JP7211384B2 - Voice recognition device, personal identification method and personal identification program - Google Patents

Description

The present invention relates to a voice recognition device, a personal identification method, and a personal identification program for recognizing speech to identify an individual.

With the development of speech recognition, machine operation by speech recognition is performed by various methods. Among them, machine operation by user's free speech is used in products called smart phones and smart speakers, and is becoming mainstream recently. Many of these are often realized by learning models created for the use of "speech recognition" and "sentence recognition" using machine learning in recent years.

In addition, due to the generalization and price reduction of voice recognition technology, it has become possible to utilize voice recognition technology not only for business purposes but also for personal purposes. As a result, it is now required to operate devices by voice instead of visually and manually. In particular, there are many requests to operate communication means represented by voice calls by voice. At that time, one of the problems is "name resolution", which is a process of linking a partner (ID) on the system with a partner to be communicated with.

Techniques for recognizing speech in a system and converting it into a character string are widely known. A technique of morphological analysis for estimating a part of speech based on an input sentence is also widely known. As a result, for example, by detecting the information "noun: Ichiro Tanaka" and "verb: telephone (to call)" for the input "call Ichiro Tanaka", it is possible to make a call to Mr. Ichiro Tanaka's phone number. be possible.

Further, Patent Literature 1 describes a telephone answering support device capable of specifying a dialogue partner without creating a large-scale database. The apparatus described in US Pat. No. 5,900,003 queries a caller database that stores information about the caller and historical information about the caller's interlocutors to obtain information about the caller and Based on the history information, the conversation partner corresponding to the content of the voice data transmitted from the caller is identified.

JP-A-2003-158579

As a premise of machine learning, it is necessary to prepare a huge amount of data for learning to create a learning model. Regarding "speech recognition", it is possible to construct a versatile model that extracts the user's utterances. However, with regard to "meaning recognition," it is difficult to uniquely recognize the meaning of a sentence because the words are blurred due to various factors such as the user's experience, environment, and relationship with others.

In order to uniquely recognize the meaning of a sentence, for example, a system is conceivable in which words and phrases recognized by a machine are presented to the user in advance, and the user is instructed using the presented words and phrases. However, it is very stressful for the user to make the user recognize and speak uniform words and phrases. In the case of a method using machine learning, for example, a method of constructing a learning model of sentence recognition for each individual is assumed. However, since it is necessary to prepare "correct answers" as learning data, and it takes time to collect learning data and perform actual operations, there were issues in terms of adaptability and responsiveness.

Also, morphological analysis is often designed on the premise that "correct sentences" are input, so it is difficult to deal with broken Japanese like spoken language. In fact, in the spoken language used for communication by Japanese people, words are often omitted from the original sentences. For example, there may be a case where only a pair of surname and position is spoken, or a case where only position is spoken.

However, except for examples such as the president, there are generally a plurality of persons with the same position within a company. Also, depending on the position (affiliation) of the person who made the statement, there are various individuals who correspond to the position. Therefore, in the case of the method of simply recognizing words, it is difficult to identify individuals only by job titles. On the other hand, it is not realistic to hold a large amount of settings and information assuming various situations in which such spoken words are uttered.

In addition, since the device described in Patent Document 1 assumes that a part of the name of the individual is known, it is difficult to identify the individual only from the title of the organization. In addition, since the apparatus described in Patent Document 1 identifies a dialogue partner based on history information, it must retain past history information, which leaves a problem in terms of storage.

Considering the characteristics of spoken language as described above, in order to improve the convenience of machine operation, it is necessary to improve the accuracy of identifying individuals by job title etc. in an organization using voice information with simple settings and logic. is preferred.

Accordingly, the present invention aims to provide a speech recognition apparatus, an individual identification method, and an individual identification program capable of improving the identification accuracy with simple settings and logic when identifying an individual existing in an organization by voice. aim.

The speech recognition apparatus of the present invention is a speech recognition apparatus for identifying individuals based on voice, and extracts target words, which are words representing roles in an organization, from character strings converted into text based on the user's voice. A target word extracted based on a target word extracting unit, personal information that associates an individual's identification information with affiliation information that is information representing the organization to which the individual belongs and the role of the individual in the organization, and the extracted target word indicates An individual identification unit that identifies individual candidates according to the relationship with the organization to which the user belongs, and an output unit that outputs the identified individual candidates, wherein the individual identification unit outputs individual candidates during working hours. is determined to be higher than the reliability of individual candidates who are not working hours, and when there are multiple individual candidates with the same identification information, the user in the organization to which the user belongs It is characterized in that candidates for individuals other than those in higher positions are preferentially selected over candidates for individuals in higher positions, and the output unit outputs the candidates for individuals based on the degree of reliability.

