JP2020170310A - Conversation analysis device, conversation analysis method and conversation analysis program - Google Patents

Abstract

To perform risk determination as an object of a conference based on an utterance of the conference.SOLUTION: A conversation analysis device 100 includes a first determination unit 110, a voice recognition unit 120, a second determination unit 130, and an evaluation unit 140. The first determination unit 110 analyzes a conversation voice, and determines a state of the conversation. The voice recognition unit 120 performs voice recognition of the conversation voice, and extracts words included in the conversation voice. The second determination unit 130 determines the abstraction degree of conversation contents in the conversation voice on the basis of the words included in the conversation voice. The evaluation unit 140 evaluates the conversation contents on the basis of the state and the abstraction degree of the conversation.SELECTED DRAWING: Figure 2

Description

The present invention relates to a conversation analyzer and the like.

It is required to evaluate the "conversation target (conversation content)" in the conversation between the operator and the customer in the call center, or in the conversation between the participants of the conference. For example, the target of conversation between the operator and the customer is inquiries about products, complaint handling, and the like. The target of conversation by the participants of the conference is the project.

As a technique for evaluating a project based on voice data recorded from a conversation during a meeting, there is a conventional technique as described below. This prior art predefines a set word and a similar word similar to the set word. FIG. 18 is a diagram showing an example of a setting word and a similar word. For example, a similar word corresponding to the set word "delayed" is "delayed, slow, ...". In the following description, the setting word and similar words similar to the setting word are collectively referred to as "setting word".

In the prior art, voice recognition is performed on voice data, and the number of appearances of a set word included in the voice data is counted. In the prior art, the risk evaluation value is specified and the conversation target is evaluated according to the number of occurrences of the set word. For example, in the prior art, the higher the number of occurrences of the set word, the larger the risk evaluation value, and the larger the risk evaluation value, the higher the risk of the conversation target (project). Further, in this conventional technique, the ratio of the maximum utterance time to the meeting time is calculated, and the larger the ratio, the larger the risk evaluation value.

JP-A-2009-146121 JP-A-2017-27102 Japanese Unexamined Patent Publication No. 2011-55160 Japanese Unexamined Patent Publication No. 2018-36868

In the conversation of the meeting, the risk of being told that the requirement specifications were "delayed" or "not completed" and the decision of the output message was "delayed" or "not completed" were spoken. The risk of the case is not the same risk. The work related to "organizing the required specifications" for the entire project takes days to weeks, and if the required specifications are not organized, it is a big risk. On the other hand, the work related to the "output message" can be solved in one hour for the entire project, and even if the output message is sent, the risk is small.

However, in the prior art, the number of times that "delayed" or "not completed" was spoken regarding the arrangement of required specifications, and the number of times that "delayed" or "not completed" was spoken regarding the output message. If the numbers are the same, each risk is evaluated as the same risk. Therefore, in the prior art, even if the presence or absence of risk can be evaluated, the degree of risk as a conversation target cannot be appropriately evaluated.

In one aspect, it is an object of the present invention to provide a conversation analyzer, a conversation analysis method, and a conversation analysis program capable of making a risk determination as a subject of a conference based on the utterance of the conference.

In the first plan, the conversation analyzer has a first determination unit, a voice recognition unit, a second determination unit, and an evaluation unit. The first determination unit analyzes the conversation voice and determines the state of the conversation. The voice recognition unit performs voice recognition on the conversational voice and extracts words included in the conversational voice. The second determination unit determines the degree of abstraction of the conversation content in the conversation voice based on the words included in the conversation voice. The evaluation unit evaluates the conversation target based on the state of conversation and the degree of abstraction.

It is possible to make a risk judgment as a target of a meeting based on the utterance of the meeting.

FIG. 1 is a diagram for explaining the processing of the conversation analyzer according to the first embodiment. FIG. 2 is a functional block diagram showing the configuration of the conversation analyzer according to the first embodiment. FIG. 3 is a diagram showing an example of the data structure of the risk evaluation value table according to the first embodiment. FIG. 4 is a diagram showing an example of the data structure of the abstraction degree determination table according to the first embodiment. FIG. 5 is a diagram showing an example of the concept DB. FIG. 6 is a diagram for explaining the processing of the second determination unit according to the first embodiment. FIG. 7 is a flowchart showing a processing procedure of the conversation analyzer according to the first embodiment. FIG. 8 is a diagram for supplementing the effect of the invention. FIG. 9 is a functional block diagram showing the configuration of the conversation analyzer according to the second embodiment. FIG. 10 is a diagram showing an example of the data structure of the setting keyword table according to the second embodiment. FIG. 11 is a diagram showing an example of the data structure of the risk evaluation value table according to the second embodiment. FIG. 12 is a diagram for explaining the processing of the generation unit according to the second embodiment. FIG. 13 is a flowchart showing a processing procedure of the conversation analyzer according to the second embodiment. FIG. 14 is a functional block diagram showing an example of the configuration of the conversation analyzer according to the third embodiment. FIG. 15 is a diagram for explaining the processing of the extraction unit according to the third embodiment. FIG. 16 is a flowchart showing a processing procedure of the conversation analyzer according to the third embodiment. FIG. 17 is a diagram showing an example of a computer hardware configuration that realizes the same functions as the conversation analyzer according to the present embodiment. FIG. 18 is a diagram showing an example of a setting word and a similar word.

