JP2019211909A - Information presentation system, information presentation method and program - Google Patents

Abstract

To provide an information presentation system that allows each of users to accurately hear voice information that a system provides, in which an alteration model changing speech production data for accurate hearing is set for each user, and which can conduct maintenance of the alteration model easily due to a small amount of data compared with a rule base.SOLUTION: An information presentation system according to the present invention comprises a hearing orientation estimation part that estimates a hearing orientation that indicates easiness of hearing of a voice of each user in a conversation with a user and creates and updates an alteration model changing speech production data provided by voice to the user according to a hearing orientation of the user and a presentation control part that corresponds to an alteration model set for each user and changes speech production data that is a voice answer from a dialog system so that it becomes the hearing orientation of each user.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information presentation system, an information presentation method, and a program.

  In recent years, due to the enhancement of the Internet environment, social network services (hereinafter referred to as SNS) have become widespread, and it has become possible to easily communicate between a plurality of users using text and images. . For example, LINE (registered trademark), Facebook (registered trademark) messenger, Slack (registered trademark), and the like are representative examples of SNS applications. These SNSs share not only the exchange of information between one-to-one users but also the information (dialogs among a plurality of users) transmitted and received among a large number of users in a predetermined group among all users in the group. It also has a function.

Examples of man-machine interactive SNS include Google Assistant (registered trademark), Amazon Alexa (registered trademark), and Line Cova (registered trademark).
In addition, each of the above-described applications is mounted on a smart speaker such as a personal computer and a smart device, Google Home (registered trademark), Amazon Echo (registered trademark), or Clova Wave (registered trademark), and each of them is synthesized by speech. In addition, information mainly using information presentation using voice is also widely used.

In the information presentation method in SNS, various ideas are made for each application.
For example, there is a presentation method for rewriting and presenting the language expression of an information sentence for the purpose of presenting an information sentence together with characters defined on the system for the presentation contents in information presentation (see, for example, Patent Document 1). .

  In addition, when information is presented to the user by voice, the information presentation voice synthesized by the system is uttered to the user via a speaker or the like for voice synthesis of voice used for presentation. . At this time, when the user listens to the utterance to be presented, there is an information presentation method for performing processing that does not feel mechanical unnaturalness to the utterance (for example, Patent Document 2 and Patent Document 3). reference).

Japanese Patent No. 6161656 Japanese Patent No. 595348 Japanese Patent No. 6232922

However, when a dialogue only by voice is performed between the user and the system, the user's hearing ability and comprehension ability for words differ because the user's attributes have diversity such as age or gender.
For this reason, in the dialogue with the system, there is a case where the information provided by the system cannot be heard or the contents of the information cannot be accurately transmitted in the dialogue because the words in the voice cannot be understood.

In addition, in order to perform a dialogue between the user and the system, a rule-based method can be used in which a dialogue is executed in advance assuming a dialogue scenario as to what kind of answer is to be given to the question from the user.
However, there is a need for manual maintenance for each of the rules set in the rule base because of the necessity to correspond to each of many users. In this maintenance, it is enormous to set up the above rules that contain similar words and expressions for words that are difficult to understand or difficult to understand, corresponding to each of the diverse users described above, and constructing it as a dialogue scenario Work is required.

  The present invention has been made in view of such circumstances, and a change model that changes utterance data so that each of the users can accurately hear the information provided by the system and can be heard accurately is provided by the user. Provided are an information presentation system, an information presentation method, and a program that are provided for each, and can easily maintain a changed model (a correction process that sequentially corresponds to a user) because the amount of data is smaller than that of a rule base.

  In order to solve the above-described problems, the information presentation system of the present invention estimates the listening orientation indicating the ease of hearing of each voice of the user in the dialog with the user, and is supplied to the user by voice. A listening orientation estimation unit that generates, predicts, and updates a change model (including each of the listening orientation estimation model and the listening orientation template model in the embodiment) that changes the speech data corresponding to the listening orientation of the user; A presentation control unit that corresponds to the change model set for each user, and that changes the utterance data, which is an answer by voice from a dialogue system, to be the listening preference of each of the users. Features.

  The information presentation system of the present invention writes and stores, for each user, a dialog history, which is a history of the dialog with each of the users, used for estimating the listening orientation. It further includes a dialog processing unit that determines a response to an utterance from the user based on a rule.

  In the information presentation system of the present invention, the listening orientation estimation unit extracts the listening orientation of the user from the evaluation of the utterance data in the dialogue of the user, and the attribute information of the user and the listening of the user Each of the orientation information indicating the orientation is written and stored for each user in the user attribute storage unit.

  The information presentation system of the present invention performs grouping for classifying each of the users corresponding to the attribute information of each of the users, and each of the classifications according to the listening orientation common to the users included for each of the classifications. The method further includes a grouping estimation unit that generates a template change model that is a change model.

  In the information presentation system of the present invention, the listening orientation estimation unit extracts the template change model of the classification corresponding to the user for the user for which the change model is not prepared, and is extracted in the dialogue. The change model corresponding to the user is generated corresponding to the listening orientation.

  The information presentation system of the present invention is characterized in that the attribute information is set as a combination of at least demographic data including age, sex, and residence of the user.

  In the information presentation system of the present invention, the change model includes at least replacement of words in the utterance data of the system response determined by the dialog processing unit, frequency and speed of speech when the utterance data is read out, and paragraph breaks. It is characterized by showing processing to be changed.

  In the information presentation system of the present invention, the presentation control unit writes and stores the change content, which is the content of the change of the utterance data by the change model, as a change history in the dialogue action storage unit, and the listening-oriented The estimation unit extracts the listening orientation of the user from the conversation history and the change history.

  In the information presentation method of the present invention, the listening orientation estimation unit estimates the listening orientation indicating the ease of hearing of each voice of the user in the dialog with the user, and the speech data supplied to the user by voice A listening orientation estimation process for generating and updating a change model to be changed corresponding to the listening orientation of the user, and a presentation control unit corresponding to the changing model set for each user, A presentation control process for changing the utterance data, which is an answer by voice from a dialogue system, to the listening orientation of each of the users, via a dialogue processing unit that determines a response to the utterance based on a rule. It is characterized by including.

  The program of the present invention estimates the listening orientation indicating the ease of hearing of each voice of the user in the dialogue with the user, and the speech data supplied to the user by the voice Listening orientation estimation means for generating and updating a change model that changes according to the listening orientation, the utterance data corresponding to the change model set for each user, and a speech response from a dialogue system, It is a program for making it function as a presentation control means to change so that it may become the said listening preference of each said user.

As described above, according to the present invention, each user can accurately hear the information provided by the system, and a change model is provided for each user to change the utterance data so that it can be heard accurately. Information presentation that facilitates maintenance of the change model (correction process that sequentially corresponds to the user) because the amount of data including the dialogue scenario that should be constructed assuming the content of the dialogue in advance compared to the rule base is small A system, an information presentation method, and a program can be provided.
In addition, because there is a grouping estimation unit, a template change model, which is a change model generalized within the group, is used for each user who does not have a change model or who has insufficient accumulation of speech data, etc. Thus, the display information can be optimized.

