzing the input image from the camera input unit 27 is performed. If it is determined that the face is not a human face, the control advances to S18, and a message “Please bring your face closer to the camera” is displayed on the display unit 29. On the other hand, if it is determined in S17 that the face is a human face, the control advances to S19, and processing for designating the utterance number as “1” is performed. As a result, at the next stage of executing this face authentication process, a determination of YES is made in S15, and the message “Please bring your face closer to the camera” in S18 is not displayed.

  Next, the operation of the subroutine program of the speech recognition process S3 will be described with reference to FIG. First, in S25, it is determined whether or not there is a voice input from a visitor. If there is no voice input yet, this voice recognition process is terminated. On the other hand, if the visitor speaks to the reception PC 2, YES is determined in S25, and the control advances to S26, morphological analysis processing is performed, syntactic analysis processing is performed in S27, semantic analysis processing is performed in S28, and S29. Then, context analysis processing is performed, and intention analysis processing (intention interpretation processing) is performed in S30. These analysis processes are executed by a natural language processing engine.

  Morphological analysis is to divide a sentence into meaningful words and discriminate parts of speech and contents using a dictionary. Parsing is a process for determining the dependency relationship between clauses. Semantic analysis is a natural language processing technique that uses a vector space model, and analyzes the relationship between a group of documents and a group of terms contained therein by generating a set of related concepts. That is. Context analysis is to check the connection of multiple sentences. Intention analysis (intention interpretation) is a process of understanding the meaning of a sentence.

  Next, the operation of the subroutine program in the DB search process S4 will be described with reference to FIG. First, in S35, it is determined whether or not there is a voice recognition result. If there is no voice recognition result, the DB search process is terminated. If there is a result of speech recognition by the speech recognition process S3, the process proceeds to control S36, and a process of determining an utterance number corresponding to the analysis result by the speech recognition process described above is performed. This process is a process of finding a matching response from among a large number of dialogue templates collected in order to find a response to the user's dialogue by template matching, and determining an utterance number corresponding to the response. Next, in S37, a process of searching and determining the utterance content corresponding to the determined utterance number from the dialogue DB (see FIG. 3B) is performed.

  Next, the control operation of the subroutine program in the speech process S5 will be described with reference to FIG. In S40, it is determined whether or not there is a searched utterance content. If there is no utterance content, this utterance process is terminated. On the other hand, if there is an utterance content retrieved in S37, the control advances to S41, and a process of uttering according to the utterance content is performed. As a result, for example, "Hello. Today Thank you for coming to our office." If the retrieved speech content of the utterance number 1 speech is emitted from the audio output unit 32 (see Figure 3 (b)).

  Next, the process of clearing the utterance content uttered from S41 is performed in S42, and it is determined whether or not the dialog is ended in S43, and this utterance is determined if the dialog is not yet ended. Processing ends.

  On the other hand, if the dialogue has ended, the control advances to S44 to perform a process of clearing the utterance number to zero. As a result, NO is determined in S15 described above, and the face recognition processing from S16 onward is executed.

  The following can be considered as the advanced types of the reception agent described above. The face image data of the visitor recognized by the face recognition process of FIG. 4A and the person in charge interviewed by the visitor at that time and the information input by the person in charge about the visitor are associated and stored in the DB. In addition, the DB may be searched from the face data input from the camera input unit at the time of revisiting, and the reception agent may be controlled to interact based on the stored information regarding the revisiting person.

  Next, the control operation of the main program of the visitor reservation web server 10 will be described with reference to FIG. First, in S50, customization information reception processing is executed. This customization information receiving process is a process of receiving the reservation frame data of each person in charge input from the reservation frame setting PC 3 and registering it in the DB (reservation frame) of the storage device 24. Next, voice recognition processing is performed in S51. This voice recognition process is the same as that shown in FIG. 4B, and detailed description will not be repeated. Next, DB search processing is performed in S52. This DB search process is a process of searching for a reservation frame of each person in charge registered in the DB (reservation frame) in S50. The speech process is performed in the next S53. This utterance process is a process for responding to a customer who visits a home page by voice. Details will be described later with reference to FIG.

  Next, a screen control process is performed by S54. This screen control process displays a calendar or a list of persons in charge for visitors to the home page. Details will be described later. Next, calendar synchronization processing is performed in S55. This calendar synchronization process is a process for recording a reservation date and time in a calendar when the date and time of the reservation is determined through a dialogue with a visitor on the home page.

  Next, the control operation of the subroutine information receiving process for customization information in S50 will be described with reference to FIG. In this customization information reception process, the reception process is performed by communicating between the reservation frame setting PC 3 and the visitor reservation web server 10. First, in the reserved frame setting PC 3, it is determined whether or not a login operation has been performed in S60. If there is no login operation, the control process of the reserved frame setting PC 3 ends. On the other hand, if the user accesses the visitor reservation web server 10 and performs a login operation, the control proceeds to S61, and the login ID and password are transmitted from the reservation frame setting PC 3 to the visitor reservation web server 10. The visitor reservation web server 10 receives the transmission information in S70, confirms the login ID and password in S71, and returns the input page to the reservation frame setting PC 3.

  The reservation frame setting PC 3 receives the input page in S62, and receives an input operation of an operator who inputs a reservation frame in accordance with the received input page in S63. Then, in S64, it is determined whether or not the operator's input operation has been completed. If the operator has not completed the input operation, the reservation frame input operation in S63 is continued. When the operator's input operation is completed, the control proceeds to S65, and the input reservation frame data is transmitted to the visitor reservation web server 10. The visitor reservation web server 10 receives the reservation frame data in S72, and stores the received reservation frame in a DB (reservation frame) in S73. The storage state of the reservation frame (calendar) is shown in FIG. In the storage device DB (reservation frame), for example, reservation frame data is stored for each day such as March 18, March 17, March 16, and so on. The reservation frame data for each day is data for specifying the reservation status (vacancy status) from 9:00 to 18:00 for each person in charge. This reservation frame (calendar) is published on the website every day (S74).

  The operation of the above-described DB search subroutine program in S52 will be described with reference to FIG. First, in S80, it is determined whether or not there is a voice recognition result. If there is no voice recognition result, the DB search process is terminated. When a user who has made a reservation operates the user terminal 17 to access the visitor reservation web server 10 and visits a website published in S74, and makes a reservation procedure by voice, the voice recognition process S51 Voice recognition is performed. As a result, a determination of YES is made in S80, and the process proceeds to S81, in which it is determined whether the person in charge and the applicant for reservation are recognized. If the user has not yet spoke with both the person in charge and the person who wants to make a reservation, a NO determination is made at S81 and control proceeds to S85. In S85, it is determined whether or not the reservation time has been recognized. If the user has not yet spoken of the reservation time, the control proceeds to S84.

  In S84, a process for determining an utterance number corresponding to the analysis result in the voice recognition process S51 is performed. For example, when the user speaks that he / she wants to make a reservation, the utterance number “1” is determined, and the utterance content “I understand. When is the day?” Is selected and uttered by the utterance processing S53. If the user speaks the reservation date and time, it is recognized by the voice recognition processing S51, and then the utterance number “2” is selected and decided, and the utterance content “Is there a request from the person in charge?” Is selected. Spoken by the utterance process S53.