Hereinafter, examples of the conversation analyzer, the conversation analysis method, and the conversation analysis program disclosed in the present application will be described in detail with reference to the drawings. The present invention is not limited to this embodiment.

FIG. 1 is a diagram for explaining the processing of the conversation analyzer according to the first embodiment. First, the conversation analyzer will explain the case of evaluating the risk of the conversation target of the conference A based on the table 10A. Table 10A is a table showing the relationship between the words uttered in the conference A and the number of times.

The conversation analyzer identifies the word corresponding to the set keyword from each word in the table 10A, and determines whether the "meeting state" is good or bad based on the number of occurrences of the word corresponding to the specified set keyword. .. For example, the conversation analyzer determines that the state of the conference is "bad" when the number of occurrences of the word corresponding to the set keyword is equal to or more than the threshold value, and the state of the conference is "good" when the number of occurrences is less than the threshold value. Is determined.

For example, the setting keywords are "delayed" and "not completed", and the threshold value of the number of occurrences is "10". Then, the conversation analyzer determines that the state of the conference A is bad because the number of occurrences "16" of the word corresponding to the set keyword in the table 10A becomes the threshold value "10" or more.

Subsequently, the conversation analyzer determines the "abstraction level of the conversation content" based on the abstraction level of each word in the table 10A. The degree of abstraction of each word is determined by the frequency of occurrence in the language corpus, and the higher the frequency of occurrence, the higher the degree of abstraction.

For example, in Table 10A, many "specifications" with a high degree of abstraction appear. Therefore, the conversation analyzer determines that the conversation content of the conference A has a high degree of abstraction. "Highly abstract conversation content" means that no specific points have been discussed at the meeting, only abstract discussions have been made.

The conversation analyzer evaluates the risk of the conversation target of the conversation A based on the judgment result of the conversation state and the judgment result of the abstraction degree of the conversation content. As described above, since the state of conversation is poor and the degree of abstraction of the conversation content is high, it can be said that it is difficult to solve the conversation target in the conference A because it is required to match the consciousness. Therefore, the conversation analyzer evaluates that the risk of the conversation target of the conference A is “high”.

Next, the case where the conversation analyzer evaluates the risk of the conversation target of the conference B based on the table 10B will be described. Table 10B is a table showing the relationship between the words uttered in the conference B and the number of times.

The conversation analyzer identifies the word corresponding to the set keyword from each word in the table 10B, and determines the quality of the "meeting state" based on the number of occurrences of the word corresponding to the specified set keyword. .. For example, the conversation analyzer determines that the state of the conference B is bad because the number of occurrences "16" of the word corresponding to the set keyword in the table 10B is equal to or more than the threshold value "10".

Subsequently, the conversation analyzer determines the "abstraction level of the conversation content" based on the abstraction level of each word in the table 10B. For example, in Table 10B, many "character codes" having a low degree of abstraction appear. Therefore, the conversation analyzer determines that the degree of abstraction of the conversation content of the conference B is low. "Low abstraction of conversation content" means that specific points are being discussed at the meeting.

The conversation analyzer evaluates the risk of the conversation target of the conversation B based on the judgment result of the conversation state and the judgment result of the abstraction degree of the conversation content. As described above, although the state of conversation is poor, the degree of abstraction of the conversation content is low, so it can be said that the points of discussion are clear, easy to communicate, and easy to solve with respect to the conversation target in the conference B. Therefore, the conversation analyzer evaluates that the risk of the conversation target of the conference B is “small”.

As described above, the conversation analyzer according to the first embodiment evaluates the risk of the conversation target by using the abstraction level of the meeting content in addition to the good or bad state of the meeting based on the number of occurrences of the set keyword. There is. For example, it may be judged that the risk is high in both conference A and conference B only by the number of occurrences of the set keyword. However, considering the degree of abstraction, conference B has a low degree of abstraction, so a concrete discussion It can be judged that the risk is small. That is, the degree of risk as a conversation target can be appropriately evaluated.

Next, an example of the configuration of the conversation analyzer according to the first embodiment will be described. FIG. 2 is a functional block diagram showing the configuration of the conversation analyzer according to the first embodiment. As shown in FIG. 2, the conversation analyzer 100 includes a first determination unit 110, a voice recognition unit 120, a second determination unit 130, and an evaluation unit 140.

The storage unit 105 has a risk evaluation value table 105a and an abstraction degree determination table 105b. The storage unit 105 corresponds to semiconductor memory elements such as RAM (Random Access Memory), ROM (Read Only Memory), and flash memory (Flash Memory), and storage devices such as HDD (Hard Disk Drive).

FIG. 3 is a diagram showing an example of the data structure of the risk evaluation value table according to the first embodiment. As shown in FIG. 3, the risk evaluation value table 105a associates the ratio with the risk evaluation value. The ratio is calculated by the first determination unit 110, which will be described later. The risk evaluation value is a value indicating the degree of poor state of the meeting, and the larger the risk evaluation value, the worse the state of the meeting.