The personal identification method of the present invention is a personal identification method for identifying an individual based on voice, and extracts a target word, which is a word representing a role in an organization, from a character string converted into text based on the user's voice. Then, based on the personal information that associates the identification information of the individual with the affiliation information that is information representing the organization to which the individual belongs and the role of the individual in that organization, the candidate of the individual indicated by the extracted target word is Identify according to the relationship with the organization to which the user belongs, determine the reliability of the individual candidate during working hours so that it is higher than the reliability of the individual candidate who is not working hours, and provide the same identification information If there are multiple candidates for individuals who have personal information, candidates for individuals other than those in higher positions in the organization to which the user belongs are preferentially selected and identified. It is characterized by outputting individual candidates based on the degree of reliability.

The personal identification program of the present invention is a personal identification program applied to a computer that identifies an individual based on voice, and the computer identifies a role in an organization from a character string converted into text based on the user's voice. Based on the target word extraction process for extracting the target word that is the word to be represented, and the personal information that associates the identification information of the individual with the affiliation information that is information representing the organization to which the individual belongs and the role of the individual in that organization, An individual identification process for identifying individual candidates indicated by the extracted target words according to the relationship with the organization to which the user belongs, and an output process for outputting the identified individual candidates, and in the individual identification process, Determine the reliability of individual candidates during working hours so that it is higher than the reliability of individual candidates who are not during working hours, and if there are multiple individual candidates with the same identification information, the user's Preferential selection of candidates for individuals other than those in higher positions than candidates for individuals in higher positions than the user in the organization to which the user belongs, and outputting candidates for individuals based on the degree of reliability in the output process. characterized by

According to the present invention, when an individual existing in an organization is identified by voice, the identification accuracy can be improved with simple settings and logic.

1 is a block diagram showing a configuration example of an embodiment of a speech recognition device according to the present invention; FIG. FIG. 4 is an explanatory diagram showing an example of organization information; FIG. 4 is an explanatory diagram showing an example of a dictionary; FIG. 4 is an explanatory diagram showing an example of a relationship between an organization and persons belonging to the organization; FIG. 4 is an explanatory diagram showing an example of information included in a dictionary; FIG. 4 is an explanatory diagram showing an example of a schedule in an organization; FIG. 2 is an explanatory diagram showing a specific configuration example of a speech recognition device; 4 is a flowchart showing an operation example of the speech recognition device; FIG. 4 is an explanatory diagram showing a specific example of processing using a speech recognition device; 1 is a block diagram showing an overview of a speech recognition device according to the present invention; FIG.

The present invention focuses on the general culture of an organization in voice-based device operation, especially in communication functions that require the identification of individuals, and adds a small amount of additional information about the organization to the list (name) to improve the user's convenience. It is intended to ensure the In addition, by holding additional information other than the information about the organization, a mechanism that can be applied in various situations is provided. BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a configuration example of an embodiment of a speech recognition device according to the present invention. The speech recognition device 100 of this embodiment is a device that identifies an individual based on speech. The speech recognition device 100 includes an input unit 10, a speech data generation unit 20, a speech recognition unit 30, a target word extraction unit 40, an individual identification unit 50, an output unit 60, and a storage unit 70. .

The input unit 10 receives speech input from the user of the speech recognition apparatus 100 . Specifically, the input unit 10 records the user's voice and converts it into an electrical signal. The input unit 10 is implemented by, for example, a microphone. Since the input unit 10 thus has a function of recording the user's voice, the input unit 10 can be called a recording unit.

Further, the input unit 10 may receive input of information for identifying the user (hereinafter referred to as user identification information) together with voice input by the user. User identification information may be input before recording, or may be generated by converting input from the outside using any personal authentication technology. Note that when the user can be identified from the input voice, the user identification information does not have to be input separately from the input of the voice.

