JP2010055307A - Conference support system and conference support method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technology for objectively evaluating deterioration in the quality of a conference. <P>SOLUTION: An opinion acquisition device (2) of a conference support system (1) creates time-series opinion results (26) based on the number of opinions that participants of a conference input for each evaluation item. A detection device (3) of the conference support system acquires a physical quantity for one or a plurality of evaluation items during the conference, and creates an evaluation result (37) including time-series quality values acquired based on the physical quantity. An evaluation device (4) of the conference support system calculates the final time-series quality value of the conference based on the opinion results and evaluation results for each evaluation item. Further, the evaluation items include the detection of either repeated statements, prohibited statements, statements not conforming to the main subject of the conference, exclusive statements, statements with sound pressures equal to or greater than a predetermined threshold value, or an elapsed time equal to or longer than a predetermined threshold value. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a conference support system and a conference support method.

In a meeting where a plurality of participants meet, the agenda is generally decided. In such a meeting, the quality of the meeting may be deteriorated by repeating the same statement or derailing the topic away from the agenda. A technology is disclosed that allows participants to convey to the moderator of the conference without hesitation that these quality degradation factors have just occurred (for example, Patent Document 1).
JP 2007-241620 A (paragraphs 0008 to 0011, FIGS. 4 and 5)

However, in the technique of Patent Document 1, even if the participants point out that there are many repetitions, “is derailed from the topic”, etc., as a moderator, whether or not there are many repetitions, derailment from the topic It is not possible to judge whether or not this is based on an objective basis. Therefore, it is difficult to give attention to the participants who are causing the quality degradation factor.
Therefore, an object of the present invention is to provide a technique for objectively evaluating a decrease in the quality of a conference.

  The opinion acquisition device of the conference support system according to the present invention creates a time-series opinion result based on the number of opinions for each evaluation item input by a conference participant. The detection device of the meeting support system acquires a physical quantity during a meeting for one or a plurality of evaluation items, and creates a time-series evaluation result including a quality value acquired based on the physical quantity. An evaluation device of a meeting support system is characterized in that a final quality value of a time-series meeting is calculated based on an opinion result and an evaluation result for each evaluation item. Furthermore, the evaluation items include any of repetitive utterances, banned utterances, utterances that do not meet the main topic of the meeting, exclusive utterances, utterances having a sound pressure equal to or higher than a predetermined threshold, and elapsed time that exceeds a predetermined threshold. It is characterized by including the detection.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to provide the technique which objectively evaluates the fall of the quality of a meeting.

  Hereinafter, the best mode for carrying out the present invention (referred to as “the present embodiment”) will be described in detail with reference to the drawings and the like.

(overall structure)
FIG. 1 is a diagram illustrating an example of the overall configuration of the conference support system according to the present embodiment.
The meeting support system 1 includes an opinion acquisition device 2, a detection device 3, and an evaluation device 4. These are connected to each other via communication interfaces (IF) 25, 36, and 49 included in each device.

The opinion acquisition device 2 is a general computer, and includes a central control device 21, a main storage device 22, an auxiliary storage device 23, a keyboard 24, and a communication interface (IF) 25. These are connected to each other by a bus. One keyboard 24 is assigned to each participant. A plurality of keyboards 24 are connected via an input / output interface (IF).
In the auxiliary storage device 23 of the opinion acquisition device 2, an opinion result 26 and an opinion reflection magnification table 27 are stored. The opinion reflection magnification setting unit 28 is a program.
In the following description, when the subject is described as “XX section is”, the central control device 21 reads out each program from the auxiliary storage device 23 and loads it into the main storage device 22 to execute the function of each program. (The same applies to the programs of the detection device 3 and the evaluation device 4).

The detection device 3 is a general computer and includes a central control device 31, a main storage device 32, an auxiliary storage device 33, and a communication interface (IF) 36. These are connected to each other by a bus. The microphone 34 is connected to the bus via an input / output interface (not shown).
An evaluation result 37 is stored in the auxiliary storage device 33 of the detection device 3. The detection unit 38 is a program. The detection unit 38 is also a program, the repetitive speech detection unit 39, the NG speech detection unit 40, the derailment speech detection unit 41, the exclusive speech detection unit 42, the emotional speech detection unit 43, and the end detection unit 44 (these are referred to as “each It may also be referred to as a “detection unit”. The main storage device 32 includes a timer 35.