FIG. 4 is a diagram showing an example of the data structure of the abstraction degree determination table according to the first embodiment. As shown in FIG. 4, the abstraction degree determination table 105b associates a word with an abstraction degree. The degree of abstraction is a value indicating the degree of abstraction of a word, and the higher the degree of abstraction, the more abstract the word is. For example, the degree of abstraction of a word is determined based on a concept DB (Data Base).

FIG. 5 is a diagram showing an example of the concept DB. In the concept DB 50, each word is defined by a concept tree structure, and the higher the concept hierarchy, the higher the degree of abstraction. The concept hierarchy "1" is the highest hierarchy, and the hierarchy becomes lower in the order of concept hierarchy 2, 3, 4, ..., 9.

In FIG. 5, the conceptual hierarchy of the words “event” and “action” is “1”, and the degree of abstraction is “9”. The concept hierarchy of the word "requirement specification" is "5", and the degree of abstraction is "5". The conceptual hierarchy of the word "member" is "6" and the abstraction level is "4". The conceptual hierarchy of the word "ID" is "7" and the abstraction level is "3". The hierarchy of the word "universal design" is "9" and the abstraction level is "1".

For example, since the word "requirement specification" hits the "requirement specification" of the concept DB 50, the abstraction degree of the word "requirement specification" is "5". Since the word "universal design" hits the concept DB50 "universal design", the degree of abstraction of the word "universal design" is "1".

When the word is a compound word, the word having the lowest conceptual hierarchy among the plurality of words included in the compound word is specified, and the conceptual hierarchy obtained by adding 1 to the conceptual hierarchy of the specified word is the compound word. It is a conceptual hierarchy. For example, the word (compound word) "member ID" hits "member" and "ID" in the concept DB 50. Of the concept hierarchy "6" of "member" and the concept hierarchy "7" of "ID" in the concept DB, the concept hierarchy "8" obtained by adding 1 to the lower concept hierarchy "7" is the word "member ID". The concept hierarchy of the above is specified, and the abstraction degree "2" of the specified concept hierarchy "8" is specified as the abstraction degree of the word "member ID".

Returning to the description of FIG. Each of the processing units 110, 120, 130, 140 in FIG. 2 can be realized by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. Further, each processing unit 110-140 can also be realized by hard-wired logic such as ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

The conversation analyzer 100 acquires the conversation voice data 20 from an external device (not shown). The conversation voice data 20 is voice data recorded from a conference conversation, for example, associating time with voice power.

The first determination unit 110 is a processing unit that analyzes the conversation voice data 20 and determines the state of the conversation. For example, the first determination unit 110 scans the voice power of the conversation voice data 20 and specifies the sections in which the voice power is equal to or higher than the threshold value Thp as the utterance sections.

The first determination unit 110 calculates the time from the start time of the first utterance section to the end time of the last utterance section as the "meeting time" among the plurality of utterance sections specified from the conversation voice data 20.

The first determination unit 110 compares the lengths of the plurality of utterance sections and identifies the longest utterance section. The first determination unit 110 calculates the ratio of the longest utterance section to the conference time based on the equation (1).

Ratio = longest utterance section / meeting time x 100 (%) ... (1)

The first determination unit 110 determines the risk evaluation value by comparing the ratio calculated based on the equation (1) with the risk evaluation value table 105a. The risk assessment value is an example of the state of conversation. The first determination unit 110 outputs the determination result of the risk evaluation value to the evaluation unit 140.

The first determination unit 110 described in the first embodiment has described the case where the risk evaluation value is determined based on the equation (1), but as described in FIG. 1, the number of occurrences of the set keyword has been described. The state of the conversation may be determined based on. For example, the first determination unit 110 calculates the ratio of the number of appearances of the set keyword to the total number of appearances of all the words included in the voice recognition result of the voice recognition unit 120, and the calculated ratio and the risk evaluation value table 105a The risk evaluation value may be determined based on.

The voice recognition unit 120 is a processing unit that analyzes the conversation voice data 20 and extracts the spoken word. The voice recognition unit 120 outputs the information of each extracted word to the second determination unit 130. The voice recognition unit 120 may use any voice recognition technology. For example, the spoken word is converted into a character string based on the characteristics of the voice included in the conversation voice data 20, and the word is extracted based on the dictionary information that defines the word.

The second determination unit 130 is a processing unit that determines the degree of abstraction of the conversation content based on the voice recognition result of the voice recognition unit 120 and the abstraction degree determination table 105b. The second determination unit 130 outputs information on the degree of abstraction of the determined conversation content to the evaluation unit 140.

For example, the second determination unit 130 compares a plurality of words included in the speech recognition result with the abstraction degree determination table 105b, and determines the number of occurrences of each word for each word registered in the abstraction degree determination table 105b. Count.