The audio data generation unit 20 generates audio data from the received audio. The audio data generation unit 20 generates audio data by, for example, performing AD (Analog-to-Digital) conversion on the converted electric signal.

The speech recognition unit 30 converts the generated speech data into text. Text conversion means a process of converting voice data into characters or character strings (single sentences) composed of character codes. A character string is a single word or a set of multiple words. The speech recognition unit 30 may convert speech data into text using commonly known speech recognition technology. Note that any language may be used for text conversion, and may be, for example, Japanese or English. In this embodiment, it is assumed that the input voice is converted into a character string as it is.

When the user identification information is not input, the input unit 10, the voice data generation unit 20, and the voice recognition unit 30 generate the user identification information by specifying the speaker using any voice recognition technology. may

The storage unit 70 stores various types of information that the speech recognition apparatus 100 uses for processing. In this embodiment, the storage unit 70 stores, in addition to the name and ID, the prerequisite information such as the role in the organization to which the individual belongs, the unique rules of the operating organization, a table corresponding to vocabulary, and the like. In the following description, a collection of these tables will be referred to as a "dictionary".

The dictionary stored in the storage unit 70 will be described in detail below. Personal information is registered in the dictionary as basic information. Personal information includes an individual's name and, as additional information, the organization to which the individual belongs and the role of the organization. In the following description, information representing the structure of an organization is referred to as organization information, and information representing an organization to which an individual belongs and the role of the individual in that organization is referred to as affiliation information. In the case of a company, the roles of individuals include, for example, job titles.

Personal information is information in which personal identification information (name, ID, etc.) and belonging information are associated with each other. Information other than the affiliation information may be associated with the personal information. Affiliation information expressed using a separator is referred to as an organization path. An organization path can be specified by any method. For example, an organization path may be specified using a separator in order from the top of the organization having a hierarchical structure.

FIG. 2 is an explanatory diagram showing an example of organization information. The organizational information exemplified in FIG. 2 is that Cop Co., Ltd. has Department D, which includes Sections A and B under Section D, and Sections X and Y under Section A. indicates that

When the organization information illustrated in FIG. 2 is represented by an organization path, for example, if the separator of the organization path is "/", the X group of the A section included in the D department of Cop Co., Ltd. is "Cop Co., Ltd. / D department / Section A / Group X". For example, for the manager of Section A, a job title can be specified such as "Cop Co., Ltd./Department D/Section A/Group X/Manager." Furthermore, by using this organization path, it is possible to designate a virtual organization such as an executive committee. For example, the Com committee of Cop Corporation can be pointed out as "Cop Corporation/Com Committee".

Although "/" is used as the separator in the above example, the symbol used for the separator is not limited to "/". This symbol may be defined uniquely for each system environment using characters or character strings that are not included in the organization. Such a method of separating character strings using characters or character strings as separators is widely known and can be implemented on various information platforms.

Also, in the above example, the case where the structure of the organization is a hierarchical structure has been described. However, when there is no hierarchy in the organization (when there is no hierarchical relationship in the organization), the structure of the organization does not necessarily have to be a hierarchical structure.

Further, the dictionary may include information in which role names are associated with organizations as related information other than personal information. Information in which the organization and the role name are associated with each other is hereinafter referred to as a role table.

Also, this role table may be used as replacement information when performing name resolution for each organization's unique (local) rule. For example, as a characteristic of an organization, not only in Japan, but actually, the relationship between the "organization name" and the "position name" that governs the organization is not set in such a way that it can be easily guessed. For example, it is easy to infer the relationship between the organization name “A Section” and the position name “Section Manager”. "Minister" is difficult to presume related. On the other hand, since these names are a fixed culture within the organization (company), a comparison table can be prepared for each company (or department). By storing a comparison table showing such relationships in a dictionary as a role table, it is possible to adapt to various corporate cultures (systems).

Furthermore, there are cases where the role name is converted into a unique name and uttered, or abbreviated and uttered. For example, the role "manager" may be abbreviated as "manager". Therefore, the dictionary may include a table (hereinafter referred to as a conversion table) in which converted (for example, omitted) role names and target words are associated with each other. In this way, by registering convertible information for an input colloquial sentence in the dictionary, name resolution can be performed by simple comparison.