  If the user who has heard it specifies, for example, Nakajima-sensei as a person in charge by voice, it is recognized by the voice recognition processing S51, and YES is determined by S81. As a result, the control advances to S82, and a process of searching the DB (reservation frame) based on the person in charge (for example, Dr. Nakajima) and the desired reservation date (for example, March 15) to determine the empty frame is performed. Referring to the DB (reservation frame) in FIG. 5 (c), Nakashima-sensei's free space on March 15 is from 13:00 to 18:00. As a result, in S83, the utterance number “3” is determined, and the word / phrase for filling the blanks a to d in the utterance content corresponding to the utterance number 3 is determined. Specifically, “Nakajima” is determined in the blank a, “March 15” in the blank b, “13:00” in the blank c, and “18:00” in the blank d. The utterance content of the utterance number 3 in which the determined phrase is filled in the blanks a to d is uttered by the utterance process S53. As a result, the content of the utterance is “Invited. Prof. Nakajima is available from 13:00 to 18:00 on March 15.” If the user who hears it says that he / she wants to make a reservation from 13:00, it is recognized by the voice recognition processing S51, a YES determination is made in S85, the control advances to S86, and the utterance number “4” is selected and determined. The phrase “13” to be filled in the blank e of the utterance content of the utterance number “4” is determined, and the utterance content in which “13” is buried in the blank e is “Acknowledged. Is uttered to the user by the utterance process S53.

  Next, the control operation of the subroutine program for the speech process in S53 will be described with reference to FIG. First, in S90, it is determined whether or not there is a voice recognition result, and if not, this utterance process is terminated. If there is a voice recognition result, the control advances to S91 to determine whether or not the utterance number is “3”. If the utterance number is not “3”, the control advances to S92 to determine whether or not the utterance number is “4”. If it is not “4”, the control advances to S93, and a process of uttering the utterance content of the utterance number is performed. The utterance number by S93 is a number other than “3” and a number other than “4”, such as “1” or “2”.

  If the utterance number is 3, the control advances to S94, where the speech contents of the utterance number 3 are filled with blanks a to d with the determined words and uttered. If the utterance number is “4”, the control advances to S96, where the blank e is determined and the utterance content of the utterance number 4 is uttered.

  After the processes of S94, S93, and S96 are completed, the control shifts to S95, and it is determined whether or not the dialogue is completed. If the dialogue has not ended, the utterance processing ends. However, when the dialogue ends, the control advances to S97, and after the processing for clearing the utterance number determined by S83, S86, etc. is performed, the utterance processing is performed. Processing ends.

  Next, the control operation of the subroutine program of the screen control process of S54 described above will be described with reference to FIG. First, in S100, it is determined whether or not the utterance number is “1”. If it is not “1”, it is determined whether or not the utterance number is “2” (S101). If it is not “2”, the screen control process ends.

  If the utterance number is “1”, control proceeds to S102, and control for displaying the calendar on the user terminal 17 is performed. As a result, the utterance of the utterance number “1” “acknowledged, when is the date” flows from the user terminal 17 and at the same time, the calendar is displayed on the display unit 40 of the user terminal 17. The user speaks toward the user terminal 17 while looking at the displayed calendar.

  On the other hand, when the utterance number is “2”, the control proceeds to S103, and a control for displaying the person-in-charge list on the display unit 40 of the user terminal 17 is performed. As a result, the user can designate the desired person in charge while browsing the person-in-charge list.

  Next, the control operation of the subroutine program of the calendar synchronization process of S55 described above will be described with reference to FIG. First, in S105, it is determined whether or not the utterance number is “4”. If it is not “4”, the calendar synchronization processing is terminated. When the utterance number is “4”, as described above, the voice of “I understand. I will make a reservation from 13:00 tomorrow” is heard from the user terminal 17, and the control advances to S106. An update process is performed to fill the DB (reservation frame) (see FIG. 5C) with a newly entered reservation, and a command for recording the reservation date and time on the calendar of the reservation person's terminal (user terminal 17). Processing to transmit to the user terminal 17 is performed.

  Next, the control operation of the main program of the terminal 4 for care recipient will be described based on FIG. The terminal 4 for the care recipient performs operation control for the care recipient in cooperation with the central management server 10. The central management server 10 executes control operations for the care recipient while exchanging information with the wireless sensor network 6 of the care recipient home 5.

  First, the care recipient terminal 4 executes customization information input processing in S110, performs voice recognition processing in S111, executes DB search processing in S112, executes speech processing in S113, and performs video in S114. Chat processing is executed, website browsing processing is executed in S115, and other processing is executed in S116. The customization information input process of S110 is a process of inputting the utterance content and the like necessary to care for the care recipient who is the owner of the terminal 4 for the care recipient. The actual input operation is performed by a care manager, a helper, or a care provider. When it is performed by a care provider, it may be input to the care provider's PC 16 and transmitted to the central management server 12. Since the voice recognition process in S111 and the DB search process in S112 are almost the same as those already described, detailed description is not repeated here. The website browsing process of S115 is a process for browsing a photo sharing site with a family, a site uploading memorable photos or videos, and the like. The other processing of S116 is for the care recipient to call the care manager, helper, or family member in an emergency or the like and exchange voice and video.

  When a care agent installed in the terminal 4 for a care recipient and a care recipient interact, the content spoken by the care recipient is voice-recognized in S111, and for the care recipient, S113. Reply with voice by utterance processing. Dialogue data with the care recipient that is processed by the voice recognition processing at S111 and the speech processing at S113 is transmitted from the care recipient terminal 4 to the central management server 12. In the central management server 12, the transmitted dialogue data is received in S120 and recorded in the database.

  On the other hand, in the wireless network 6 installed in the care recipient's home 5, a process of collecting detection data from each sensor assigned with a sensor ID and storing the detection data in association with each sensor ID is performed. In S136, it is determined whether or not there is an emergency transmission in the detected data. If not, the process proceeds to S139 to determine whether or not the periodic transmission timing of the detection data has come. If it is not the transmission time, the control returns to S135 and the detection data storing process is continued. In the course of the loop of S135 → S136 → S139 → S135, when there is an emergency transmission in the detection data, or when the regular transmission time comes, the process proceeds to control S137, and the collected detection data is set for each sensor ID. Processing to transmit to the central management server 12 is performed. In S138, the memory of the detected data is cleared, and then the process returns to S135.

  In the central management server 12, the sensor detection data transmitted from the wireless sensor network is received in S121, and the wireless sensor data is recorded in the database. Next, in S122, there is data to be dealt with urgently. If not, the process returns to S120. On the other hand, when data that should be dealt with urgently is found, the control proceeds to S123, and processing for notifying various contacts is performed. Specifically, when there is abnormal data such as when the temperature sensor detects an abnormally high temperature or when the human sensor does not detect a care recipient at all after a predetermined time, a care manager or helper or Notify the caregiver and the hospital or family that there is a risk of abnormality.