FIG. 6 is a diagram for explaining the processing of the second determination unit according to the first embodiment. In the example shown in FIG. 6, the number of appearances of the word "requirement specification" is "6", the number of appearances of the word "member ID" is "1", and the number of appearances of the word "output message" is "1". Since the abstraction level of the word "requirement specification" is "10", the abstraction level of the word "member ID" is "2", and the abstraction level of the word "output message" is "4", the total abstraction level is "66". It becomes. The average degree of abstraction of each word is "8.25". The second determination unit 130 outputs the average of the abstraction degree to the evaluation unit 140 as information on the abstraction degree of the determination result.

The evaluation unit 140 is a conversation target included in the conversation voice data 20 based on the conversation state (risk evaluation value) output from the first determination unit 110 and the abstraction degree output from the second determination unit 130. It is a processing department that evaluates the risk of.

For example, the evaluation unit 140 evaluates that the risk of the conversation target is "high risk" when the value obtained by multiplying the risk evaluation value and the degree of abstraction is equal to or greater than the reference evaluation value. On the other hand, when the value obtained by multiplying the risk evaluation value by the degree of abstraction is less than the standard evaluation value, the evaluation unit 140 evaluates that the risk of the conversation target is "small risk".

The evaluation unit 140 may output the evaluation result to a display device (not shown) and display the evaluation result, or may notify the evaluation result to an external device (not shown) via the network.

Next, an example of the processing procedure of the conversation analyzer 100 according to the first embodiment will be described. FIG. 7 is a flowchart showing a processing procedure of the conversation analyzer according to the first embodiment. As shown in FIG. 7, the conversation analyzer 100 acquires the conversation voice data 20 (step S101).

The first determination unit 110 of the conversation analyzer 100 determines the conversation state (risk evaluation value) based on the ratio of the longest utterance time to the conference time and the risk evaluation value table 105a (step S102).

The voice recognition unit 120 of the conversation analyzer 100 executes voice recognition on the conversation voice data 20 and extracts the spoken word (step S103). The second determination unit 130 of the conversation analyzer 100 determines the degree of abstraction of the conversation content based on the uttered word and the abstraction degree determination table 105b (step S104).

The evaluation unit 140 of the conversation analyzer 100 comprehensively evaluates the risk of the conversation target based on the conversation state (risk evaluation value) and the degree of abstraction of the conversation content (step S105). The evaluation unit 140 notifies the evaluation result (step S106).

Next, the effect of the conversation analyzer 100 according to the first embodiment will be described. The conversation analyzer 100 evaluates the risk of the conversation target by using the abstraction level of the conference content in addition to the state of the conference determined by the first determination unit 110. This makes it possible to appropriately evaluate the degree of risk as a conversation target.

Here, as a supplement to the effect of the invention, the relationship between the magnitude of risk and the frequency of occurrence of words will be described. FIG. 8 is a diagram for supplementing the effect of the invention. For example, the word "stop processing" appears in various places in the related documents 30a, 30b, and 30c, and it can be said that the frequency of appearance is high and the degree of abstraction is high. The higher the level of abstraction, the wider the range of points to be improved and the greater the risk. If the degree of abstraction is high, there is a risk of misunderstanding "what kind of cancellation process is the cancellation process" in terms of concreteness.

It can be said that the word "character code" appears only in a specific related document 30c among the related documents 30a, 30b, and 30c, and the frequency of appearance is low and the degree of abstraction is low. If the level of abstraction is low, the range of points to be improved is limited, and the risk is considered to be small.

As explained in FIG. 8, if the center of the discussion of the content of the conversation with a large risk evaluation value (the state of the conversation is bad) is a word with a high degree of abstraction, it can be estimated that the degree of risk is also large.

By the way, the second determination unit 130 of the conversation analyzer 100 according to the first embodiment specifies the abstraction degree of the word based on the abstraction degree determination table 105b generated based on the concept DB 50. It is not limited. The second determination unit 130 may directly compare the word (or compound word) with the concept DB 50 to specify the degree of abstraction corresponding to the word (or compound word).

FIG. 9 is a functional block diagram showing the configuration of the conversation analyzer according to the second embodiment. As shown in FIG. 9, the conversation analyzer 200 includes a storage unit 205, a voice recognition unit 210, a first determination unit 220, a second determination unit 230, an evaluation unit 240, and a generation unit 250.

The storage unit 205 has a setting keyword table 205a, a risk evaluation value table 205b, and an abstraction degree determination table 205c. The storage unit 205 corresponds to semiconductor memory elements such as RAM, ROM, and flash memory, and storage devices such as HDD.

The setting keyword table 205a is a table that defines setting keywords to be extracted for determining the state of conversation. FIG. 10 is a diagram showing an example of the data structure of the setting keyword table according to the second embodiment. As shown in FIG. 10, various setting keywords are registered in the setting keyword table 205a.

FIG. 11 is a diagram showing an example of the data structure of the risk evaluation value table according to the second embodiment. As shown in FIG. 11, the risk evaluation value table 205b associates the number of occurrences with the risk evaluation value. The number of occurrences indicates the number of appearances of the set keyword among the words included in the voice recognition result of the conversation voice data 20. The risk evaluation value is a value indicating the degree of poor state of the meeting, and the larger the risk evaluation value, the worse the state of the meeting.