Organization information and affiliation information can generally be acquired from a personnel system (not shown) owned by the organization. Therefore, the organization information and affiliation information included in the dictionary may be generated from the personnel system. In addition, personnel systems may have the ability to manage personal schedules. Therefore, the dictionary may include a personal schedule.

FIG. 3 is an explanatory diagram showing an example of a dictionary stored in the storage unit 70. As shown in FIG. The dictionary illustrated in FIG. 3 includes a personal information table T1 (also referred to as a list) holding personal information, an organization/position table T2 corresponding to the role table, and a schedule table T3.

For example, the personal information table T1 illustrated in FIG. 3 indicates that an ID and a personal name, which are personal identification information, and affiliation information, which is additional information, are associated with each other. Also, the organization/position table T2 illustrated in FIG. 3 is a table in which positions in the organization are associated with each other. For example, the example shown in FIG. 3 indicates that the "group" in the organization has positions called "general manager" and "deputy manager". Further, the schedule table T3 associates individuals included in the personal information table T1 with the times when the individuals are enrolled.

In the example shown in FIG. 3, the personal information table is associated with one piece of affiliation information, but a plurality of pieces of affiliation information may be associated with the personal information table. For example, when new and old affiliation information is required due to organizational restructuring or organization name change, the personal information table T1 may hold both old and new affiliation information. Also, in an organization, apart from the actual affiliation, another organization such as a committee may be established. In the case of such a committee, for example, a role such as "XX committee member" can be considered as a position. In this way, affiliation information indicating a so-called virtual organization may be separately associated with the personal information table. By associating such affiliation information, it becomes possible to perform a search in the same manner as for personal affiliation.

Also, the role table may include the extent to which the target word indicates the role-likeness in addition to the organization and role name. Hereinafter, this degree is referred to as reliability. The reliability can be said to be the likelihood that the role will be chosen. Similarly, the affiliation information of personal information may also be given a degree of reliability indicating the role-likeness of the individual. In the initial state, for example, the same reliability may be set for all items.

The target word extraction unit 40 extracts words representing roles in an organization (hereinafter referred to as target words) from the character strings converted into text by the speech recognition unit 30 . For example, if there is a position called "section manager" in the organization, the target word extraction unit 40 may extract the word "section manager" from the text string. Moreover, the target word extraction unit 40 may extract not only the target word but also personal identification information (ID and name). This is because, if individual identification information can be directly extracted, the accuracy of identifying the target word can be improved.

The target word extraction unit 40 may, for example, divide the character string converted into text into words and phrases by morphological analysis, and extract the target word by estimating the part of speech. It should be noted that many methods of morphological analysis have been commercialized, and the method is not limited here. Individual identification (name resolution) is performed using the target words extracted here.

Further, when the storage unit 70 holds the above-described role table, the role table includes role names. Target words included in the role table may be extracted from the character string.

Furthermore, when the storage unit 70 holds the conversion table described above, since the conversion table includes words indicating omitted role names, the target word extraction unit 40 refers to the conversion table. and extract the omitted role name from the character string, and convert the extracted role name into the target word.

Based on personal information (that is, information in which personal identification information and affiliation information are associated), the individual identification unit 50 identifies individual candidates indicated by the extracted target words in relation to the organization to which the user belongs. Identify accordingly. Specifically, the individual identifying unit 50 identifies, as candidates, individuals having roles indicated by the target words, among individuals who partially or wholly match the structure of the organization to which the user belongs.

The individual identification unit 50 may, for example, perform name resolution of target words extracted by morphological analysis or the like by matching with a dictionary (role table and personal information) to narrow down individual candidates. For example, when the affiliation information is represented by the above-described organization path, the individual identification unit 50 determines whether part or all of the structure matches. You can judge whether or not In addition, the individual identification unit 50 may identify individual candidates by individual names or identification information such as IDs.

The processing performed by the personal identification unit 50 of this embodiment will be described below using a specific example. FIG. 4 is an explanatory diagram showing an example of a relationship between an organization and persons belonging to the organization. FIG. 5 is an explanatory diagram showing an example of information included in the dictionary. In the example shown in FIG. 5, the dictionary includes a personal information table T4 (name list) and an organization/post table T5.

It is assumed that the user AAA is specified as the user identification information prior to the processing performed by the individual specifying unit 50 . In the example shown in FIG. 4, user AAA belongs to X group of A section.