It is a block diagram which shows the structural example of the information presentation system with which a user and a system interact by one Embodiment of this invention. It is a block diagram which shows the structural example of the dialogue process part 102 in this embodiment. It is a figure which shows the structural example of the dialogue history table memorize | stored in the dialogue history memory | storage part. It is a block diagram which shows the structural example of the listening orientation estimation part 103 in this embodiment. It is a figure which shows the structural example of the user attribute table memorize | stored in the user attribute memory | storage part. It is a figure which shows the structural example of the dialogue action table memorize | stored in the dialogue action memory | storage part 109. FIG. FIG. 7 is a conceptual diagram illustrating word replacement processing in the action illustrated in FIG. 6. It is a figure which shows the structural example of the grouping table memorize | stored in the grouping memory | storage part. It is a flowchart which shows the operation example of the dialogue system using the information presentation system of this embodiment. It is a conceptual diagram which shows the other structural example of the information presentation system with which a user and a system interact by one Embodiment of this invention.

  The present invention relates to an interactive system in which, for example, when a user asks a question, the system side notifies an answer corresponding to the question by voice obtained by synthesizing speech data, or a dialogue between users. Further, the present invention relates to a configuration in which the utterance data notified by the system side is changed corresponding to each of the user's hearing ability and word comprehension ability so that the user can easily hear the answer from the dialogue system and understand the contents.

In response to the user's hearing ability, for example, when speech data is synthesized and output as speech, the frequency of the speech, the playback speed, the segmentation of the phrase to be reproduced, the duration of the segmentation, and the like are changed. Here, the delimiter is silence that is inserted between phrases or between words when the speech data is reproduced as speech. The delimiter time width indicates the length of silent time inserted between clauses or between words.
Further, in response to the user's ability to understand words, for example, the technical term is changed to be replaced with a word that is another synonym (or synonym or synonym) generally used.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the following description corresponding to FIG. 1, the dialogue between the user and the system will be described as an example.
FIG. 1 is a block diagram illustrating a configuration example of an information presentation system in which a user and a system interact with each other according to an embodiment of the present invention.
In FIG. 1, the information presentation system 1 includes an information presentation server 10 and a user terminal 11.
Each of the information presentation server 10 and the user terminal 11 transmits and receives data via a network 500 that is an information communication network including the Internet.

  The information presentation server 10 outputs an answer corresponding to the user's question and the like supplied via the user terminal 11 to the user terminal 11 by voice data. Here, the audio data indicates data obtained by digitizing audio in a predetermined audio file format (such as an uncompressed audio format, an irreversible compressed audio format, and a lossless compressed audio format). The information presentation server 10 is, for example, a general purpose computer or a personal computer.

  The user terminal 11 serves as a platform for using a virtual personal assistant (VPA) such as Amazon Alexa (registered trademark), Apple Siri (registered trademark), Google Assistant (registered trademark), etc. A smart speaker or the like that performs a dialogue with the information presentation system 1. Further, the user terminal 11 may be a mobile terminal such as a smartphone or a tablet terminal, or a personal computer, and is an interactive application in which a request and a response are performed only by voice without displaying an image (including text characters) on the display unit. It may be applied to.

  The information presentation server 10 includes a data input / output unit 101, a dialogue processing unit 102, a listening orientation estimation unit 103, a presentation control unit 104, a speech synthesis unit 105, a grouping estimation unit 106, a dialogue history storage unit 107, a user attribute storage unit 108, Each of the interactive action storage unit 109, the grouping storage unit 110, and the language knowledge storage unit 111 is provided.

The data input / output unit 101 is an external input interface that inputs data including voice data of voice input by the user to the user terminal 11 via the network 500.
The data input / output unit 101 is also an external output interface that outputs data including voice data such as an answer to the contents of voice data from the user to the user terminal 11 via the network 500.
Further, the data input / output unit 101 has a function of acquiring data (including voice data) directly from input means including a microphone, a keyboard, and various sensors, such as a control signal for operating the information presentation server 10. doing.

  The dialogue processing unit 102 analyzes voice data from the user and estimates the contents of the user's voice data. Then, the dialogue processing unit 102 generates utterance data of a text sentence as an answer message (answer in dialogue) for the estimated content.

FIG. 2 is a block diagram illustrating a configuration example of the dialogue processing unit 102 in the present embodiment. In FIG. 2, the dialogue processing unit 102 includes an analysis unit 1021, a dialogue management unit 1022, and a generation unit 1023.
The analysis unit 1021 converts the voice data supplied from the data input / output unit 101 into text, that is, converts the text into text, and writes it in the dialogue history storage unit 107 for storage.

  In addition, the dialogue processing unit 102 performs morphological analysis of the text sentence after text conversion, performs numerical processing such as keyword extraction from the obtained morpheme, vectorization of the morpheme, and performs conversion processing of the text sentence into numerical data. Do. This conversion processing is performed using natural language processing technology or machine learning technology. In the present embodiment, for example, a technique such as keyword extraction by a tf (term frequency) -idf (inverse document frequency) method, vectorization by word2vec and doc2vec may be used. The dialogue processing unit 102 outputs the extracted keyword or vectorized numerical data to the dialogue management unit 1022.

  FIG. 3 is a diagram illustrating a configuration example of a dialogue history table stored in the dialogue history storage unit 107. In FIG. 3, the dialog history table includes fields for message time, speaker ID (Identification), message text, listening-oriented phrase flag, message ID1, and message ID2 for each record.

  Here, the message time indicates the time when voice data is input in the message or the utterance data is output in the message. The speaker ID indicates identification information of a speaker who outputs voice data or speech data (a user ID described later if the speaker is a user). For example, “U — 001” in the speaker ID is user identification information for identifying the user who is the speaker. In the speaker ID, “C — 001” is system identification information for identifying the system where the speaker is the information presentation server 10 (system). When the speaker is the information presentation server 10, there are a plurality of different system identification information because a different system answers for each type of content of the user's voice data (weather, health, science, etc.).

  In the message body, text data such as character data obtained by converting voice data into text or speech data is described. The listening-oriented phrase flag is a flag that is given to the message text indicating that the content of the voice data from the user is not a question and the voice based on the utterance data from the information presentation server 10 cannot be understood. Here, when the listening-oriented phrase flag is “0”, that is, when the flag is not set, it indicates that the corresponding message body is a normal conversation phrase.

On the other hand, when the listening-oriented phrase flag is “1”, that is, when the flag is set, the user does not understand the contents of the system voice for the corresponding message body (cannot hear or understand the contents). It shows that it is an intention phrase.
For example, in FIG. 3, the message body in which the listening-oriented phrase flag is “1” is “Since it could not be heard, please make another request”, and the voice that the information presentation server 10 outputs from the user terminal 11 cannot be heard. (According to listening-oriented parameters such as frequency, volume, and separation). This listening-oriented phrase is a word (corresponding to a positive word and a negative word described later) indicating the degree of ease of hearing of the audio content by the user with respect to the audio content that is a response to the user's request from the information presentation server 10, Or it is a synonym (synonym, synonym), and is registered as a predetermined phrase by an expert beforehand.