  On the other hand, if the family member of the care recipient, the care manager, or the helper operates the user terminal 17 to access the central management server 12 and requests to view the recorded information of the care recipient, YES is determined in S130. Then, the control proceeds to S131, and a browsing request is transmitted to the central management server 12. In the centralized management server 12, the browsing request is received in S124, and the process of transmitting the interaction information with the care recipient recorded in the database and the sensor detection information from the wireless sensor network to the user terminal 17 is executed in S125. The The user terminal 17 that has received the transmission data in S132 performs control to display the received recording data on the display unit 40. As a result, there is an advantage that even the family of a care recipient who lives in a remote place can grasp the state of the care recipient.

  As a result of the customization information input process in S110 described above, the data of the utterance content as shown in FIG. 9A is stored in the DB (utterance) of the EEPROM 36 of the care recipient terminal 4 in association with the utterance number.

  Next, the control operation of the subroutine program in the above-described utterance process of S113 will be described with reference to FIG. First, it is determined whether or not there is a voice recognition result in S145, and if not, it is determined whether or not the utterance time has come in S146, and if not, whether or not the utterance is ended in S147. If this determination is made and the processing has not ended, the utterance processing ends.

  If the utterance time set in advance by the customization information input process (S110) is reached, a determination of YES is made in S146, and the process proceeds to S148, where the utterance is selected and the utterance is selected according to the current time. The For example, the utterance number “1” is selected when it is time for the care recipient to wake up in the morning (for example, 7:00), and the voice “Wake up. It's morning.” Shown in FIG. It is emitted from the voice output unit 43 of the person terminal 4.

  If the care recipient speaks to the care recipient terminal 4, the voice is processed by the voice recognition process (S 111), a determination of YES is made in S 145, the control advances to S 149, and DB (utterance) is made in S 112 described above. Control is performed to utter the utterance content retrieved from. For example, the following dialogue is possible according to the utterance content stored in the DB (utterance) in FIG.

Care Agent: Get up. It's morning.
Care recipient: Good morning. It ’s good weather today.
Care Agent: Did you sleep well this morning?
Caregiver: Yes, I was able to sleep well.
Nursing agent: Did you eat breakfast?
Care recipient: Yeah.
Care Agent: Did you take medicine?
Care recipient: Yes. Your medicine is almost gone.
Care Agent: I understand. Report to care manager. The room temperature is too high. Please turn on the air conditioner.
Yes, I turned on the air conditioner. I'd like to video chat with my son after a long time.
Care Agent: I understand. Displays the video chat screen.
Caregiver: I want to browse a photo sharing site this time.
Care Agent: I understand. Browse photo sharing sites.

  In addition, the care agent may be controlled so that the care recipient is informed of the name and arrival time of the helper by inputting and setting the helper schedule at the office. The helper may control to send the business report to the office or the family of the care recipient by speaking to the care recipient terminal 4 and reporting the business. Furthermore, the care agent may transmit a helper's work report or business report to the family of the care recipient. Furthermore, the care agent may report to the family confirmations such as “whether or not you have taken a drug” or “whether or not you have eaten” to relieve anxiety about the health of the care recipient. In addition, a database of helper attendance reports may be created so that the business establishment and the family of the cared person can view it.

  Next, in the utterance process, it is determined in S150 whether or not the utterance number is “5”. If it is “5”, as described above, “I understand. I will report it to the care manager.” , And the control advances to S151 to perform processing for reporting to the care manager. The content of the report is “Lack of medicine remaining” during the conversation. Next, the control advances to S147 to determine whether or not the utterance has been completed, and at the stage of completion, processing for clearing the utterance number is performed in S152.

  Next, the control operation of the subroutine program for the video chat process in S114 will be described with reference to FIG. First, it is determined whether or not the utterance number is “7” in S155, and if it is not “7”, it is determined whether or not the chat flag is turned on in S156. This video chat process ends.

  On the other hand, when the utterance number is “7”, as described above, the voice “I understand. Display the video chat screen.” Is uttered and the control proceeds to S157, and the “7” Is cleared and the chat flag is turned on. This chat flag is ON until it is cleared in S163, which will be described later. During that time, a determination of YES is made in S156, and the control advances to S158.

  In S158, processing for displaying a video chat menu screen is performed. As a result, a video chat menu screen is displayed on the display section 42 of the care recipient 4. In this video chat menu screen, an icon representing a pre-registered partner capable of video chat is displayed, and the icon of the partner to chat with can be selected by clicking the icon. Whether or not the chat partner has been selected is determined in S159, and if not, the menu screen display control in S158 is continued. When the care recipient performs a chat partner selection operation, the control advances to S160, and a process for calling the selected partner is performed. Next, in S161, a process for performing a chat exchange with the selected partner is performed, and it is determined whether or not the chat is terminated in S162, and the chat exchange process in S161 is continuously performed until the chat is terminated. . Then, when the chat is finished, the control proceeds to S163, a process for clearing the chat flag is performed, and this video chat process is finished.

  Next, the control operation of the subroutine program for browsing the website in S115 will be described with reference to FIG. First, in S170, if the utterance number is not “8”, it is determined whether or not the browsing flag is turned on in S175. If not, the website browsing process is ended.

  When the utterance number is “8”, as described above, the utterance “I understand. Display the video chat screen” is uttered, and the control advances to S171 to clear the utterance number. Then, processing for turning on the viewing flag is performed. As a result, a determination of YES is made in S175 until the viewing flag is cleared in S174 described later, and the control proceeds to S172.

  In S172, a process of accessing the photo sharing site and browsing the photo sharing site is performed. Then, in S173, it is determined whether or not browsing is completed, and the browsing process in S172 is continuously performed until the browsing is completed. Then, at the stage of completion, the control proceeds to S174, and the browsing flag is cleared.

  Next, the control operation of the care learning process by the artificial intelligence server 13 will be described with reference to FIG. By S183, a process of periodically receiving a conversation record with a care recipient, a wireless sensor data record, and the like from the central management server 12 is performed. On the other hand, in the PC 16 of a specialist (care provider), a care manager, a helper or a caregiver who visits the care recipient home 5 inputs the condition of the care recipient in the PC 16. In the PC 16 to which the condition is input, the input condition is transmitted to the artificial intelligence server 13 for each care recipient ID. (S181). And after performing other processing by S182, control returns to S180.

  The artificial intelligence server 13 determines whether or not the condition of the care recipient has been transmitted from the care provider's PC 16 in S184, and if it has not been transmitted yet, control proceeds to S191. On the other hand, control proceeds to S185 at the stage of transmission, and the received data is classified for each care recipient ID and additionally stored in the learning DB 14. The learning DB 14 also stores data received in S183. Then, the process of extracting only the data of the care recipient ID of which the condition has deteriorated (NG) is performed in S186, and the process of finding the NG pattern NGPi using data mining is performed in S187. For example, when the data of the condition of the care recipient received in S184 is “dementia is progressing”, the process of S187 is the latest dialog record data corresponding to the care recipient ID. The wireless sensor data recording data is extracted, and data mining is performed to find a characteristic pattern NGPi.

  In this pattern NGPi, the more a lot of care recipient data is, the easier it is to find an accurate pattern. Therefore, the learning DB 14 collects and stores data sent from the terminal 4 for the care recipient nationwide and the PC 16 of the care provider.