First, a user AAA who wants to communicate vocalizes his/her desire to make a voice recording, and the voice data generated from the recorded voice is converted into text as it is. Then, it is assumed that morphological analysis is performed on the speech data converted into text, and nouns, which are referents, are extracted. Here, it is assumed that the user AAA utters "section manager" to call the section manager.

The individual identification unit 50 searches whether the extracted reference word exists in a comparison table (role table) of positions included in a predefined dictionary. By searching the organization/position table T5 illustrated in FIG. 5, it is found that "section manager" is the position of the organization "section". Therefore, the individual identification unit 50 identifies the user AAA's affiliation information "Cop Co., Ltd./D department/Section A/Group X" to the organization "Section" (that is, "Cop Co., Ltd./D department/Section A"). Individuals having common affiliation information are extracted from the personal information table T4. In the example shown in FIG. 5, users with user IDs from 0001 to 0007 (that is, users DDD, users CCC, users BBB, and users EEE) are extracted.

Then, the individual identifying unit 50 identifies, as candidates, individuals whose role is the extracted referent from among the individuals belonging to the extracted organization. In the example shown in FIG. 5, the role of user DDD in the organization "Cop Co., Ltd./Dept./A Section" is "Section Manager", so the individual identifying unit 50 identifies user DDD as a candidate.

That is, although the organization illustrated in FIG. 4 also includes a section manager (user FFF) of section B, the individual identification unit 50 identifies section manager DDD belonging to "section A" as a candidate.

Although it is desirable that the extracted candidates can be uniquely narrowed down, it is also assumed that a plurality of candidates may be listed. Therefore, the individual identification unit 50 may determine the reliability for each candidate. For example, if the role table includes trust, the individual identifier 50 may determine the trust of the identified candidate from the role table based on the candidate's position in the organization. Further, for example, when the affiliation information of the personal information includes the reliability, the individual identification unit 50 may determine the reliability from the affiliation information of the identified individual.

Furthermore, when personal information includes a plurality of pieces of affiliation information (for example, old and new affiliation information) and a reliability level is set for each piece of affiliation information, the individual identification unit 50 determines which of the plurality of pieces of affiliation information is identified. Reliability may be determined from personal affiliation information.

Also, the individual identification unit 50 may calculate the reliability based on an individual's schedule. FIG. 6 is an explanatory diagram showing an example of a schedule in an organization. In the example shown in FIG. 6, the working hours of the X group are from 9:00 to 16:00, and the working hours of the Y group are from 17:00 to 24:00. In such a schedule, there is a high possibility that the person with whom user AAA wants to communicate at 12:00 is a member of team X (user CCC, user BBB). Therefore, the individual identifying unit 50 may determine the reliability of individuals present during working hours (time spent in the organization) to be higher than the reliability of individuals outside of working hours.

It should be noted that multiple confidence levels may be determined when identifying individual candidates. In this case, the individual identification unit 50 may perform arithmetic processing (for example, addition, multiplication, etc.) on the determined reliability levels to determine the final reliability level.

The output unit 60 outputs the identified individual candidates. The output unit 60 may output one or all of the individual candidates. For example, when the individual identifying unit 50 calculates the reliability of individual candidates, the output unit 60 may output individual candidates based on the identified reliability. The output unit 60 may output, for example, one person with the highest reliability, or may output a predetermined number of persons (or all) in descending order of reliability.

Also, the output unit 60 may convert the specified candidate into an ID based on a dictionary. Such IDs can be used for various applications on the system (for example, communications such as telephone calls).

Alternatively, the output unit 60 may output only the identification information of the candidate with the highest reliability. By outputting only such identification information, another application that uses this identification information can uniquely identify the object to be processed. This is because there is a possibility that some recognition errors can be tolerated within the organization assumed in this embodiment.

FIG. 7 is an explanatory diagram showing a specific configuration example of the speech recognition device 100 of this embodiment. A voice of the user 101 is converted into an electric signal by the microphone 102 . This microphone 102 corresponds to the input unit 10 in FIG.

Further, the processing unit 103 and the processing unit 104 illustrated in FIG. 7 perform processing performed by the speech data generation unit 20, the speech recognition unit 30, the target word extraction unit 40, the individual identification unit 50, and the output unit 60 illustrated in FIG. do some or all Also, the application 105 executes actual operations based on the results output by the processing unit 103 or the processing unit 104 .