Message ID1 is identification information indicating the message text in the same record. On the other hand, the message ID 2 is identification information that is identification information of the immediately preceding message (that is, message ID 1) with respect to the message in the message body indicated by the message ID 1 in the dialogue between the user and the system.
Therefore, when confirming the relationship before and after in the dialogue between the user and the system, the message ID 2 in the record of the message body to be confirmed is read, and the message ID 1 that is the same as this message ID 2 is searched to retrieve the message to be confirmed. The message body of the message immediately before the body text can be searched, and each message body of continuous conversation can be easily confirmed.

  For example, message ID2: M180101001 is corresponding to message ID1: M180101003 with speaker ID C_001. For this reason, it is possible to easily search that the question with “How about this weekend?” Is “I want to go to ○○” of message ID 1: M180101001. The fact that each of message ID1 and message ID2 is not a continuous number indicates that another dialog may be performed in the middle, and messages in one dialog are not always input continuously.

  Returning to FIG. 2, the dialogue management unit 1022 refers to the dialogue history storage unit 107 based on the message ID 1 supplied together with each of the keyword and numerical data, and determines the user status (from the relationship between the message ID 1 and the message ID 2. Define whether the phrase is a conversational phrase or a listening-oriented phrase. When the message body is a conversation phrase, the dialogue management unit 1022 guides a system response to the user by the information presentation server 10 (for example, dialogue act such as genre designation, location designation, weather designation, traffic timetable designation, etc. Type).

  That is, the dialogue management unit 1022 may have a configuration in which responses (responses) from the system side are made into a database in response to a request (question) as a conversation phrase from a user, and a framework such as machine learning or reinforcement learning is used. It is also possible to generate a dialogue model that outputs the content of the response corresponding to the request and use it as a configuration. Since the processing of the dialog management unit 1022 is the same as that of a dialog system that is a generally known technique, detailed description thereof is omitted.

  When the dialogue management unit 1022 detects that the message ID is either a conversation phrase or a listening-oriented phrase from the keyword and numerical data of each message body of the message ID1, the dialogue management history table 107 of the dialogue history storage unit 107 Operate the listening-oriented phrase flag in the corresponding record. At this time, when it is determined that the message body is a conversation phrase, the dialogue management unit 1022 sets the listening-oriented phrase flag to “0” and does not set the flag. On the other hand, when it is determined that the message body is a listening-oriented phrase, the dialogue management unit 1022 sets the listening-oriented phrase flag to “1” and sets a flag.

  The generation unit 1023 generates utterance data, which is a response sentence of text data, using a system response model corresponding to the guideline of the system response output from the dialogue management unit 1022. That is, the generation unit 1023 inputs keywords extracted from the message body of the request, vectorized numerical data, and the like to the system response model, and generates utterance data having contents corresponding to the request. The system response model is generated so that utterance data of a response corresponding to the content of the request can be obtained by machine learning with teacher data of a known technique.

In addition, the generation unit 1023 does not use a system response model obtained by machine learning, but uses an external API (Application Programming Interface) for a previously created sentence template (sentence template). A method of obtaining necessary information and applying it to complete a sentence may be used.
For example, when the request is a train time, the generation unit 1023 searches for the train time using the external API of the timetable search based on the information of the departure station, the arrival station, and a predetermined time, and the search result Is inserted at a predetermined position of the template to generate response speech data.

In the present embodiment, either a system response model method based on machine learning or a method of filling a sentence template (sentence template) with external API information may be used.
Moreover, in this embodiment, it is good also as a structure which has the database in which the request and the response corresponding to this request were written beforehand. In this case, since the dialogue management unit 1022 extracts a response corresponding to the request from the database, the generation unit 1023 is not necessary.
The generation unit 1023 outputs at least the listening-oriented phrase flag data to the listening-oriented estimation unit 103 together with the generated utterance data.

Returning to FIG. 1, the listening orientation estimation unit 103 estimates the listening orientation of the user from the user attribute information (user attribute information), the conversation history, and the like, and provides a presentation guideline when supplying speech data to the user as speech. presume.
FIG. 4 is a block diagram illustrating a configuration example of the listening orientation estimation unit 103 in the present embodiment. In FIG. 4, a model construction unit 1031 and a listening orientation management unit 1032 are provided.
The model construction unit 1031 estimates a user's listening intention, and indicates a guideline (for example, an adjustment amount such as a volume, a reading speed, and a separation, which are listening-oriented parameters) for changing the characteristics of speech data and voice when presenting. A listening orientation estimation model is generated that performs the determining process. In the present embodiment, the listening-oriented parameter may be simply referred to as a parameter.

The listening orientation estimation model inputs the user's attribute information and text data of speech data, and outputs the parameters to be changed in listening orientation and the change amount of the parameters to be changed or other words to be replaced as estimation results. To do.
Also, the listening-oriented template model, like the listening-oriented estimation model, inputs the group attribute information and the text data of the utterance data, so that the change-target parameter and the change amount of the change-target parameter, Alternatively, another word to be replaced is output as an estimation result.

  Here, the model construction unit 1031 generates (constructs) mathematical formulas or rules for estimating the listening orientation as the listening orientation estimation model, and sequentially updates them. In the present embodiment, the listening orientation is a parameter corresponding to the user's hearing ability. When the user listens to the voice synthesized with the speech data, the frequency of the voice that is easy to hear, the playback speed, the segmentation of the phrase to be reproduced, this segmentation It shows the time width and so on.

  When the model construction unit 1031 generates the listening orientation estimation model, each dialogue of the dialogue history storage unit 107 and the dialogue action storage unit 109, as will be described later, using an algorithm such as machine learning, subject learning, or a neural network. Based on the history data in the change of the contents and listening-oriented parameters, the listening-oriented parameters in the listening-oriented estimation model and the estimation algorithm for estimating the amount of change of the parameters are optimized.

  That is, the model construction unit 1031 uses parameter estimation for estimating candidate words to be replaced, frequencies related to listening orientation in the sentence of the entire utterance data, reading speed, segmentation, and the like as mathematical formulas and rules necessary for estimation. Of the conversation history storage unit 107 and the dialogue action storage unit 109, and the history data in the change of the listening-oriented parameters are used as teacher data, and the listening-oriented estimation model (or a listening method described later) (Oriented template model) is constructed or updated.

In addition, listening orientation also includes the knowledge of whether or not the user can understand the meaning of the word, and it is replaced (changed) with other common synonyms (or synonyms and synonyms) that the user can understand. Is also included as one of the parameters.
Also, for listening orientation, if the words in the utterance data contain a fuzzing sound or a plosive sound corresponding to the frequency of the utterance described above, even if the frequency at the time of reading is lowered, the sound is spoken. Since it includes a high frequency, it is a synonym (or a synonym, a synonym), and it is also included as one of the parameters to replace it with another word that does not include a scratching sound or a plosive sound.