  Next, the control advances to S188, where it is determined whether the found NGPi is different from the one found so far. If they are the same, the control proceeds to S191, but if they are newly found this time, the control proceeds to S189, and a process of additionally storing NGPi as an NG pattern in the artificial intelligence DB 15 is performed. Next, the control advances to S190, and the care recipient ID having the newly found NGPi pattern is searched and extracted from the learning DB 14, and NG notification is made. This NG notification is, for example, notified to a family member, a care manager, a helper, or a caregiver of an NG who needs care.

  Next, control proceeds to S191, where it is determined whether or not a regular check time has come. If the regular check time has come, control proceeds to S192. In S192, a care recipient ID having an NG pattern is extracted from the care recipient data added after the previous regular check (data received in S183) and an NG notification process is performed. This NG notification is the same as the notification in S190 described above. The “NG pattern” in S192 is all NGPi (i = 1, 2, 3,...) That have been found so far stored in the artificial intelligence DB 15. Fuzzy inference may be used as means for finding out this abnormal pattern NGPi. Fuzzy reasoning is an inference method that imitates human sensory reasoning, and is an approximate reasoning that can output an inference result inferred from a rule even for an input different from a predetermined rule. By setting and adjusting the inference rules and membership functions, work (tuning work) is carried out so that the inference result is the optimum value. Using this fuzzy inference, for example, predicting the timing of getting sick from time-series information such as meals, water, medicines, and exercises, and performing prior prevention and advice. Next, other processing is performed by S193, and this care learning processing is completed.

  As described above, if a new abnormal pattern NGPi is found by the care learning process by the artificial intelligence server 13, the learning DB 14 is immediately searched using the new abnormal pattern NGPi, and the new abnormal pattern NGPi is searched. An abnormal care recipient that matches the above is determined and an emergency notification is made (S190). Furthermore, all the abnormal patterns NGPi (i = 1, 1) discovered so far stored in the artificial intelligence DB 15 for new data (unchecked data) transmitted from the central management server 12 to the artificial intelligence server 13. 2, 3,...) Is applied to determine whether it is abnormal or not, and if an abnormal care recipient is found, an NG notification is immediately made (S 192). Thereby, there exists an advantage which can discover abnormality of a care recipient early.

  Next, the control operation of the subroutine program of the other process (S193) by the artificial intelligence server 13 and the control operation of the subroutine program of the other process (S116) of the care recipient terminal 4 will be described with reference to FIG.

  The artificial intelligence server 13 performs processing for creating customized data suitable for NGPi in S200. For example, if NGPi discovered as a result of data mining for a care recipient who has suffered from dementia is “I am not interested in what I have been interested in until now” Create utterance data for "Ask a question".

  Next, the process which transmits the created customization data to the terminal 4 for a care recipient is performed by S201. The care recipient terminal 4 receives the transmitted data in S204, and determines whether or not further customization is performed in S205. For example, when the data received in S204 is the utterance data of “Ask the question of interest”, the care recipient's interest items (for example, hobbies) are stored in the care recipient terminal 4. In such a case, the customization process is performed so as to ask questions such as hobbies of the care recipient (S206). When NO is determined in S205 or after the process of S206 is performed, the control proceeds to S207, and the process of storing the customized data in DB (utterance) (see FIG. 9A) is performed. . By performing such a dialogue reflecting the machine learning result, it is possible to find an NGi pattern of a sign that the aforementioned dementia worsens, and to take some measures before the dementia worsens.

  Next, the control operation of the main program of the site agent web server 11 will be described with reference to FIG. The site agent web server 11 executes a control operation while communicating with the company PC 7 and the user terminal 17. First, if the company PC 7 accesses the site agent web server 11 and inputs and transmits various information of the company as customization information (S210), the site agent web server 11 receives the customization information through S212. The information is stored in the database (S212). Next, the site agent web server 11 performs processing for displaying a web page in S213, performs voice recognition processing in S214, performs DB search processing in S215, performs speech processing in S216, and performs other processing in S217. Then, the control returns to S212.

  In the web page display process of S213, if the user terminal 17 accesses the site agent web server 11 and browses the web page (S218), it communicates with the user terminal 17 to communicate with the company web page (homepage). Is transmitted to the user terminal 17 for browsing. On the other hand, the company PC performs other processing in S211 and the user terminal 17 performs other processing in S219.

  Next, the subroutine program control operations of the customization information input process of S210 and the customization information storage process of S212 will be described with reference to FIG. These processes are for causing the site agent web server 11 to store company information for creating a company homepage. First, in S225, it is determined whether or not the company information input operation has been performed by the company PC 7. When an input operation is performed, a process of transmitting the input company information to the site agent web server 11 is performed through S226. The site agent web server 11 receives the transmitted company information in S231, and stores the received company information in the customization DB in S232. As company information stored in the customization DB, for example, as shown in FIG. 14, there are various types of information such as company outline, business content, product description, and the like.

  Next, it is determined whether or not the utterance content is input (or changed) by the company PC 7 (S227). This utterance content is utterance data prepared for a site agent character uttered and interacted with a visitor to be displayed to a user visiting a company homepage. If there is an operation for inputting (or changing) the utterance content, the control proceeds to S228, and processing for transmitting the input utterance content to the site agent web server 11 is performed. The site agent web server 11 receives the transmitted utterance content (S233), and performs a process of storing the received utterance content in the customization DB (or a process of changing to the received utterance content) through S234. The site agent web server 11 employs a TTS (Text To Speech) system and reads out the text data on the customization DB in FIG. 14 in real time, so that the contents of the character can be easily corrected. This TTS (Text To Speech) system may be employed not only in the site agent web server 11 but also in the terminal 4 for care recipient, the PC 8 for employee consultation, the web server 10 for customer reservation, and the like.

  Next, the company PC 7 determines whether or not there is an operation for customizing the character representing the site agent (S229). If the customization operation has been performed, a process of transmitting the customized character to the site agent web server 11 is performed in S230. The site agent web server 11 receives the transmitted customized character (S235), and stores the customized character in the customization DB in S236.

  Referring to FIG. 14, a specific example of the utterance contents stored in the customization DB in S234 is, for example, in the case of utterance number “1”, “I finally feel like spring, I am Shannon.” Utterance number “2 "This is an event at the museum this week!" A specific example of the customized character stored in S236 is data such as rotation and jump designating the movement of the character, such as a character image of a dog mark stored in the customized character column of FIG.

  Next, the control operation of the subroutine program for the web page display process shown in S213 will be described with reference to FIG. In S240, it is determined whether or not there is an access to the home page. If the user operates the user terminal 17 to access the site agent web server 11 and browses the home page, a determination of YES is made in S240 and the control advances to S241, where the character is overlaid on the home page and the user terminal is overlaid. Control for displaying on the screen 17 is performed. As a result, the dog character image stored in the storage area of the customized character in the customization DB in FIG. 14 is overlaid on the home page and displayed on the user terminal 17.