For example, when the number of occurrences is "Na or more and less than Nb", the risk evaluation value is "0". When the number of appearances is "Nb or more and less than Nc", the risk evaluation value is "1". When the number of appearances is "Nc or more", the risk evaluation value is "2". Here, the values of Na, Nb, and Nc are preset values, and the magnitude relationship is Na <Nb <Nc. The risk evaluation value may be specified using the two threshold values Na and Nb. For example, if it is less than Na, the risk evaluation value may be "0", if it is Na or more and less than Nb, the risk evaluation value may be "1", and if it is Nb or more, the risk evaluation value may be "2".

The abstraction degree determination table 205c is a table that associates words with the abstraction degree. The data structure of the abstraction degree determination table 205c is the same as the data structure of the abstraction degree determination table 105b described with reference to FIG.

Returning to the description of FIG. Each of the processing units 210, 220, 230, 240 in FIG. 9 can be realized by a CPU, MPU, or the like. Further, each processing unit 210-240 can be realized by hard-wired logic such as ASIC or FPGA.

The conversation analyzer 200 acquires the conversation voice data 20 from an external device (not shown). The conversation voice data 20 is voice data recorded from a conference conversation, for example, associating time with voice power.

The voice recognition unit 210 is a processing unit that analyzes the conversation voice data 20 and extracts the spoken word. The voice recognition unit 210 outputs the information of each extracted word to the first determination unit 220 and the second determination unit 230. The voice recognition unit 210 may use any voice recognition technology. For example, the spoken word is converted into a character string based on the characteristics of the voice included in the conversation voice data 20, and the word is extracted based on the dictionary information that defines the word.

The first determination unit 220 is a processing unit that determines the state of conversation based on the voice recognition result of the conversation voice data 20. For example, the first determination unit 220 counts the number of occurrences of the set keyword in each word of the voice recognition result based on the voice recognition result of the voice recognition unit 210 and the set keyword table 205a. The first determination unit 220 compares the counted number of occurrences with the risk evaluation value table 205b to determine the risk evaluation value (conversation state). The first determination unit 220 outputs the state of conversation that is the determination result to the evaluation unit 240.

The second determination unit 230 is a processing unit that determines the degree of abstraction of the conversation content based on the voice recognition result of the voice recognition unit 210 and the abstraction degree determination table 205c. The second determination unit 230 outputs information on the degree of abstraction of the determined conversation content to the evaluation unit 240. The process in which the second determination unit 230 determines the abstraction degree of the conversation content is the same as the process in which the second determination unit 130 described in the first embodiment determines the abstraction degree of the conversation content.

The evaluation unit 240 is a conversation target included in the conversation voice data 20 based on the conversation state (risk evaluation value) output from the first judgment unit 220 and the abstraction degree output from the second judgment unit 230. It is a processing department that evaluates the risk of. The process in which the evaluation unit 240 evaluates the risk is the same as the process in which the evaluation unit 140 evaluates the risk.

The generation unit 250 is a processing unit that acquires the language corpus 25 and generates the abstraction degree determination table 205c. The language corpus 25 is a language corpus related to the conversation target. For example, if the conversation target is related to a conference, the language corpus 25 is a language corpus related to the conference.

The generation unit 250 analyzes the language corpus 25 for morphological elements and totals the frequency of appearance of each morpheme (hereinafter, word). The generation unit 250 calculates the abstraction degree for each word in the abstraction degree determination table 205c based on the frequency of occurrence of the words.

FIG. 12 is a diagram for explaining the processing of the generation unit according to the second embodiment. Table 10C of FIG. 12 is a table for associating words with the frequency of occurrence of words in the language corpus 25. The generation unit 250 may calculate the abstraction degree of each word based on the equation (2), or may calculate the rank in ascending order of appearance frequency as the abstraction degree.

Abstraction = log (frequency of appearance) x 2 ... (2)

For example, the degree of abstraction of each word calculated based on the equation (2) is expressed as the first degree of abstraction. The degree of abstraction of each word calculated based on the order of appearance frequency is referred to as the second degree of abstraction. The generation unit 250 registers the abstraction degree of either the first abstraction degree or the second abstraction degree in the abstraction degree determination table 205c.

Next, an example of the processing procedure of the conversation analyzer 200 according to the second embodiment will be described. FIG. 13 is a flowchart showing a processing procedure of the conversation analyzer according to the second embodiment. As shown in FIG. 13, the conversation analyzer 200 acquires the conversation voice data 20 (step S201). The conversation analyzer 200 executes voice recognition on the conversation voice data 20 and extracts the spoken word (step S202).

The first determination unit 220 of the conversation analyzer 200 counts the number of occurrences of the set keyword based on the spoken word, and determines the voice state (risk evaluation value) (step S203).

The second determination unit 230 of the conversation analyzer 200 determines the degree of abstraction of the conversation content based on the uttered word and the abstraction degree determination table 205c (step S204).

The evaluation unit 240 of the conversation analyzer 200 comprehensively evaluates the risk of the conversation target based on the conversation state (risk evaluation value) and the abstraction level of the conversation content (step S205). The evaluation unit 240 notifies the evaluation result (step S206).