  The listening orientation estimation model described above adjusts each parameter such as the frequency of audio that is easy to hear, the playback speed, the frequency of segmentation of a phrase to be reproduced, and the time width of this segment, for example, corresponding to the user attribute information of each user A commonly used synonym that sets a quantity and replaces a specialized or esoteric word so that it can be understood by the user, including the setting of other words that can be understood, as well as fuzzing or popping sounds The replacement word may be configured as a database (database configuration), such as setting of other words that are synonyms (or synonyms, synonyms) and do not include a fuzzing sound or a plosive sound.

  The listening orientation estimation model may be configured as a machine learning model that performs machine learning using teacher data (machine learning model configuration). In the case of this machine learning model configuration, the model construction unit 1031 uses the data accumulated in each of the dialogue history storage unit 107, the user attribute storage unit 108, the dialogue action storage unit 109, and the grouping storage unit 110 as teacher data. By learning, a listening orientation estimation model for estimating the process of changing each parameter in the listening orientation for each user is generated.

  The listening orientation management unit 1032 uses the listening orientation estimation model generated (derived) for each user by the model construction unit 1031 to provide the presentation control unit 104 with the content of the utterance data change process corresponding to each user. Output. Here, the change process indicates a process of adjusting the frequency of the voice, the playback speed, the frequency of the segment of the phrase to be reproduced, the time width of the segment, and the replacement with a word that is easy to hear. .

In addition, the listening orientation management unit 1032 may not generate a listening orientation estimation model corresponding to the user who has input the requested voice data. In this case, the listening orientation management unit 1032 uses the listening orientation template model prepared in advance as a template, and presents the content of the utterance data change processing corresponding to the user for which the listening orientation estimation model has not been generated. Output to 104.
In addition, the listening orientation management unit 1032 uses the listening orientation template model of a similar group of users using grouping information and the like to be described later, and the content of the utterance data change processing corresponding to the user for which the listening orientation estimation model has not been generated May be output to the presentation control unit 104.

  The presentation control unit 104 relates to the replacement of words in the change of the utterance data supplied from the listening orientation estimation unit 103, and uses other replacement expressions stored in the language knowledge storage unit 111 to replace the target word. Extract words. This replacement table shows the correspondence between a word and a synonymous word that is replaced with this word. For example, as described above, the presentation control unit 104 converts a word having a fuzzing sound or a plosive sound such as “this time” for “this week”, “Saturday or Sunday” for “weekend”, The process of replacing with a word without a plosive is performed with reference to the replacement table.

Then, the presentation control unit 104 outputs the utterance data subjected to word replacement to the speech synthesis unit 105.
In addition, the listening orientation estimation unit 103 is configured not only to replace words but also to replace sentences (sentences) that include a plosive sound or scratching sound with synonymous (synonymous) sentences that do not include a plosive sound or scratching sound. May be.

  Here, the linguistic knowledge storage unit 111 is a synonym of a word that is difficult to understand based on linguistic knowledge about ease of listening as a listening-oriented parameter, and is another word that is commonly used and easy to understand. Other words (which may be a phrase as described above) that are synonyms of a word including a fuzzing sound or a plosive sound and do not include a flawing sound and a plosive sound are accumulated.

  For example, knowledge and corpora (text and utterances are collected on a large scale to create a database for medical staff and caregivers with older people (elderly people). Language knowledge storage unit 111 as a combination with another word that replaces a predetermined word using a dictionary of synonyms (synonyms, synonyms), a thesaurus (superior concept and subordinate concept of words), etc. Are stored in advance or additionally.

The speech synthesizer 105 performs speech synthesis on the speech data supplied from the presentation control unit 104 in response to a parameter to be changed and the amount of adjustment of the parameter, and speech content as a response to the user by speech synthesis. Is generated. At this time, the speech synthesizer 105, for example, as a listening-oriented parameter corresponding to the user's hearing ability and its adjustment amount, when the user listens to the speech synthesized from the speech data, the frequency of the voice that is easy to hear, playback The speech synthesis is performed by changing the reading speed at the time of reading, the segmentation of the phrase to be reproduced, the time width of the segmentation, and the like.
Then, the voice synthesis unit 105 outputs the voice content generated by the voice synthesis to the user terminal 11 via the data input / output unit 101.

The grouping estimation unit 106 searches the group attribute table for a group corresponding to the attribute data of the user whose history has newly occurred.
Then, the grouping estimation unit 106 adds the information to the group table of the searched group in the grouping storage unit 110 and stores it.

  FIG. 5 is a diagram illustrating a configuration example of a user attribute table stored in the user attribute storage unit 108. 5, the user attribute table has columns of user attribute items such as user ID, age, sex, volume, reading speed, separation, installation environment,... For each record. The user ID is identification information for identifying each user who uses the information presentation system 1 using the user terminal 11. The age indicates the age of the user identified by the corresponding user ID. The gender indicates whether the user identified by the corresponding user ID is male or female.

  The volume indicates the volume level (large, medium, or small) of the sound that the user identified by the corresponding user ID is easy to hear (can be heard). The reading speed indicates a voice speed level (fast, normal, or slow) that the user identified by the corresponding user ID can easily hear as an utterance. The delimiter indicates the amount (large, normal, small) of the number of phrases that are pronounced by the user identified by the corresponding user ID.

  In addition, the installation environment is the place where the user terminal 11 is installed, that is, the environment when listening to the voice, whether the room is large and the voice is easily propagated, the room is small and easily echoed, or other sounds are mixed The user's voice listening environment, such as whether there is a low possibility of being mixed or a high possibility of mixing other sounds, is shown. The user attribute storage unit 108 stores and stores a listening orientation estimation model corresponding to the user ID for each user.

  FIG. 6 is a diagram illustrating a configuration example of a dialogue action table stored in the dialogue action storage unit 109. In FIG. 6, the user attribute table has columns of items such as time, user ID, action type, action ID, implementation content, message ID,... For each record. The time indicates the time when a process (action) for applying some change to the speech data is performed. The user ID is identification information for identifying each user who uses the information presentation system 1 using the user terminal 11. The action type indicates either a system-initiated action (active) performed by the system side or a user-initiated action (passive) performed in response to a request from the user side. .

  The action ID is identification information for identifying the type of change of each of the system-driven action or the user-driven action. In FIG. 6, for example, action ID: A001 is “word replacement”, action ID: A003 indicates “change in reading speed”, and action ID: A004 indicates “change in break”. The execution content indicates the content of the change actually performed on the utterance data as an action. In FIG. 6, as an example of action ID: A001, “this week → this time” replaces the word “this week” with a similar word “this time”, and “weekend → Saturday, Sunday” changes the word “weekend” to “ It shows that the words are replaced with similar words (words having similar meanings) such as “Saturday and Sunday”.

  Here, the pronunciation of “shu” in “surround” is a rubbing sound and includes a high frequency component. Therefore, for a user who cannot easily hear a high frequency, it is necessary to replace it with a similar word that does not include the rubbing sound. . In addition, for users who are difficult to hear high frequencies, words containing plosives also contain high frequencies, so it is necessary to replace them with similar words that do not contain plosives, as in the case of scratching sounds.