  Next, the control operation of the subroutine program for the DB search process shown in S215 of FIG. 13A will be described with reference to FIG. It is determined whether or not the page has been moved in S245. Usually, a home page is hierarchically prepared with a top page and a plurality of pages of various genres linked to the top page. In S245, the determination is made from one of the plurality of pages to another. It is determined whether or not the user has performed an operation of moving to a page. If there is no page movement operation, the control proceeds to S246, where it is determined whether there is a voice recognition result. If there is no page movement operation, this DB search process ends.

  If it is determined that there is a voice recognition result as a result of the user's voice recognition process in S214 described above, control proceeds to S247, and an utterance number corresponding to the intention analyzed as a result of the voice recognition process is determined. Processing is done. Specific examples of the utterance numbers are shown in the utterance content storage area in the customization DB of FIG. 14, and utterance numbers 1, 2, 3, 4,... Are stored corresponding to various utterance contents. Yes.

  On the other hand, when the user operates the user terminal 17 to perform a page moving operation, the control proceeds to S248, and a process for determining an utterance number corresponding to the web page of the moving destination is performed. For example, when a movement operation to a web page on which event information is posted is performed, an utterance number “2” is determined, and an utterance such as “An event will be held at the museum this week!” Is uttered by the character.

  After the control of S248 or S247, the process proceeds to S249, and processing for searching the customized DB for the utterance content and the character movement corresponding to the determined utterance number is performed. For example, if the utterance number “1” is selected, the rotation of the character is also selected as “1”, and while the dog mark character rotates, “I finally feel like spring, I am Shannon.” Is uttered. Also, when utterance number “2” is selected, the character movement “2” jump is selected, and the dog mark character jumps while the utterance “There will be an event at the museum this week!” Be emitted.

  Next, the control operation of the subroutine program for the speech process shown in S216 of FIG. 13A will be described with reference to FIG. In S255, it is determined whether or not there is a searched utterance content and a character movement. If there is no utterance, the utterance processing is terminated. If the utterance content and the character movement have already been searched for in S249 described above, a determination of YES is made in S255, and the control advances to S256, where control is performed to move the character while speaking according to the utterance content. As a result, as described above, the character of the dog mark overlaid on the home page rotates on the user terminal 17 so that the utterance “I finally feel like spring, I am Shannon” is uttered. Is done. The controlled utterance contents and character movements in S256 are cleared in S257, and it is determined in the next S258 whether or not the dialogue has ended. If the dialog has not ended, the utterance process ends. However, when the dialog ends, control proceeds to S259, and after the utterance number determined in S248 or S247 is cleared, the utterance process ends.

  Next, FIG. 16 (a) shows the operation control of the subroutine program for other processing of the company PC shown in S211 of FIG. 13 and the control operation of the subroutine program of other processing shown in S217 of the site agent web server 11. This will be explained based on. This other process is to improve the content of the utterance content stored in the utterance content storage area in the customization DB of FIG. 14 so that a satisfactory dialogue can be made for the user who visited the homepage. It is.

  First, in S269, it is determined whether or not there is an utterance number corresponding to the intention analyzed as a result of the speech recognition by the speech recognition process (S214). For example, when a user asks a question that was not expected at all, the utterance number and utterance content for the question are not registered in the customization DB in advance, and there is no utterance number corresponding to the user's question intention It becomes. In such a case, the site agent character (the dog's mark) will respond with a voice such as “I don't know” for the time being. In such a case, a determination of YES is made in S269, the process proceeds to control S270, and processing for registering the intention and voice in the storage device 24 of the site agent web server 11 is performed.

  On the other hand, in the other processing of the company PC 7, first, in S265, it is determined whether or not there has been an operation for browsing the registration information of the site agent web server. If the company PC 7 is operated to access the site agent web server 11 and browse the registered information, the control advances to S266, and a process of transmitting a browse request to the site agent web server 11 is performed. In the site agent web server 11, if the browsing request is transmitted, a determination of YES is made in S 271 and the control advances to S 272, and the registration information of the unanalyzable intention and voice registered in the storage device 24 in S 270 is stored in the company. Processing to transmit to the PC 7 is performed.

  The company PC 7 receives the transmission information in S267, and controls to display the received registration information on the company PC 7 in S268. The person in charge of the company's homepage who sees the registration information, that is, the question of the user that the site agent could not handle, can customize it to supplement the utterance content and utterance number so that it can also handle the conversation content of the user who cannot respond An information input process (S210) is performed. By repeatedly performing such customization information replenishment input processing, the content of utterances in the customization DB is gradually enhanced, and a dialog that can give satisfaction to the visiting user becomes possible.

  Next, an advanced control operation of the DB search process will be described with reference to FIG. This advanced type of DB search process is an extension of the DB search process shown in FIG.

  First, in S280, it is determined whether or not a user agent has visited. The user agent is a software agent that operates for a specific user. Various detection signals from a wearable computer worn by the user or interacted with the user for a certain period of time. Is a software agent that stores information related to the user (such as a user's profile) as knowledge data. In other words, it is the user's buddy artificial intelligence. For example, as shown in FIG. 16 (c), the user agent 48 uses information related to the user's profile as knowledge data, such as date of birth, gender, occupation, hobby, favorite music, favorite movie, etc. Etc. are remembered. The user agent 48 is composed of a mobile agent, moves to the place 46 of the site agent web server 11, communicates (meetings) with the site agent 47 on the place 46, and works in cooperation with each other. . That is, the site agent 47 and the user agent 48 constitute a multi-agent system.

  When the user agent 48, which is a mobile agent, moves on the place 46 of the site agent web server 11, a determination of YES is made in S280, and the control advances to S281. In S281, the site agent 47 meets with the user agent 48 to obtain a user profile. Next, in S282, a process for searching for information matching the user based on the obtained profile and determining an utterance number is performed. On the other hand, when there is no visit by the user agent, NO is determined in S280 and the control shifts to S245 (see FIG. 15B). By performing such control, when there is a visit by a user agent, there is an advantage that it is easy to provide information matching the homepage visitor (user) by effectively using the knowledge data of the user agent. is there. For example, in the case of a restaurant homepage, it is possible to recommend dishes that reflect the user's preferences.

  Next, the control operation of the main program of the employee consultation PC 8 equipped with artificial intelligence will be described with reference to FIG. A speech recognition process is performed in S290, an utterance process is performed in S291, a customization information input process is performed in S292, a DB search process is performed in S293, and a feedback learning process is performed in S294.

Since the speech recognition process in S290 is substantially the same as the speech recognition process in FIG. 4B, detailed description is not repeated here. As shown in the consultation DB in FIG. 17B, the employee consultation PC 7 stores the response contents in association with the response content storage areas A, B, C.
For example, in association with A, “I can get a 3-day off for remorse. Paid leave starts from the second year of joining the company.”
Corresponding to B, “The technology currently under development is deep learning.”
Corresponding to C, "Personnel evaluation criteria is ..."
Etc.

Weights m, n, o, p, q,... Are assigned to the contents of the answers A, B, C,. . These weights include:
m: Contents that are difficult to hear from supervisors n: Work rules o: Special departments p: Research and development q: Personnel relations, etc.