Next, the effect of the conversation analyzer 200 according to the second embodiment will be described. The conversation analyzer 100 evaluates the risk of the conversation target by using the abstraction degree of the conference content in addition to the state of the conference based on the number of occurrences of the set keyword determined by the first determination unit 110. This makes it possible to appropriately evaluate the degree of risk as a conversation target.

FIG. 14 is a functional block diagram showing an example of the configuration of the conversation analyzer according to the third embodiment. As shown in FIG. 14, the conversation analyzer 300 includes a storage unit 305, a first determination unit 310, an extraction unit 320, a voice recognition unit 330, a second determination unit 340, and an evaluation unit 350. ..

The storage unit 305 has a risk evaluation value table 305a and an abstraction degree determination table 305b. The storage unit 305 corresponds to semiconductor memory elements such as RAM, ROM, and flash memory, and storage devices such as HDD.

The risk evaluation value table 305a is a table that associates the ratio with the risk evaluation value. The description of the ratio and the risk evaluation value is the same as the description given in the risk evaluation value table 105a of Example 1. The data structure of the risk evaluation value table 305a is the same as the data structure of the risk evaluation value table 105a described with reference to FIG.

The abstraction degree determination table 305b is a table that associates words with the abstraction degree. The description of the degree of abstraction is the same as the description given in the level of abstraction determination table 105b of the first embodiment. The data structure of the abstraction degree determination table 305b is the same as the data structure of the abstraction degree determination table 305b described with reference to FIG.

The conversation analyzer 300 acquires the conversation voice data 20 from an external device (not shown). The conversation voice data 20 is voice data recorded from a conference conversation, for example, associating time with voice power.

The first determination unit 310 is a processing unit that analyzes the conversation voice data 20 and determines the state of the conversation. The first determination unit 310 calculates the ratio of the longest utterance section to the conference time in the same manner as the first determination unit 110. The first determination unit 310 compares the calculated ratio with the risk evaluation value table 305a to determine the risk evaluation value. The first determination unit 110 outputs the determination result of the risk evaluation value (conversation state) to the evaluation unit 350.

Further, the first determination unit 310 outputs the information of the "longest utterance section" used when calculating the ratio to the extraction unit 320.

The extraction unit 320 is a processing unit that extracts a section that contributes to the determination of the state of conversation by the first determination unit 310 as a “section of interest”. FIG. 15 is a diagram for explaining the processing of the extraction unit according to the third embodiment. For example, assuming that the longest utterance section is time t ₁ to time t ₂ , the extraction unit 320 extracts time t _1a to time t _2a as the section of interest. The time t _1a is a time before a predetermined time (for example, 5 minutes) from t ₁ . The time t _2a is a time after a predetermined time (for example, 5 minutes) after t ₂ . The extraction unit 320 outputs the information of the section of interest to the second determination unit 340.

The voice recognition unit 330 is a processing unit that analyzes the conversation voice data 20 and extracts the spoken word. The voice recognition unit 330 outputs the information of each extracted word to the second determination unit 340. The voice recognition unit 330 may use any voice recognition technology. For example, the spoken word is converted into a character string based on the characteristics of the voice included in the conversation voice data 20, and the word is extracted based on the dictionary information that defines the word.

The second determination unit 340 is a processing unit that determines the degree of abstraction of the conversation content based on the words uttered in the section of interest and the abstraction degree determination table 305b among the words of the voice recognition result of the voice recognition unit 330. Is. The second determination unit 340 outputs information on the degree of abstraction of the determined conversation content to the evaluation unit 350.

For example, the second determination unit 340 compares a plurality of words included in the section of interest with the abstraction degree determination table 305b, and counts the number of occurrences of each word for each word registered in the abstraction degree determination table 305b. To do. The second determination unit 340 determines the degree of abstraction based on the number of occurrences of each counted word. The process of determining the degree of abstraction based on the number of occurrences of each counted word is the same as the process of the second determination unit 130 described in the first embodiment.

The evaluation unit 350 is a conversation target included in the conversation voice data 20 based on the conversation state (risk evaluation value) output from the first judgment unit 310 and the abstraction degree output from the second judgment unit 340. It is a processing department that evaluates the risk of. The process in which the evaluation unit 350 evaluates the risk is the same as the process in which the evaluation unit 140 evaluates the risk.

Next, an example of the processing procedure of the conversation analyzer 300 according to the third embodiment will be described. FIG. 16 is a flowchart showing a processing procedure of the conversation analyzer according to the third embodiment. As shown in FIG. 16, the conversation analyzer 300 acquires the conversation voice data 20 (step S301).

The first determination unit 110 of the conversation analyzer 300 determines the conversation state (risk evaluation value) based on the ratio of the longest utterance time to the conference time and the risk evaluation value table 305a (step S302). The extraction unit 320 of the conversation analyzer 300 extracts the section of interest that contributed to the determination of the conversation state (step S303).

The voice recognition unit 330 of the conversation analyzer 300 executes voice recognition on the conversation voice data 20 and extracts the spoken word (step S304). The second determination unit 340 of the conversation analyzer 300 determines the degree of abstraction of the conversation content based on the words uttered during the section of interest and the abstraction degree determination table 305b (step S305).