Further, as an example of the action ID: A003, “speed: −” indicates a process in which the utterance reading speed is reduced. Further, as an example of the action ID: A004, “delimiter: +” provides a predetermined time between phrases in the utterance data, that is, a predetermined time (interval) after reading one phrase when reading out. To increase the frequency of reading the next phrase.
The message ID is identification information indicating the message in which the action is performed in the same record, and is the same identification information as the message ID 1 in FIG.

FIG. 7 is a conceptual diagram illustrating word replacement processing in the action shown in FIG. In FIG. 7, the speech uttered by each of the user who is a speaker and the user terminal 11 is visualized and explained as character data.
FIG. 7A shows word replacement as a system-led action. When the user 301 makes a request (question) “sound balloon?” In the speech balloon 351 to the user terminal 11 by voice, the information presentation server 10 reads “Saturday this week…” in the speech balloon 451. Is returned as a response. At this time, the listening orientation management unit 1032 refers to the listening orientation estimation model corresponding to the user 301, but it is not described that the user 301 is subjected to a change process on the utterance data. Therefore, the utterance data supplied by the dialogue processing unit 102 is used as a response (answer) as it is.

On the other hand, when the user 302 makes a request (question) “Oh wa?” Of the speech balloon 351 to the user terminal 11 by voice, the information presentation server 10 utterance data “Saturday of this week ...” of the speech balloon 451. Is replied as a response after changing the balloon 452 to “next Saturday…”. At this time, it is described that the listening orientation management unit 1032 refers to the listening orientation estimation model corresponding to the user 301 and performs processing for changing the utterance data for the user 302. The utterance data supplied by the processing unit 102 is changed according to the listening orientation estimation model.
That is, in the attribute information, the user 302 (for example, age 70s) is higher in age than the user 301 (for example, age 20s), and the user 302 is more likely to hear the voice having a lower frequency, so the listening orientation estimation is performed. The model has a replacement process for words including plosives and scratches.

  FIG. 7B shows word replacement as a user-initiated action. Although not shown, when the user 303 (for example, age 70's) makes a request (question) by voice to the user terminal 11, the information presentation server 10 displays “ The utterance data supplied by the dialogue processing unit 102 "Saturday of this week ..." is directly used as a response. However, in response to the voice “Saturday of this week ...”, the user 303 inputs a request (request) for a listening-oriented phrase “Eh?

  Therefore, the listening orientation management unit 1032 refers to the listening orientation estimation model corresponding to the user 301, and changes to the utterance data supplied to the dialogue processing unit 102 corresponding to the listening orientation phrase for the user 302. Processing is in progress. As a result, the information presentation server 10 outputs again to the user terminal 11 the voice data in which “Saturday of this week…” of the speech data balloon 453 is changed to “Saturday of this week…” of the balloon 454.

  FIG. 8 is a diagram illustrating a configuration example of a grouping table stored in the grouping storage unit 110. FIG. 8A shows a configuration example of a group attribute information table indicating group attribute information. In FIG. 8A, the group attribute information table is provided with columns of items such as group ID, age, sex, and residence for each record as an example. The group ID is identification information for identifying each group. The age indicates the range of the age of the users making up the group.

For example, the group ID: G — 001 indicates that the user is a collection of users included at least in the range of 60 to 75 years old. Similarly, the group ID: G — 002 indicates that it is a collection of users that are at least in the range of 10 to 20 years old. The gender indicates whether the human gender constituting the group identified by the corresponding group ID is male or female. The residence indicates which region the residence of the user constituting the group identified by the corresponding group ID is.
Further, for each group identified by this group ID, a listening-oriented template model corresponding to the above-described attributes of the users constituting each group is written and stored in the grouping storage unit 110 in advance.

  FIG. 8B shows a configuration example of a grouping table to which users belonging to each group ID are assigned. Each record has columns for group ID, user ID, age, gender, volume, reading speed, separation, and installation environment. The group ID is identification information for identifying each group. The user ID is identification information indicating users classified into the group indicated by the group ID in the same record, and is the same identification information as the user ID of the user attribute table in FIG.

  The age indicates the age of the user identified by the corresponding user ID. The gender indicates whether the user identified by the corresponding user ID is male or female. Here, when the gender is not included in the attribute in the grouping, there are both male and female users in the group.

  The volume indicates the volume level (large, medium, or small) of the sound that the user identified by the corresponding user ID is easy to hear (can be heard). The reading speed indicates a voice speed level (fast, normal, or slow) that the user identified by the corresponding user ID can easily hear as an utterance. The delimiter indicates the amount (large, normal, small) of the number of phrases that are pronounced by the user identified by the corresponding user ID.

  In addition, the installation environment is the place where the user terminal 11 is installed, that is, the environment when listening to the voice, whether the room is large and the voice is easily propagated, the room is small and easily echoed, or other sounds are mixed The user's voice listening environment, such as whether there is a low possibility of being mixed or a high possibility of mixing other sounds, is shown.

As described above, in each group, users having the same attributes as the group attributes (group attributes) defined in the grouping table are classified.
The types of attributes in the above-mentioned grouping are the types of attributes presented by scholars and doctors who are familiar with human speech audibility, or experts who are familiar with listening orientations of nursing facility staff (such as nurses and caregivers). It may be used.
Further, the attribute type in the grouping may be set by performing a process such as clustering using the attributes of a plurality of users as the feature amount and extracting the attribute type of the feature amount that can most clearly classify the user verb.

Even if the type of attribute used for grouping is extracted by any of the above-mentioned processes, it is necessary to change parameters such as volume, reading speed, and division in the listening orientation of the listening-oriented template model in accordance with the clinical trial of the expert. The adjustment amount of each parameter when changing may be set.
For the user immediately after starting the use of the information presentation server 10 in the present embodiment, whether the information presentation server 10 needs to change the listening-oriented parameter described above, and the data of the parameter adjustment amount at the time of the change, It cannot be extracted sufficiently from the user's response to listening orientation.

For this reason, the listening orientation estimation unit 103 searches the grouping table for a group that is close to the user's attribute for a user whose listening-oriented parameter data has not been sufficiently extracted, and obtains the group obtained by the search. The necessity of listening-oriented parameters or the amount of parameter change is estimated using a listening-oriented template model.
Based on the listening orientation template model used for the user's attributes, the listening orientation estimation unit 103 changes whether or not each parameter in the listening orientation needs to be changed and the adjustment amount data when the change is made to the voice from the user To obtain each user's listening orientation estimation model. At this time, as described above, the listening orientation estimation unit 103 may generate the listening orientation estimation model by performing optimization processing by machine learning on the listening orientation template model.