In the example shown in FIG. 7, it is assumed that the processing unit 103 performs processing locally, and the processing unit 104 performs processing online. The processing performed by the processing units 103 and 104 is determined according to the environment in which the processing by the speech recognition apparatus 100 of this embodiment is performed.

In general, it is desirable that processing (recording processing) for performing AD conversion of audio data converted into electric signals be locally arranged. Processing that requires a large number of operations, such as speech recognition processing, is generally performed in an online environment where resources can be easily secured. However, for example, in an environment where the analog audio signal itself can be transmitted online as it is, the recording process may be performed online. On the other hand, if there is an environment in which speech recognition processing can be implemented locally, speech recognition processing may be performed locally.

The speech data generation unit 20, the speech recognition unit 30, the target word extraction unit 40, the individual identification unit 50, and the output unit 60 are implemented by a computer processor (e.g., CPU (Central Processing Unit), GPU (Graphics
Processing Unit)). For example, the program is stored in the storage unit 70, the processor reads the program, and according to the program, the voice data generation unit 20, the voice recognition unit 30, the target word extraction unit 40, the individual identification unit 50, and the output unit 60 may operate as Also, the functions of the speech recognition device 100 may be provided in a SaaS (Software as a Service) format.

Further, the speech data generation unit 20, the speech recognition unit 30, the target word extraction unit 40, the individual identification unit 50, and the output unit 60 may each be realized by dedicated hardware. Also, part or all of each component of each device may be implemented by general-purpose or dedicated circuitry, processors, etc., or combinations thereof. These may be composed of a single chip, or may be composed of multiple chips connected via a bus. A part or all of each component of each device may be implemented by a combination of the above-described circuits and the like and programs.

Further, when part or all of each component of the speech recognition device 100 is realized by a plurality of information processing devices, circuits, etc., the plurality of information processing devices, circuits, etc. may be centrally arranged, They may be distributed. For example, the information processing device, circuits, and the like may be implemented as a form in which each is connected via a communication network, such as a client-server system, a cloud computing system, or the like.

Also, the storage unit 70 is realized by, for example, a magnetic disk or the like.

Next, an operation example of this embodiment will be described. FIG. 8 is a flow chart showing an operation example of the speech recognition device 100 of this embodiment. The target word extracting unit 40 extracts a target word from the character string converted into text based on the user's voice (step S11). Based on the personal information, the individual identification unit 50 identifies individual candidates indicated by the extracted target words according to the relationship with the organization to which the user belongs (step S12). The output unit 60 outputs the identified individual candidates (step S13).

Next, specific processing using the speech recognition apparatus 100 of this embodiment will be described. FIG. 9 is an explanatory diagram showing a specific example of processing using the speech recognition apparatus 100 of this embodiment.

First, the input unit 10 records voice, and the voice data generation unit 20 generates voice data from the recorded voice (step S101). Next, the speech recognition unit 30 performs speech recognition from the speech data using, for example, information for speaker recognition, and outputs simple sentences (character strings) and speaker information (step S102). Then, the speech recognition unit 30 performs morphological analysis and outputs the noun text phrase and speaker information (step S103). The target word extraction unit 40, for example, refers to the name list (user list) 171, job titles 172, schedule 173, etc. included in the dictionary 170, and outputs partner candidates and reliability (step S104). The individual identification unit 50 outputs individual identification information (ID) based on the output candidates and reliability.

The output ID is used in each application (step S106). In step S103, if a verb or a phrase representing an action (function) is extracted, each application may perform processing according to the verb. At that time, each application may perform similar reading replacement (conversion) processing for verbs and other words. This makes it possible to mechanically absorb and perform appropriate actions even when different expressions (for example, accent) are used for the same behavior depending on the individual's experience and environment (situation and culture). . This is also applicable to languages other than Japanese.

Note that machine learning may be used to generate and update such dictionaries. For example, in the initial state of learning, static information as described above is considered necessary, but machine learning is expected to improve the accuracy of the dictionary. Also, although the present embodiment can be implemented by a computer system, part of the processing may be performed manually. For example, verbs (actions) are extracted based on the minutes document that has already been created, and by presenting information that has been narrowed down to candidates, the final decision and execution are left to the user. A configuration can also be envisioned.