The evaluation unit 350 of the conversation analyzer 300 comprehensively evaluates the risk of the conversation target based on the conversation state (risk evaluation value) and the abstraction level of the conversation content (step S306). The evaluation unit 350 notifies the evaluation result (step S307).

Next, the effect of the conversation analyzer 300 according to the third embodiment will be described. When determining the degree of abstraction of the content of the conference, the conversation analyzer 300 determines the degree of abstraction using the words uttered in the section of interest that contributed to the determination of the state of the conference, and uses it for risk evaluation of the conversation target. This makes it possible to evaluate the magnitude of risk using the level of abstraction of the section closely related to the state of the meeting.

By the way, the conversation analyzer 300 according to the third embodiment has extracted the section of interest that contributed to the determination of the state of conversation by using the longest utterance section, but the present invention is not limited to this. For example, the extraction unit 320 of the conversation analyzer 300 may extract the section of interest based on the utterance section including the set keyword. In this case, the first determination unit 310 acquires the voice recognition result from the voice recognition unit 330, determines whether or not the set keyword is included in the utterance section, and obtains information on the utterance section including the set keyword. , Output to the extraction unit 320.

Next, an example of a computer hardware configuration that realizes the same functions as the conversation analyzer 100 (200, 300) shown in this embodiment will be described. FIG. 17 is a diagram showing an example of a computer hardware configuration that realizes the same functions as the conversation analyzer according to the present embodiment.

As shown in FIG. 17, the computer 500 includes a CPU 501 that executes various arithmetic processes, an input device 502 that receives data input from a user, and a display 503. Further, the computer 500 has a reading device 504 that reads a program or the like from a storage medium, and an interface device 505 that exchanges data with an external device or the like via a wired or wireless network. The computer 500 has a RAM 506 for temporarily storing various information and a hard disk device 507. Then, each device 501 to 507 is connected to the bus 508.

The hard disk device 507 includes a voice recognition program 507a, a first determination program 507b, an extraction program 507c, a second determination program 507d, and an evaluation program 507e. The CPU 501 reads out the voice recognition program 507a, the first determination program 507b, the extraction program 507c, the second determination program 507d, and the evaluation program 507e and deploys them in the RAM 506.

The voice recognition program 507a functions as a voice recognition process 506a. The first determination program 507b functions as the first determination process 506b. The extraction program 507c functions as the extraction process 506c. The second determination program 507d functions as the second determination process 506d. The evaluation program 507e functions as the evaluation process 506e.

The processing of the voice recognition process 506a corresponds to the processing of the voice recognition units 120, 210, 330. The processing of the first determination process 506b corresponds to the processing of the first determination units 110, 220, 310. The processing of the extraction process 506c corresponds to the processing of the extraction unit 320. The processing of the second determination process 506d corresponds to the processing of the second determination units 130, 230, 340. The processing of the evaluation process 506e corresponds to the processing of the evaluation units 140, 240, 350.

It should be noted that the programs 507a to 507e do not necessarily have to be stored in the hard disk device 507 from the beginning. For example, each program is stored in a "portable physical medium" such as a flexible disk (FD), a CD-ROM, a DVD disk, a magneto-optical disk, or an IC card inserted into the computer 500. Then, the computer 500 may read and execute each of the programs 507a to 507e.

The following additional notes will be further disclosed with respect to the embodiments including each of the above embodiments.

(Appendix 1) The first judgment unit that analyzes the conversation voice and judges the state of the conversation,
A voice recognition unit that performs voice recognition on the conversation voice and extracts words included in the conversation voice.
A second determination unit that determines the degree of abstraction of the conversation content in the conversation voice based on the words included in the conversation voice.
A conversation analyzer characterized by having an evaluation unit that evaluates the content of the conversation based on the state of the conversation and the degree of abstraction.

(Appendix 2) The first determination unit is based on the number of occurrences of a predetermined word included in the conversation voice or the longest utterance time among a plurality of utterance times included in the conversation voice. The conversation analyzer according to Appendix 1, wherein an evaluation value indicating the degree of risk of the content is determined as the state of the conversation.

(Appendix 3) The conversation analyzer according to Appendix 1 or 2, wherein the second determination unit determines the degree of abstraction based on the frequency of appearance of words included in the language corpus.

(Appendix 4) The second determination unit is characterized in that the abstraction degree is determined based on the conceptual structure information located in the upper layer as the more abstract words are and the words included in the conversation voice. The conversation analyzer according to Appendix 2.

(Appendix 5) Among the conversation voice sections, the second determination unit further includes an extraction unit that extracts the section that contributed to the determination of the conversation state as the interest section, and the second determination unit extracts words included in the interest section. The conversation analyzer according to any one of Supplementary notes 1 to 4, wherein the degree of abstraction of the conversation content is determined based on the conversation content.

(Appendix 6) A conversation analysis method executed by a computer.
Analyze the conversation voice to determine the state of the conversation,
Voice recognition is performed on the conversation voice, and words included in the conversation voice are extracted.
Based on the words contained in the conversation voice, the degree of abstraction of the conversation content in the conversation voice is determined.
A conversation analysis method characterized by executing a process of evaluating the conversation content based on the state of the conversation and the degree of abstraction.