  These weights are assigned as weighting coefficients to the contents of each answer in five stages of 1 to 5. For example, in the case of answer A, “You can get 3 days off for abduction. Paid leave starts from the second year of joining the company.” M: The highest weighting “5” for content that is difficult to ask the boss, n: Heaviest weight “5” for work rules, a: lightest weight “1” for specific departments, p: lightest weight “1” for research and development, q : For personnel relations, the lightest weight is “1”. Each stored data of these consultation DBs is classified and stored by a customization information input process described later.

  The control operation of the subroutine program for the speech processing in S291 in FIG. 17A will be described with reference to FIG. In S298, it is determined whether or not it is a narrowing-down stage. Judgment is made based on this narrowing-down stage and whether the artificial intelligence installed on the PC 8 for employee consultation is in the middle of asking a question about weighting in order to narrow down the answer contents. In this case, YES is determined in S298, and the control advances to S299.

  In S299, a process of sequentially speaking according to the weighting content is performed. Specifically, referring to the consultation DB shown in FIG. 17B, first, the m question “whether it is difficult for the boss to ask” is given to the employee, and then “n” Talk to employees about whether or not it is, then talk to employees about whether or not it is related to a specific department, and then check whether it is related to research and development. The question “whether” or not is spoken to the employee, and then the question “whether or not it is a personnel relation” q is spoken to the employee. Employees answer these questions in five levels, 1-5. For example, for an utterance of “whether or not it is difficult for the boss to hear”, “5” if it is exactly the same, “4” if it is somewhat difficult to hear, “3” if it is neither, “2” if the content is easy to hear and “1” if the content is easy to hear. In this way, the weighted questions are sequentially spoken, and the answers from the employees are analyzed by the voice recognition process (S290).

  Here, for convenience of explanation, the DB search processing of FIG. This DB search process is a subroutine program shown in the DB search process of S293 in FIG. 17A described above. First, in S305, it is determined whether or not the acquisition of the speech recognition result for weighting has been completed. The The answer to the question about all the weights m, n, o, p, q... In the consultation DB is made by the employee, and when the acquisition of the voice recognition results is completed, YES is determined in S305. Proceeding to S306, processing is performed for searching and obtaining the answer contents close to the obtained weighting from the consultation DB. For example, if the answer contents from the employee are weighted answer contents with weights m and n, and o, p, and q are light weighted answer contents, the answer contents of “A” rank higher Will be attached.

  Next, the DB search end flag is turned on in S308. As a result, in the utterance process of FIG. 18A, as a result of determining whether or not the DB search end flag is ON in S300, the answer is YES and the control proceeds to S301. In S301, control is performed to sequentially speak and display the response contents ranked in S307 described above. This makes it possible to respond to employee consultations. Next, in S302, processing for clearing the DB search end flag is performed.

  Next, the control operation of the subroutine program for the DB search process of S293 in FIG. 17A will be described with reference to FIG. This customization information input process is a process for storing data in the consultation DB in FIG. First, in S310, it is determined whether or not there is an instruction to add answer contents. If there is not, the process ends. If there is, the process of storing the added answer contents in the DB together with the weighting in S311. Is made. As a specific method of this additional input, an operator who wants to perform customization information input processing consults with the employee consultation PC 8 in the same manner as a normal employee, and is installed on the employee consultation PC 8 in response to the consultation. Artificial intelligence utters questions about weighting sequentially as described above. The operator answers the answer sequentially, and the answer contents are finally ranked and presented to the operator. The operator sees it, and if there is an answer content to be added, issues an addition instruction to the artificial intelligence, and additionally records the newly added answer content in association with the weighted answer by the operator in the consultation DB.

  As a result, a consultation DB is automatically constructed through a dialogue with artificial intelligence, and there is an advantage in that it saves the operator from having to construct a complicated hierarchical DB.

  Next, the control operation of the subroutine program for the feedback learning process indicated by S294 in FIG. 17A will be described with reference to FIG. This feedback learning process is a control for correcting the weighting of the answer contents according to the responses from the employees who have seen the answer contents ranked and presented according to the consultations from a plurality of employees. For example, when an employee is satisfied with the response content in response to a consultation with multiple (for example, 10 or more) employees, the response content that the multiple (10 or more) employees answered about weighting and the operator first Is a process of correcting the weighting according to the average of the weighted answer contents of a plurality of employees when the weighting that is answered at the stage of the customization information input process is deviated.

  First, in S315, it is determined whether or not the consultant has obtained the desired answer. If the desired answer has not been obtained, this process ends. On the other hand, if the counselor obtains the desired answer, the control advances to S316 to determine whether the weighting of the desired answer needs to be corrected. This is determined based on whether or not the answer contents for the weights of a plurality of employees and the weights for the answer contents stored in the consultation DB at the present time are more than the allowable range. If the deviation is more than the allowable range, the control advances to S317 and a process of correcting the weight is performed.

  With this feedback learning process, the more weighted consultations can be made, the more accurate the weights can be corrected, and answers suitable for the employees of the company can be made. Note that the above-mentioned method of using a DB with weighting coefficients attached to each answer content to sequentially issue questions about weighting to the consultant and narrow down the appropriate answer content from the answers is limited to the use for employee consultation. For example, various uses such as legal consultation, diagnosis of a disease name from a medical condition, and consultation of job hunting can be considered. In addition, a function of performing reinforcement learning may be provided so that the minimum number of questions can be narrowed down to appropriate answer contents. Reinforcement learning is performed by giving a higher reward as the number of questions is reduced to narrow down the appropriate answer contents.

  Next, a control operation in which the user browses and purchases the machine learning result information transmitted to the upload server 19 in S203 of FIG. 12 will be described with reference to FIG. If the machine learning result information is transmitted to the up-store server 19 in S203 of FIG. 12, it is determined that the learning result is received in S320, and the control advances to S321. In S321, the received learning result is classified by genre and registered in the DB.

  FIG. 1 shows a state where the learning results are classified into genres and registered in the DB. The learning results are stored in the DB 20 of the learning result upstore 18 in correspondence with various genres (classifications) such as nursing care, infant education, and so on. The learning result includes, for example, customized data such as an NG pattern found by data mining and utterance content corresponding to the NG pattern. The genre of learning results uploaded to this learning result up-store is not limited to nursing care and early childhood education. For example, learning results for patients with depression, manic depression or psychosis, or the weighting described above. The result of learning (for example, legal consultation or illness consultation) learned by a method of narrowing down answers and answering consultations may be uploaded.

  On the other hand, in step S330, the user terminal 17 determines whether or not to access the genre-specific learning result display site, and the user operates the user terminal 17 to access the up-store server 19 to display the genre-specific learning result display site. If a browsing operation is performed, the control advances to S331, and processing for transmitting a browsing request to the upload server 19 is performed.

  In the upstore server 19, it is determined whether or not there is a request for browsing the learning result by genre in S <b> 322, and if a browsing request is transmitted from the user terminal 17, the control proceeds to S <b> 323 and Processing to transmit to the terminal 17 is performed.