  FIG. 9 is a flowchart showing an operation example of the dialogue system using the information presentation system of the present embodiment. The operation of the flowchart of FIG. 9 is performed by, for example, accessing the information presentation server 10 in the information presentation system 1 and transmitting a voice request from the user terminal 11 such as a smart speaker to the information presentation server 10 by voice. It is started when a dialogue with the information presentation server 10 is performed. The following description of the operation of the flowchart of FIG. 9 shows that each of the listening orientation template models corresponding to the listening orientation for each group is generated in the listening orientation estimation unit 103 as described above and stored in the grouping storage unit 110. In the state of being.

Step S101:
The data input / output unit 101 determines whether audio data is supplied from any of the user terminals 11. Then, when audio data is supplied from any of the user terminals 11, the data input / output unit 101 advances the process to step S2. On the other hand, if no audio data is supplied from any user terminal 11, the data input / output unit 101 repeats the process of step S101.

At this time, for example, the user inputs a request for inquiring of the schedule such as a concert or the like to the user terminal 11 by voice. And the user terminal 11 accesses the information presentation server 10, and transmits each of audio | voice data and the user ID of the user who input this audio | voice. In this case, the data input / output unit 101 detects that audio data is supplied from any user terminal 11, and advances the process to step S102.
Then, when proceeding to step S102, the data input / output unit 101 outputs the input voice data to the dialogue processing unit 102. Further, the data input / output unit 101 outputs the input user ID to the listening orientation estimation unit 103.

Step S102:
When the user ID is supplied from the data input / output unit 101, the listening orientation estimation unit 103 determines whether or not the user indicated by the user ID has a conversation history. That is, the listening orientation estimation unit 103 refers to the user attribute storage unit 108 and determines whether or not a listening orientation estimation model is stored corresponding to this user ID. That is, if the user has no history of dialogue, the listening orientation estimation model is not generated from the listening orientation template model.
At this time, when the listening orientation estimation unit 103 stores the listening orientation estimation model corresponding to the user in the user attribute storage unit 108, the processing proceeds to step S103. On the other hand, when the listening orientation estimation unit 103 does not store the listening orientation estimation model corresponding to the user in the user attribute storage unit 108, the listening orientation estimation unit 103 advances the process to step S104.

Step S103:
The listening orientation estimation unit 103 reads the listening orientation estimation model corresponding to the user ID from the user attribute storage unit 108.

Step S104:
The listening orientation estimation unit 103 refers to the user attribute storage unit 108 and reads the user attribute information corresponding to the user ID.
Then, the listening orientation estimation unit 103 searches the group attribute information table of the grouping storage unit 110 for a group having attribute information close to the read attribute information, and reads the listening orientation template model of the group obtained by the search.
Further, the grouping estimation unit 106 adds the information to the group table of the searched group in the grouping storage unit 110 and stores it.

Step S105:
The dialogue processing unit 102 converts voice data into text data, performs morphological analysis, and determines whether the voice data is a conversation phrase or a listening-oriented phrase from the obtained word or phrase. . The fact that the voice data is a listening-oriented phrase means that the user is requesting optimization of the voice of the speech data (response) (change according to his / her listening preference).

Therefore, the dialog processing unit 102 determines whether or not the user requests optimization of the speech data voice in step S105.
Then, when the user does not request the optimization of the speech data, the dialogue processing unit 102 advances the process to step S106. On the other hand, the dialogue processing unit 102 advances the process to step S107 when the user requests optimization of the speech data.
At this time, the dialogue processing unit 102 writes the text data of the input voice data, the flag of the listening-oriented phrase in the case of the listening-oriented phrase, and the message ID in the dialogue history table in the dialogue history storage unit 107. .

Step S106:
Since the input voice data is a conversation phrase, the dialogue processing unit 102 generates utterance data corresponding to the voice data of the request using each word obtained by morphological analysis of the text sentence of the voice data.
Then, the listening orientation estimation unit 103 estimates the change processing for the utterance data led by the system using the listening orientation estimation model or the listening orientation template model, that is, the parameter to be changed from among the listening orientation parameters of the user, Estimate the amount of change (or word replacement).
The listening orientation estimation unit 103 also outputs to the presentation control unit 104 the parameters to be changed selected from the listening orientation parameters and the amount of change (or replacement word) of the parameters.

Step S107:
Since the input voice data is a listening-oriented phrase, at this point, the conversation phrase as a response to the request has already been obtained in the previous conversation phrase loop in this flowchart.
For this reason, the listening orientation estimation unit 103 adjusts the parameter to be changed from the listening orientation parameters and the amount of change of this parameter by using the listening orientation estimation model or the listening orientation template model. Output.

  At this time, the listening orientation estimation unit 103 writes and stores the word replacement process in the dialog action table of the dialog action storage unit 109. At this time, the listening orientation estimation unit 103 describes whether the action type is system-driven or user-driven. In addition, the listening orientation estimation unit 103 describes the action ID given to each action in advance, and describes the implementation content corresponding to the action ID (see the interactive action table in FIG. 6 for a description example). When the implementation content is word replacement (action ID: A001), which word is replaced by what word by the presentation control unit 104 is described in the implementation content column of the dialogue action table.

Here, for example, when the parameter to be changed is a volume, a larger volume is set as a change amount (adjustment of the change amount) with respect to a size that is previously defined as a change amount from the normal volume. In addition, when the parameter to be changed is the reading speed, the slower speed is set as the changing amount with respect to the slow speed defined as the changing amount from the normal reading speed in advance. In addition, when the parameter to be changed is a delimiter, a larger delimiter frequency is set as the change amount than the delimiter frequency defined in advance as the amount of change from the normal delimiter frequency.
In addition, the amount of change for each of these parameters may be changed collectively, or the amount of change may be adjusted by determining the order of change each time the flowchart loop is repeated.

Step S108:
In response to the request for word replacement processing in the listening orientation supplied from the listening orientation estimation unit 103, the presentation control unit 104 extracts each of the words having a fuzzing sound and a plosive sound included in the text data of the utterance data. Then, the presentation control unit 104 extracts replacement words corresponding to the extracted words with reference to the replacement table of the language knowledge storage unit 111.
Then, the presentation control unit 104 outputs, to the speech synthesizer 105, the speech-oriented parameters and the amount of change of the parameters of the speech data that has been replaced with words that are easy to hear.

The speech synthesizer 105 performs speech synthesis on the speech data supplied from the presentation control unit 104 in response to a parameter to be changed and the amount of adjustment of the parameter, and speech content as a response to the user by speech synthesis. Is generated.
Then, the voice synthesis unit 105 outputs the voice content generated by the voice synthesis to the user terminal 11 via the data input / output unit 101.

Step S109:
The listening orientation estimation unit 103 refers to the message ID in the dialogue behavior table of the dialogue behavior storage unit 109 and extracts a message ID continuous to the message ID from the dialogue history table of the dialogue history storage unit 107.
Then, when the listening-oriented phrase flag of the message body corresponding to the extracted message is “0”, the listening-oriented estimation unit 103 determines that the listening-oriented parameter change or word replacement has succeeded. On the other hand, when the listening-oriented phrase flag of the message body corresponding to the extracted message is “1”, the listening-oriented estimation unit 103 changes the listening-oriented parameter or replaces the word to improve the ease of listening. Is determined to be insufficient.