(Appendix 7) The process of determining the state of the conversation is based on the number of occurrences of a predetermined word included in the conversation voice or the longest utterance time among a plurality of utterance times included in the conversation voice. The conversation analysis method according to Appendix 6, wherein an evaluation value indicating the degree of risk of the conversation content is determined as the state of the conversation.

(Appendix 8) The conversation analysis method according to Appendix 6 or 7, wherein the process of determining the degree of abstraction determines the degree of abstraction based on the frequency of appearance of words included in the language corpus.

(Appendix 9) In the process of determining the degree of abstraction, the degree of abstraction is determined based on the conceptual structure information located in the higher layer of the more abstract words and the words included in the conversation voice. The conversation analysis method according to Appendix 7, characterized by the above.

(Appendix 10) Of the conversation voice sections, the process of extracting the section that contributed to the determination of the conversation state as the section of interest is further executed, and the process of determining the degree of abstraction is a word included in the section of interest. The conversation analysis method according to any one of Appendix 6 to 9, wherein the degree of abstraction of the conversation content is determined based on the above.

(Appendix 11) To the computer
Analyze the conversation voice to determine the state of the conversation,
Voice recognition is performed on the conversation voice, and words included in the conversation voice are extracted.
Based on the words contained in the conversation voice, the degree of abstraction of the conversation content in the conversation voice is determined.
A conversation analysis program characterized by executing a process of evaluating the conversation content based on the state of the conversation and the degree of abstraction.

(Appendix 12) The process of determining the state of the conversation is based on the number of occurrences of a predetermined word included in the conversation voice or the longest utterance time among a plurality of utterance times included in the conversation voice. The conversation analysis program according to Appendix 11, wherein an evaluation value indicating the degree of risk of the conversation content is determined as the state of the conversation.

(Appendix 13) The conversation analysis program according to Appendix 11 or 12, wherein the process of determining the degree of abstraction determines the degree of abstraction based on the frequency of appearance of words included in the language corpus.

(Appendix 14) In the process of determining the degree of abstraction, the degree of abstraction is determined based on the conceptual structure information located in the higher layer of the more abstract words and the words included in the conversation voice. The conversation analysis program according to Appendix 12, which comprises the above.

(Appendix 15) Of the conversation voice sections, a process of further executing a process of extracting a section contributing to the determination of the conversation state as a section of interest and determining the degree of abstraction is included in the section of interest. The conversation analysis program according to any one of Supplementary note 11 to 14, wherein the degree of abstraction of the conversation content is determined based on the words.

100, 200, 300 Conversation analyzer 105, 205, 305 Storage unit 105a, 205b, 305a Risk evaluation value table 105b, 205c, 305b Abstraction degree judgment table 110, 220, 310 First judgment unit 120, 210, 330 Speech recognition unit 130, 230, 340 Second judgment unit 140, 240, 350 Evaluation unit 205a Setting keyword table 320 Extraction unit

Claims (7)

The first judgment unit that analyzes the conversation voice and judges the state of the conversation,
A voice recognition unit that performs voice recognition on the conversation voice and extracts words included in the conversation voice.
A second determination unit that determines the degree of abstraction of the conversation content in the conversation voice based on the words included in the conversation voice.
A conversation analyzer characterized by having an evaluation unit that evaluates the content of the conversation based on the state of the conversation and the degree of abstraction. The first determination unit determines the risk of the conversation content based on the number of occurrences of a predetermined word included in the conversation voice or the longest utterance time among a plurality of utterance times included in the conversation voice. The conversation analyzer according to claim 1, wherein an evaluation value indicating the degree is determined as the state of the conversation. The conversation analyzer according to claim 1 or 2, wherein the second determination unit determines the degree of abstraction based on the frequency of appearance of words included in the language corpus. The second determination unit is characterized in that the degree of abstraction is determined based on the conceptual structure information located in the upper layer as the more abstract words are and the words included in the conversation voice. 2. The conversation analyzer according to 2. The second determination unit further includes an extraction unit that extracts a section of the conversation voice section that has contributed to the determination of the conversation state as a section of interest, and the second determination unit is based on a word included in the section of interest. The conversation analyzer according to any one of claims 1 to 4, wherein the degree of abstraction of the conversation content is determined. A computer-executed conversation analysis method
Analyze the conversation voice to determine the state of the conversation,
Voice recognition is performed on the conversation voice, and words included in the conversation voice are extracted.
Based on the words contained in the conversation voice, the degree of abstraction of the conversation content in the conversation voice is determined.
A conversation analysis method characterized by executing a process of evaluating the conversation content based on the state of the conversation and the degree of abstraction. On the computer
Analyze the conversation voice to determine the state of the conversation,
Voice recognition is performed on the conversation voice, and words included in the conversation voice are extracted.
Based on the words contained in the conversation voice, the degree of abstraction of the conversation content in the conversation voice is determined.
A conversation analysis program characterized by executing a process of evaluating the conversation content based on the state of the conversation and the degree of abstraction.

Images