  In the user terminal 17 that has received it in S332, the control proceeds to 333, and the received genre-specific learning result is displayed on the user terminal 17. Next, in S334, it is determined whether or not the user has performed a purchase operation. If there is something that the user wants to purchase after seeing the display of the machine learning result by the artificial intelligence, the purchase operation is performed by specifying the machine learning result desired to be purchased. Then, a determination of YES is made in S334, the control proceeds to S335, and the purchase request is transmitted to the upstore server 19. In the upstore server 19, it is determined whether or not a purchase request is received in S324. If a purchase request is transmitted from the user terminal 17, the control proceeds to S325, and the approval process and upstore in S336 on the user terminal side are performed. In the approval process in S325 on the server 19 side, the approval process is performed between the two.

  Next, a process of transmitting the learning result requested in S326 to the user terminal 17 is performed. In the user terminal 17, the learning result is downloaded in S337.

  A company operating the learning result up-store 18 in which the up-store server 19 is installed accepts machine learning result data from machine learning companies in various fields that generate machine learning result data by artificial intelligence and sells it to the user. It is a person who develops a business, in other words, a contractor who conducts a machine learning platform business. Many machine learning companies play on this platform to generate and upload various machine learning result data, and only the machine learning result data that the user likes is downloaded and purchased by the user. In other words, on this platform, an ecosystem (ecosystem) that is governed by the natural principle of natural selection by free competition is nurtured. As a result, a new distribution market of machine learning result data is created, and excellent machine learning result data desired by the market is uploaded and distributed, which can promote dramatic development of the machine learning field of artificial intelligence.

The genre of machine learning result data targeted by this machine learning platform is, for example, for legal consultation, to narrow down disease names from medical conditions, for meal menus, for sightseeing spot selection, for date course selection, for homes using artificial intelligence Anything may be used, such as for teachers, go, shogi, chess, etc.
The machine learning result data is classified into an OS part (basic data part) and an application data part, and the OS part (basic data part) is incorporated into a user terminal or server, or the OS part (basic data part) from the cloud. These functions and services may be provided, and only the application data portion may be uploaded to the learning result upstore 18 (upstore server 19) for distribution. Further, instead of classifying into an OS part (basic data part) and an application data part, it may be classified into a part that requires customization (personalization) and a part that does not require customization (personalization) for each user.

The invention of the machine learning platform has the following configuration.
Upload accepting means (eg, S320) for accepting upload of result data of machine learning by artificial intelligence;
Storage means (for example, S321) for storing machine learning result data received by the upload reception means;
Display means (for example, S323) for displaying the machine learning result data stored in the storage means on a web page;
Request accepting means (for example, S324) for accepting the request (for example, purchase request) when the user who browsed the web page wants to use the result data of the machine learning;
A payment means (for example, S325) for making a payment to the user who issued the request when the request is accepted by the request acceptance means;
A machine learning marketplace system comprising: transfer means (for example, S326) for transferring ownership of the machine learning result data, which is a target of the request received by the request receiving means, to a purchase user.

  In addition, the machine learning marketplace system provides the user with the result data of the machine learning that is the target of the request received by the request receiving unit instead of or in addition to the transfer unit. The lending means may be provided. In addition, the machine learning marketplace system permits the user to use the machine learning result data, which is the target of the request received by the request receiving means, instead of or in addition to the transferring means. May be provided. In such a case, the result data of the machine learning becomes more personalized by matching the user by utilizing the user for artificial intelligence. Therefore, when lending (or licensing) to a user, after storing the original machine learning result data before lending (or licensing), separately storing the machine learning result data dedicated to the user, Control is performed so that machine learning result data dedicated to the user is used only by the user. As a result, the more it is used, the more it becomes machine learning result data personalized to the user, and it can be nurtured into artificial intelligence that performs work that matches the user.

  Moreover, you may further provide the artificial intelligence engine transfer means for transferring the ownership of the artificial intelligence engine for utilizing the machine learning result data to which the ownership is transferred to the purchase user to the purchase user.

  Further, instead of the artificial intelligence engine transfer means or in addition to the artificial intelligence engine transfer means, an artificial intelligence engine lending means for lending the artificial intelligence engine to the purchase user may be provided. Further, instead of the artificial intelligence engine transfer means or in addition to the artificial intelligence engine transfer means, a use permission means for permitting the purchase user to use the artificial intelligence engine may be provided.

  Moreover, you may further provide the use means on the cloud for using at least one of the result data of the machine learning and the artificial intelligence engine for the user on the cloud.

Next, a modification of the embodiment described above will be described.
(1) A microphone array is installed on the wall or ceiling of the home 5 requiring care, voices of care recipients (users) are collected by the microphone array, and voice recognition processing by the care recipient terminal 4 (S110 in FIG. 8). Etc.) may be performed. The microphone array has a function of performing sound source separation based on audio signals collected by a plurality of microphones and removing audio signals from a predetermined distance. With this function, the voice of the care recipient is discriminated from other voices (noise) in the voice recognition process (S110 in FIG. 8, etc.), and voice recognition is performed based on the classified voice. Thereby, there exists an advantage which can improve the precision of voice recognition of a care recipient (user), without being confused by noise. The noise suppression function by the microphone array is to remove noise existing on a straight line from two microphones by applying a two-point received sound source separation method based on a simultaneous equation solution between frames. The two-point sound source separation method based on the simultaneous inter-frame equation solution corrects the delay between the microphones, and the signals recorded by the two microphones become the sum of the distance attenuations of the sound sources. And subtracting the signal from a certain distance. Since processing in the time domain is possible and signal can be removed by simple subtraction, there is an advantage that it can be expected to be realized with a small device with reduced calculation cost.

  The general-purpose microphone environment has a large performance difference including surrounding systems, and it takes time and effort to individually construct device drivers depending on the type of computer. In order to eliminate such inconvenience, the microphone array system incorporates a real-time signal processing device in the microphone body. As a result, a general-purpose driver using USB audio can be used on any Mac machine, Windows machine, or Linux machine.

(2) Further, the sound source position in the room is estimated using microphones installed at a plurality of locations such as the walls and ceiling of the home requiring care 5 and the sound source position information is read in advance by the voice recognition system so that the same sound source can be used. May be detected, and an improvement in the speech recognition rate and a highly accurate recognition of the conversation content may be realized by a speech recognition process (S110 in FIG. 8, etc.). For example, the sound source position information of the television emitted from a certain position of the care recipient's home 5 is read in advance by the voice recognition system, and control is performed so as to remove the sound (noise) from the sound source position of the television.

(3) Instead of or in addition to the two-point received sound source separation method based on the inter-frame simultaneous equation solution, a general sound source separation method using independent component analysis may be introduced by increasing the number of microphones. Independent component analysis is a calculation method for separating a multivariate signal into a plurality of additive components, and each component is assumed to be a non-Gaussian signal and statistically independent from each other. This is a special case of blind signal separation.

(4) Further, general recognition criteria including vowels and consonant acoustic patterns learned using the data generated by a large number of speakers by an unspecified speaker speech recognition system are stored in a database, and care recipients (users) The general recognition criteria may be modified based on the acoustic properties included in the speech collected as a sample from (1) to perform speaker adaptation. The recognition criterion modified by the speaker adaptation is stored in the database, and speech recognition is performed by the speech recognition process (S110 in FIG. 8) using the modified recognition criterion, thereby improving the speech recognition rate. And highly accurate recognition of conversation contents may be realized. This speaker adaptation may be speaker adaptation in speech recognition using a Gaussian mixture distribution model, or may be performed using a deep neural network (deep learning).