The evaluation device 4 is a general computer and includes a central control device 45, a main storage device 46, an auxiliary storage device 47, and a communication interface (IF) 49. These are connected to each other by a bus. The display 48 is connected to the bus via an input / output interface (not shown). It is assumed that one or a plurality of displays 48 are arranged at a place where all the participants can see. However, you may arrange | position in the place which can be visually recognized only with respect to the presenter.
The auxiliary storage device 47 of the evaluation device 4 includes an opinion result 26 (same as the opinion result 26 created by the opinion acquisition device 2), an evaluation result 37 (same as the evaluation result 37 created by the detection device 3), A quality border 50 and a display result 51 are stored. The final quality value calculation unit 52 and the man-hour calculation unit 53 are programs.

  1 shows an example in which the opinion acquisition device 2, the detection device 3, and the evaluation device 4 are independent of each other, two or three of them may be configured as one computer.

(Evaluation items, functions of each device)
In the conference support system 1 of the present embodiment, the following six items are evaluated.
That is, repetitive speech (same speech), NG speech (prohibited speech), derailment speech (non-conforming speech), exclusive speech (a speech that a particular participant monopolizes the meeting time) ), Emotional speech (a speech having a sound pressure equal to or higher than a predetermined threshold), and the elapsed time from the start of the conference. These six items may be referred to as “evaluation items” hereinafter.
The participant means a member other than the moderator among members attending the conference using the conference support system 1 of the present invention (the same applies hereinafter).

The function of each device will be roughly described.
The opinion acquisition device 2 receives a subjective evaluation for each evaluation item from the participant and transmits it to the evaluation device 4.
The detection device 3 transmits an objective evaluation for each evaluation item to the evaluation device 4 based on the information acquired from the microphone 34 and the timer 35.
The evaluation device 4 comprehensively evaluates and displays the quality of the conference by combining the information received from the opinion acquisition device 2 and the detection device 3.

FIG. 2A shows an example of the opinion button correspondence table according to the present embodiment.
The opinion buttons are six types of keys provided on the keyboard 24 in order to receive a participant's subjective opinion regarding each evaluation item.
The opinion button correspondence table 61 associates opinion contents with their identification numbers, and the same information is stored in the auxiliary storage devices 23, 33, and 47 of the opinion acquisition device 2, the detection device 3, and the evaluation device 4. Remembered (not shown).
The identification number in the identification number column 62 is a number that identifies an evaluation item, and is also a number that identifies each detection unit 39 to 44 of the detection device 3 that executes processing for each evaluation item.
The opinion content in the opinion content column 63 is a character string that briefly expresses factors that degrade the quality of the conference in each evaluation item. This opinion content is displayed in the vicinity of the opinion button on the keyboard 24 or is known to the participants in advance.
The detection unit name in the detection unit column 64 is the name of each of the detection units 39 to 44 of the detection device 3 that executes processing for each evaluation item. The detection unit name does not need to be disclosed to the participants.

FIG. 2B is a diagram showing an example of the opinion reflection magnification table according to the present embodiment.
The vertical axis 65 stores the number of opinions, and the horizontal axis 66 stores the identification number.
The number of opinions on the vertical axis 65 is the number of times for each evaluation item that the participant has pressed the opinion button.
The identification number on the horizontal axis 66 is the same as the identification number in FIG.
Opinion reflection magnification is stored at the intersection of the vertical axis 65 and the horizontal axis 66. The opinion reflection magnification is a number by which the quality value (described later) is multiplied when the number of opinions on the vertical axis 65 is generated for the evaluation item indicated by the identification number on the horizontal axis 66.
Incidentally, there is “× 4” at the intersection of the fourth row and the first column. This is because if the opinion button is pressed three times for the opinion content “Same utterance” by a participant (may be due to one participant or multiple participants) The quality value calculated by the “detection unit” 39 indicates that the quality value is quadrupled.
The opinion reflection magnification can be set to an arbitrary value by the user. In general, the larger the number of opinions, the larger the value that reflects the opinion is set. Even if the number of opinions is the same, the opinion reflection magnification may be set differently if the identification numbers are different. For example, in an example where repeated utterances are strictly controlled and derailment utterances are relatively acceptable, even if the number of opinions is the same “3”, the opinion reflection magnification “× 10” is displayed in the column of the identification number 1. An opinion reflection magnification “× 1” may be set in the column. In this embodiment, the opinion reflection magnification table 27 is given.

FIG. 2C is a diagram illustrating an example of an opinion result according to the present embodiment.
The opinion result 26 is created by the opinion acquisition device 2 and transmitted to the evaluation device 4.
In association with the identification number stored in the identification number column 67, the opinion number column 68 stores the number of opinions, and the opinion reflection magnification column 69 stores the opinion reflection magnification.
The identification number in the identification number column 67 is the same as the identification number in FIG.
The number of opinions in the opinion number column 68 is the same as the number of opinions in FIG.
The opinion reflection magnification in the opinion reflection magnification column 69 is the same as the opinion reflection magnification in FIG. Here, the opinion reflection magnification, which is the result of searching the opinion reflection magnification table 27 based on the actual number of opinions, is stored.
By the way, the example of the opinion result 26 in FIG. 2 (c) shows that the opinion buttons are four times, once for the opinion contents “same statement”, “you must not speak”, etc. As a result of pressing, the quality values calculated by the repeated speech detection unit 39, the NG speech detection unit 40,.