The listening orientation estimation unit 103 has, for example, the above listening preference phrase flag “1” and the listening orientation parameter changed by the user and the amount of change of the parameter, and the listening orientation estimation model corresponding to the user. Is corrected to suit the user's listening preference.
Further, the listening orientation estimation unit 103 refers to the group table of the grouping storage unit 110, extracts the changed listening orientation parameters and the amount of parameter change common to each user constituting the group, and listens to the listening orientation. A correction process for adapting the template model to the listening preference of the users included in the group is performed.

  At this time, the listening orientation estimation unit 103 performs, for example, a morphological analysis of the message body corresponding to the message ID for which the listening orientation phrase flag is set, and extracts a positive word or a negative word. A process of adding “1” to the value (incrementing the evaluation value) is performed. On the other hand, in the case of a negative word, a process of subtracting “1” from the evaluation value for the change (decrementing the evaluation value) is performed. Then, when the evaluation value exceeds a predetermined threshold value, the listening orientation estimation unit 103 reflects the changed listening orientation parameter change amount (or the replaced word) in the listening orientation estimation model and the listening orientation template model. You may comprise.

  Also, the listening orientation estimation unit 103 does not extract negative words or positive words, for example, but changes the listening orientation parameters and outputs the audio content, and then “is it easy to hear?“ Yes ”/ Text data such as “Please answer with“ No ”” or “Would you like to read it a little more slowly? Please answer as it is” / “Please answer slowly” is voice-synthesized by the voice synthesizer 105 and changed to the confirmation voice content. In addition, the listening orientation estimation unit 103 transmits the confirmation audio content to the user terminal 11 via the data input / output unit 101. This algorithm corresponds to user-initiated listening-oriented parameter changes.

Then, the listening orientation estimation unit 103 inputs a user's answer to the above-described confirmation audio content. At this time, the listening intention estimation unit 103 inputs the answer data obtained by converting the voice data of the answer by the user into the text by the dialogue processing unit 102.
Then, for example, when the answer data to the question “Is it easy to hear?” Is “Yes”, the listening orientation estimation unit 103 determines that the listening orientation parameter has been successfully changed. On the other hand, if the answer data to the question “Is it easy to hear?” Is “No”, it is determined that the change of the listening-oriented parameter was not successful.

As a result, the listening orientation estimation unit 103 reflects the successful listening orientation parameter changing process in the listening orientation estimation model and the listening orientation template model when successful.
On the other hand, if the change is unsuccessful, the listening intention estimation unit 103 generates again the audio content in which other listening-oriented parameters are changed, and outputs it as a response to the user.

Further, for example, when the answer data to the question “Let's read a little more slowly?” Is “No change”, the listening orientation estimation unit 103 determines that the change of the reading speed, which is a listening orientation parameter, has succeeded. On the other hand, if the answer data to the question “Do you want to read it a little more slowly?” Is “slow”, it is determined that it was not successful because the amount of change in the reading speed, which is a listening-oriented parameter, is small.
Thus, the listening orientation estimation unit 103 reflects the reading speed change process, which is a successful listening orientation parameter, in the listening orientation estimation model and the listening orientation template model when successful.

On the other hand, when the change is unsuccessful, the listening orientation estimation unit 103 increases the amount of change in the reading speed that is a listening-oriented parameter, that is, generates audio content that has been changed to further reduce the reading speed, Output as a response to the user.
As described above, after making changes such as listening-oriented parameters such as frequency, reading speed, break and word replacement, the user is asked whether each change is appropriate, and listening-oriented parameters A configuration may be adopted in which the success / failure of the change is confirmed, and the confirmation result is reflected in the listening orientation estimation model and the listening orientation template model.

  In addition, the listening orientation estimation unit 103 counts the occurrence frequency at which the listening orientation phrase flag is “1”, and when the occurrence frequency count for performing the same listening orientation parameter change exceeds a predetermined set value. Based on a parameter whose occurrence frequency exceeds a predetermined set value, the parameter and the amount of parameter change may be reflected in the listening orientation estimation model and the listening orientation template model.

Step S110:
The presentation control unit 104 writes the text data of the utterance data in the dialog history table of the dialog history storage unit 107 in the message body, adds the message ID1, and stores it. At this time, the presentation control unit 104 writes and stores the system identification information of the system that performs the response in the column of the speaker ID.
Since the presentation control unit 104 is a conversation phrase, the listening-oriented phrase flag is set to “0”, and the message ID 1 of the message body of the voice data of the connected user is written and stored in the message ID 2 column.

  With the above-described configuration and operation, according to the present embodiment, the information presentation server 10 that responds to the user's request can listen to the information provided by the audio content so that each user can accurately hear the information. Compared to the conventional rule-based configuration of estimating listening orientation for each user or each group because the estimation of orientation is performed using the listening orientation estimation model for each user or the listening orientation template model for each group. As a result, the amount of data can be reduced, and since the amount of data is small, each of the listening orientation estimation model and the listening orientation template model (a correction process corresponding to the user in sequence) can be easily performed.

  In addition, according to the present embodiment, by the listening orientation estimation model corresponding to each attribute information of the user, the types of parameters that improve the ease of listening in the listening orientation of the user, and the amount of change (adjustment amount) of these parameters, And a word including a fuzzing sound or a plosive sound in the utterance data that is a response to the user's request is extracted, and a fretting sound and a plosive sound that are synonyms (or synonyms or synonyms) are extracted corresponding to the context of the utterance data. Since it is replaced with another word that is not included, the voice content obtained by voice synthesis of the utterance data can be made easy to hear by the user.

  In addition, according to the present embodiment, by the listening orientation estimation model corresponding to each attribute information of the user, the types of parameters that improve the ease of listening in the listening orientation of the user, and the amount of change (adjustment amount) of these parameters, When speech synthesis is performed on speech data that is a response to a user request, the speech content that is speech synthesized speech is changed to the user in order to change the frequency of speech to be spoken, the reading speed, and the segmentation. Can be easily heard.

  In addition, according to the present embodiment, for a user who has no conversation history, a listening-oriented template model generated corresponding to a group composed of other users whose attribute information is similar to this user is used. Speech that is synthesized from speech data in order to replace a word containing a fuzzing sound or a plosive sound in the utterance data described above with another word that does not contain a fuzzing sound and a plosive sound that are synonyms corresponding to the context of the utterance data. Since the content is a voice that can be easily heard by the user, it is possible to improve the ease of listening to the voice content of the response even for a user who has no history.

  In addition, according to the present embodiment, for a user who has no conversation history, a listening-oriented template model generated corresponding to a group composed of other users whose attribute information is similar to this user is used. When speech data that is a response to a user's request is synthesized, the frequency of the speech that is spoken, the reading speed, and the segmentation are changed, so that the user can easily hear the audio content that is the synthesized speech data. Since the voice of the state is used, it is possible to improve the ease of listening to the voice content of the response even for a user who has no history.