(5) In the speech recognition processing shown in this embodiment (for example, S51 in FIG. 4B, S51 in FIG. 5A, S111 in FIG. 8, S290 in FIG. 17, etc.), synonyms are made equivalent and recognized. A synonym recognition function may be provided. A buffer corresponding to the deformation of the verb is implemented, and for example, even if the user says "I don't drink", it is controlled so that it is synonymous with "I don't drink". As a buffer corresponding to the deformation of the verb, for example, the following synonym list data is stored, and “do not drink” and “do not drink” are both recognized as negative words for “drink”.
[Verb basic form] [Verb variant] [Affirmation (1) / Negation (2)]
Drinking drinking 1
Drinking drinking 1
Drinking drinking 1
Drunk drunk 1
I want to drink 1
Do not drink 2
Drink not drink 2
Drink not drink 2
:::

(6) As a modification of the customization information input process (FIG. 3C, S110 in FIG. 8, S210 in FIG. 13, etc.) shown in the present embodiment, for example, the input is made on the intuitive management screen shown in FIG. May be. FIG. 20 shows a modification of the customization information input process in the case of S110 in FIG. 8, and is a diagram of an input management screen displayed on the display of the terminal 4 for the care recipient or the PC 16 of the care provider. . As shown in FIG. 20, a speech balloon is displayed on the management screen, and the utterance content to the user is input into the speech balloon, and the branch of the speech balloon of the question to the user is branched according to the content of the user's reply. First, a further utterance content area is displayed. Specifically, the utterance content of “How are you?” Is input in the speech balloon, the branch is divided into a YES arrow and a NO arrow, and the utterance content of “Good” is in the speech balloon ahead of the YES arrow. And the content of the utterance "Do you want to contact your family?" In this way, on the management screen, the utterance content area such as a speech balloon and the utterance content area of the question to the user (speech balloon) are branched by arrows according to the response contents of the user's YES, NO, and each branch destination. In order to display and input a further utterance content area (speech balloon), it is possible to input by an intuitive management screen, and there is an advantage that the conversation content can be freely input and edited. This intuitive management screen input is not limited to the conversation content input for the care agent shown in S110 of FIG. 8, but is applied to the conversation content input for various purposes such as reception agent, visitor reservation, and site agent. it can.

  The conversation content (speech content) input as described above is stored in a database (for example, the DB (speech) of the EEPROM 36 of the terminal 4 for care recipient) and is real-time by the TTS (Text To Speech) system described above. Read aloud. Specifically, for example, the utterance content (text data) of “How are you?” Is first searched by the DB search process of S112 of FIG. 8, and the text data of “How are you?” Is acquired by the utterance process of S113. Read aloud. The response of the care recipient (user) to that is recognized by the voice recognition process of S111, and it is determined whether the response is a positive response (YES) or a negative response (NO). In the case of an affirmative response (YES), the utterance content (text data) of “It was good” ahead of the YES arrow is searched and read out, while in the case of a negative response (NO), NO Search for the utterance content (text data) of "Do you want to contact your family" at the end of the arrow and read it out.

(7) Artificial intelligence such as various agents shown in this embodiment is made into an API (Application Programming Interface), which is versatile and can be used on robots or tablets of other companies other than artificial intelligence production companies. Also good. Further, artificial intelligence may be converted into a cloud and provided to the user as a cloud computing service. At that time, the artificial intelligence on the cloud side should be generalized that has not been customized, and the user himself or a specialist will input the information for customization according to the various applications described above, and the database on the cloud side or the user terminal side Store it in the database. When a user uses artificial intelligence on the cloud side via a user terminal or a robot, the user's customization information is read from the database and the general-purpose artificial intelligence is operated using the customization information. As a result, although it is general-purpose artificial intelligence on the cloud side, there is an advantage that a customized artificial intelligence service can be provided to the user who uses it.

  As mentioned above, although embodiment of this invention was explained in full detail, this invention is not limited to these embodiment, A various change is possible in the range which does not deviate from the meaning of this invention.

1 Internet 2 Company reception PC
3 PC for reservation setting frame
4 Caregiver terminal 5 Caregiver home 6 Wireless sensor network 7 Company PC
8 PC for employee consultation
9 Artificial Intelligence Service Company 10 Customer Reservation Web Server 11 Site Agent Web Server 12 Central Management Server 13 Artificial Intelligence Server 14 Learning DB
15 Artificial Intelligence DB
16 PCs of various specialists
17 User terminal 18 Learning result upstore 19 Upstore server 20 Database

Claims (6)

A computer system that uses a software agent to respond to a person by voice,
Speech recognition means for recognizing human speech;
Utterance means for performing utterance corresponding to the recognition result by the voice recognition means;
Storage means for storing information for generating utterance contents as a database;
Utterance content generation means for generating utterance content using information stored in the storage means,
The utterance means performs an utterance according to the utterance content generated by the utterance content generation means,
The storage means is a computer system in which a user can input and store information customized for a specific application using the software agent. The voice recognition means includes a discrimination means for discriminating the user's voice from other voices, performs voice recognition based on the voice distinguished by the discrimination means,
The computer system according to claim 1, wherein the distinction unit includes a microphone array for performing sound source separation based on audio signals collected by a plurality of microphones and removing an audio signal from a predetermined distance. The storage means includes an input operation means for a user to input information customized for a specific application using the software agent on a management screen,
The input operation means displays an utterance content area on the management screen and inputs the utterance content to the user in the utterance content area, and responds to the utterance content area of the question to the user according to the response content of the user Branching and displaying a further utterance content area at each branch destination, and having a flowchart type input means for inputting the utterance content according to the response content in the utterance content area,
The computer system according to claim 1 or 2, wherein the utterance content generation means includes means for generating utterance contents in an utterance content area sequentially along the flowchart input by the flowchart type input means. The storage means stores the name of an employee and a telephone number for calling the employee in association with each other in order to use the software agent for accepting a visitor,
When the name of the interviewee spoken by the visitor is recognized by the voice recognition means, an automatic for indexing the telephone number corresponding to the name from the storage means and calling the employee of the telephone number by telephone The computer system according to claim 1, further comprising a calling unit. Data accumulation means for collecting and accumulating dialogue history data consisting of recognition results by the voice recognition means and utterance contents by the utterance means from a plurality of software agents via a network;
Machine learning means for machine learning using the accumulated data by the data accumulation means as machine learning data;
Providing means for providing a machine learning result by the machine learning means to the software agent;
The computer system according to claim 1, wherein the software agent performs control for reflecting the provided machine learning result in generation of utterance content by the utterance content generation unit. A program that uses software agents to respond to people by voice,
On the computer,
Speech recognition means for recognizing human speech;
Utterance means for speaking to people,
Storage means for storing, as a database, information for generating utterance content corresponding to the recognition result by the voice recognition means;
Using the information stored in the storage means to function as utterance content generation means for generating utterance contents corresponding to the recognition result by the voice recognition means,
The utterance means performs an utterance according to the utterance content generated by the utterance content generation means,
The storage means is a program in which a user can input and store information customized for a specific application using the software agent.

Images