Next, a modified example of the speech recognition device 100 of this embodiment will be described. In the above embodiment, a case has been described in which a target word is extracted from a character string converted into text based on voice, and individual candidates are specified based on the target word. In this modified example, a method for identifying individuals having the same identification information (for example, same surname and same name) will be described.

For example, in the organization illustrated in FIG. 4, there are two individuals whose personal names are "DDD". One is the manager of Section A, and the other is "DDD" of Section Y of Section A. In this situation, assume that user AAA speaks the personal name "DDD". In this case, the individual targeted by user AAA is more likely to be 'DDD' of Y group in A section than 'DDD' who is the manager of A section. This is because Japanese people tend to call others less by their names (that is, they designate them by their job titles) if they have a higher position. Therefore, the individual identification unit 50 may identify individual candidates by expanding the search range step by step to include groups (for example, groups) other than the group to which the individual belongs.

As described above, in the present embodiment, the target word extraction unit 40 extracts the target word from the character string converted into text based on the user's voice, and the individual identification unit 50 extracts the target word based on the personal information. Individual candidates indicated by the given target words are identified according to the relationship with the organization to which the user belongs. Then, the output unit 60 outputs the identified individual candidates. Therefore, when an individual existing in an organization is identified by voice, the identification accuracy can be improved with simple settings and logic.

That is, in the present embodiment, attention is focused on Japanese habits, and the uncertainty of the user's instructions is compensated for by combining databases. For example, in recent years, in order to solve such a problem, it is also performed to estimate the relationship by machine learning from the usual speech and behavior of individual users. While machine learning requires a large amount of calculation, the configuration of the present embodiment has high versatility and uses static data that can be easily automatically generated, such as organization configuration information. Therefore, the user's convenience can be further improved with a smaller amount of calculation. In addition, since the configuration of the present embodiment does not require learning, for example, even if there is a change in the system at the workplace or the like, the change can be reflected immediately.

In addition, in this embodiment, the machine automatically grasps the meaning of utterances with expressions similar to those used for communication between humans and operates. Therefore, by using the speech recognition apparatus of this embodiment, it is possible to provide highly adaptable applications. Also, the technique of limiting the target based on the meaning of the statement can be used, for example, for automatically complementing the referent indicated by the word in speech. In addition to human-to-human interaction, machine operation can also be realized by using a similar method for human-to-machine interaction.

Next, an outline of the present invention will be described. FIG. 10 is a block diagram showing an outline of a speech recognition device according to the invention. The speech recognition device 80 according to the present invention is a speech recognition device (for example, the speech recognition device 100) that identifies an individual based on voice, and uses a character string converted into text based on the user's voice to determine the role in the organization. A target word extraction unit 81 (for example, the target word extraction unit 40) that extracts a target word that is a word representing (for example, a position), an individual identification information (for example, name, ID), an organization to which the individual belongs Individuals who identify individual candidates indicated by the extracted target words according to their relationship with the organization to which the user belongs It includes an identification unit 82 (for example, the individual identification unit 50) and an output unit 83 (for example, the output unit 60) that outputs candidates for the identified individual.

With such a configuration, when an individual existing in an organization is identified by voice, the identification accuracy can be improved with simple settings and logic.

Specifically, the individual identification unit 82 may identify, as a candidate, an individual having a role indicated by the target word, among individuals who partially or wholly match the structure of the organization to which the user belongs.

Further, based on the role of the identified candidate in the organization, the individual identification unit 82 selects the reliability of the identified candidate from a reliability table in which the role in the organization is associated with the reliability indicating the likelihood of the role of the target word. and the output unit 83 may output individual candidates based on the identified confidence level.

Specifically, the personal information may include a plurality of pieces of affiliation information, and a degree of reliability may be assigned to each piece of affiliation information. Then, the individual identification unit 82 may determine the reliability from the affiliation information of the identified individual among the plurality of affiliation information.

At that time, the output unit 83 may output the identification information of the individual candidate with the highest reliability.

In addition, the target word extracting unit 81 refers to a conversion table that associates the omitted role names with target words, extracts the omitted role names from the character strings converted into text, and extracts the extracted role names. may be converted to the corresponding target term.

INDUSTRIAL APPLICABILITY The present invention is preferably applied to a voice recognition device that recognizes voice and identifies an individual. For example, assuming voice communication first, the present invention can be suitably applied to a system that operates based on vague verbal instructions from a person.