  Further, according to the present embodiment, each of the listening orientation estimation model and the listening orientation template model is stored in the dialogue history storage unit 107 and the dialogue action storage unit 109, respectively, on the request side (user) and the response side (information Since the history data in the dialogue with the presentation server 10) is sequentially changed, it is possible to improve the ease of hearing the voice corresponding to the user or group attribute information.

  In this embodiment, the response side has been described as a computer dialogue system. However, when each of the request side and the response side is a user (human), the mutual hearing between the interacting users is improved. In addition, a configuration may be used in which a process for replacing words in the utterance data on the response side in the dialogue and a parameter changing process for improving the ease of listening at the time of speech synthesis are performed by a listening orientation estimation model corresponding to each user.

FIG. 10 is a conceptual diagram showing another configuration example of the information presentation system in which the user and the system interact with each other according to the embodiment of the present invention.
The information presentation system 1A includes an information presentation server 10A, a user terminal 11_1, a user terminal 11_2, a user terminal 11_3, a user terminal 11_4, a dialogue server 12_1, a dialogue server 12_2, a dialogue server 12_3, and a dialogue server 12_3 via a network 500. It is connected.
Each of the user terminal 11_1 and the user terminal 11_2 is the smart speaker already described, and the user inputs a request by voice and notifies the user of the response from the information presentation server 10A as voice content.

On the other hand, the user terminal 11_3 is a mobile terminal such as a smartphone or a tablet computer, and is provided with a display screen. The user terminal 11_4 is a personal computer and is provided with a display screen.
For each of the user terminal 11_3 and the user terminal 11_4 having a display screen, the information presentation server 10A is not an audio content but an image content (character content, moving image, or stamp image) that can be visually recognized visually. Etc.) (switching the output) and outputting as a response to the request.

The information presentation server 10A has the same configuration as the information presentation server 10 in FIG. 1 described above, but does not have a dialog system function in the dialog processing unit 102.
Each of the dialogue server 12_2, the dialogue server 12_3, and the dialogue server 12_3 is a device that replaces the function of the dialogue system of the dialogue processing unit 102 in the information presentation server 10. Each of the dialogue server 12_2, the dialogue server 12_3, and the dialogue server 12_3 is a dialogue system that performs, for example, weather forecast confirmation, transportation time confirmation, and user plan confirmation.

  In the case of this configuration, the information presentation server 10A inputs response text data as utterance data from each of the dialogue server 12_2, the dialogue server 12_3, and the dialogue server 12_3. Make changes that improve ease of listening.

  In the present invention, a program for realizing the functions of the information presentation server 10 in FIG. 1 and the information presentation server 10A in FIG. 10 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is recorded. May be read by the computer system and executed to change the audio content so that the user can more easily hear it. Here, the “computer system” includes an OS and hardware such as peripheral devices.

  The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, and a CD-ROM, and a storage device such as a hard disk built in the computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

Moreover, the information presentation server 10 including the dialogue server 12 and the user terminal 11 realize the transmission of the program described above with respect to the system and program shown in FIG. However, without being limited thereto, for example, the function of the information presentation server 10 including the dialogue server 12 may be mounted on the user terminal 11 without using a network if possible.
Moreover, it is good also as a structure which implement | achieves each function of the information presentation server 10 and the dialogue server 12 by one server instead of the structure which each of the information presentation server 10 and the dialogue server 12 provides as an independent apparatus.

DESCRIPTION OF SYMBOLS 1,1A ... Information presentation system 10, 10A ... Information presentation server 11, 11_1, 11_2, 11_3, 11_4 ... User terminal 12_1, 12_2, 12_3 ... Dialog server 101 ... Data input / output part 102 ... Dialog processing part 103 ... Audition intention estimation Unit 104 ... presentation control unit 105 ... speech synthesizer 106 ... grouping estimation unit 107 ... dialog history storage unit 108 ... user attribute storage unit 109 ... dialogue action storage unit 110 ... grouping storage unit 111 ... language knowledge storage unit 500 ... network 1021 ... Analysis unit 1022 ... Dialog management unit 1023 ... Generation unit 1031 ... Model construction unit 1032 ... Listening orientation management unit 1023 ... Generation unit

Claims (10)

In dialogue with the user, the listening orientation indicating the ease of hearing of each voice of the user is estimated, and utterance data supplied by voice to the user is changed corresponding to the listening preference of the user. A listening orientation estimator for generating and updating a change model;
A presentation control unit that corresponds to the change model set for each user, and that changes the utterance data, which is an answer by voice from a dialogue system, to be the listening-oriented for each of the users. Characteristic information presentation system. Rules for utterances from users, which are used when estimating the listening orientation, and that the dialogue history which is the history of the dialogue with each of the users is written and stored for each user in the dialogue history storage unit The information presentation system according to claim 1, further comprising: a dialogue processing unit that determines a response based on the information. The listening orientation estimation unit
From the evaluation of the utterance data in the dialog of the user, the user's listening intention is extracted, and each of the attribute information of the user and the intention information indicating the listening intention of the user is stored in the user attribute storage unit. The information presentation system according to claim 1, wherein the information is written and stored for each user. In accordance with the attribute information of each of the users, grouping that classifies each of the users is performed, and a template modification model that is a modification model for each of the classifications according to the listening orientation common to the users included in each classification The information presentation system according to claim 3, further comprising: a grouping estimation unit that generates The listening orientation estimation unit
For the user for whom the change model is not prepared, the template change model of the classification corresponding to the user is extracted, and corresponding to the listening orientation extracted in the dialogue, the user corresponding to the user is extracted. The information presentation system according to claim 4, wherein a change model is generated. The attribute information is
The information presentation system according to any one of claims 3 to 5, wherein the information presentation system is set as a combination of at least demographic data including age, sex, and residence of the user. The change model is
The processing for changing at least the replacement of words in the utterance data determined through the dialogue processing unit, the frequency and speed of speech when reading the utterance data, and the division of phrases is shown. Information presentation system described. The presentation control unit
The change content which is the content changed the utterance data by the change model is written and stored as a change history in the dialogue action storage unit,
The listening orientation estimation unit
The information presentation system according to any one of claims 1 to 7, wherein the listening orientation of the user is extracted from the history of the conversation and the change history. The listening orientation estimation unit estimates the listening orientation indicating the ease of hearing of each voice of the user in the dialog with the user, and the speech data supplied to the user by voice is used as the listening orientation of the user. Listening-oriented estimation process for generating and updating a change model that changes in response to
The presentation control unit responds by voice from the dialogue system via the dialogue processing unit that corresponds to the change model set for each user and determines a response to the utterance from the user based on the rule. A presentation control step of changing the utterance data so as to be listening-oriented for each of the users. Computer
In dialogue with the user, the listening orientation indicating the ease of hearing of each voice of the user is estimated, and utterance data supplied by voice to the user is changed corresponding to the listening preference of the user. Listening-oriented estimation means for generating and updating change models,
Corresponding to the change model set for each user, the utterance data, which is an answer by voice from a dialogue system, is made to function as a presentation control means for changing the user to be listening-oriented. program.

Images