REFERENCE SIGNS LIST 10 input unit 20 speech data generation unit 30 speech recognition unit 40 target word extraction unit 50 individual identification unit 60 output unit 70 storage unit 100 speech recognition device

Claims (10)

A speech recognition device for identifying individuals based on speech,
a target word extraction unit for extracting target words, which are words representing roles in an organization, from character strings converted into text based on user's voice;
Based on the personal information that associates the identification information of the individual with the affiliation information that is information representing the organization to which the individual belongs and the role of the individual in the organization, the candidate for the individual indicated by the extracted target word an individual identification part that identifies according to the relationship with the organization to which the user belongs;
an output unit that outputs the identified individual candidate,
The individual identification unit determines the reliability of the individual candidate during working hours to be higher than the reliability of the individual candidate who is not during the working hours, and there are a plurality of individual candidates having the same identification information. In this case, preferentially select candidates for individuals other than those in higher positions than candidates for individuals in higher positions than the user in the organization to which the user belongs,
The speech recognition device, wherein the output unit outputs individual candidates based on the reliability. 2. The speech recognition apparatus according to claim 1, wherein the individual identifying unit identifies, as a candidate, an individual who has a role indicated by the target word from among individuals who partially or wholly match the structure of the organization to which the user belongs. Based on the role of the identified candidate in the organization, the individual identification unit determines the reliability of the identified candidate from a reliability table that associates the role in the organization with the reliability indicating that the target word is likely to play the role. decide and
3. The speech recognition apparatus according to claim 1, wherein the output unit outputs individual candidates based on the specified reliability. Personal information includes multiple pieces of affiliation information, and reliability is assigned to each piece of affiliation information.
4. The speech recognition apparatus according to claim 3, wherein the individual identification unit determines the reliability from the affiliation information of the identified individual among the plurality of affiliation information. 5. The speech recognition device according to any one of claims 1 to 4, wherein the output unit outputs the identification information of the individual candidate with the highest reliability. The target word extraction unit refers to a conversion table that associates abbreviated role names with target words, extracts the abbreviated role names from the textual character string, and associates the extracted role names with each other. 6. The speech recognition device according to any one of claims 1 to 5, wherein the target word is converted into a target word. A personal identification method for identifying an individual based on voice, comprising:
Extracting target words, which are words representing roles in an organization, from character strings converted into text based on the user's voice,
Based on the personal information that associates the identification information of the individual with the affiliation information that is information representing the organization to which the individual belongs and the role of the individual in the organization, the candidate for the individual indicated by the extracted target word Identify according to the relationship with the organization to which the user belongs,
The reliability of candidates for individuals during working hours is determined to be higher than the reliability for candidates for individuals who are not during working hours, and when there are multiple candidates for individuals with the same identification information, the said user preferentially select candidates for individuals other than those in higher positions than candidates in higher positions than the user in the organization to which the user belongs,
An individual identification method, characterized by outputting an identified individual candidate based on the reliability. 8. The personal identification method according to claim 7, wherein, among individuals who partially or wholly match the structure of the organization to which the user belongs, individuals having roles indicated by the target words are specified as candidates. A personal identification program applied to a computer that identifies a person based on sound, comprising:
to the computer;
Target word extraction processing for extracting target words, which are words representing roles in an organization, from character strings converted into text based on the user's voice,
Based on the personal information that associates the identification information of the individual with the affiliation information that is information representing the organization to which the individual belongs and the role of the individual in the organization, the candidate for the individual indicated by the extracted target word Individual identification processing that identifies according to the relationship with the organization to which the user belongs, and
Execute output processing for outputting the identified individual candidates,
In the individual identification process, the reliability of the candidate for the individual during working hours is determined to be higher than the reliability for the candidate for the individual who is not during the working hours, and there are multiple candidates for the individual having the same identification information. In this case, preferentially select candidates for individuals other than those in higher positions than candidates for individuals in higher positions than the user in the organization to which the user belongs,
An individual identification program for outputting individual candidates based on the reliability in the output process. to the computer,
10. The personal identification program according to claim 9, wherein in the personal identification processing, among individuals who partially or wholly match the structure of the organization to which the user belongs, the individual having the role indicated by the target word is identified as a candidate.

Images