FIG. 3A is a diagram illustrating an example of an evaluation result according to the present embodiment.
The evaluation result 37 is created by the detection device 3 and transmitted to the evaluation device 4.
In association with the identification number stored in the identification number column 72, the quality value column 73 stores quality values, and the reference information column 74 stores reference information.
The identification number name in the identification number column 72 is the same as the identification number in FIG.
The quality value in the quality value column 73 is a numerical value indicating the quality of the conference acquired by each of the detection units 39 to 44 indicated by the identification number (details will be described later).
The reference information in the reference information column 74 is a numerical value acquired by each of the detection units 39 to 44 indicated by the identification number, an extracted character string, and other information.
For example, the repeated speech detection unit 39 indicates that the total number of comparisons in a certain time zone is 64, and the words “array”, “placement”, “check”, and “placement, check” Suppose that it is detected that the word is repeated (detailed later).
In this case, in the first line, the reference information column 74 stores the heading “repetition count” with the number of repetitions “4”, and the repetition degree “6.3% (4 ÷ 64)” is “repetition”. The heading “degree” is stored and stored, and the repeated word “array” and the like are stored with the heading “repetitive content”. Further, the quality value column 73 stores the quality value “17.5” acquired by the repetitive speech detection unit 39.

FIG. 3B is a diagram showing an example of a quality border according to the present embodiment.
A comment is stored in the comment field 76 in association with the range of the final quality value stored in the final quality value field 75.
The final quality value 75 is a value obtained by multiplying the quality value calculated by the detection device 3 for each evaluation item by the opinion reflection magnification acquired by the opinion acquisition device 2 and summing the products.
For example, the quality value in the quality value column 73 of the evaluation result 37 (FIG. 3A) is “17.5, 16.7, −12, −10, −20, 0”, and the opinion result 26 (FIG. 2). When the opinion reflection magnification in the opinion reflection magnification column 69 of (c)) is “5, 2, 3, 1, 1, 2”, the final quality value is (17.5) × 5 + (16.7) × 2 + (− 12) × 3 + (− 10) × 1 + (− 20) × 1 + 0 × 2 = 54.9.
The range of the final quality value in the final quality value column 75 is a range that the final quality value calculated in this way can take. The range of final quality values is stored continuously with a predetermined width.
The comment in the comment field 76 is an arbitrary character string to be displayed on the display 48 as a part of the display result 51 (detailed later). As the final quality value becomes smaller (low quality), the meaning of the output comment becomes negative. Although there is no comment for the final quality value “0”, an arbitrary character string may be set. The width of the range of the final quality value and the number of steps can be arbitrarily set by the user. In this embodiment, the quality border 50 is given.

FIG. 4 is a diagram illustrating an example of a display result according to the present embodiment.
The display result 51 is information displayed on the display 48.
In association with the number of opinions stored in the opinion number column 77, the subjective opinion content column 78 contains the subjective opinion content, the subjective / objective result column 79 shows the subjective / objective result, and the man-hour column 80 shows the man-hour. Is remembered.
The number of opinions in the opinion number column 77 is the number of opinions having the largest value among the numbers of opinions in FIG.
The subjective opinion content in the subjective opinion content column 78 is the opinion content corresponding to the number of opinions stored in the opinion number column 77 in FIG. 4 among the opinion content in FIG. When the number of opinions is “0”, “No opinion”.
The subjective / objective result in the subjective / objective result column 79 corresponds to the number of opinions stored in the comment number column 77 of FIG. 4 in the comment of FIG. 3 (b) and the reference information of FIG. 3 (a). It is the character string produced based on the reference information which each detection part 39-44 produced.
This subjective / objective result is information based on objective information detected by the detection device 3, unlike the subjective opinion content. Therefore, if the participant visually recognizes the subjective / objective result, the quality of the conference can be maintained.
The man-hour in the man-hour column 80 is a total value of products obtained by multiplying each participant's hourly wage by the elapsed time of the conference (the time from the start time to the present time). There are cases where a positive value is taken and cases where a negative value is taken (details will be described later).

The record of the display result 51 is stored in association with the elapsed time (speech balloons 81 to 84). For example, when the display result 51 is displayed immediately after the start of the conference, 20 minutes later, and 40 minutes later, records are added in the order of (a), (b), (c), for example, as time passes. In this case, since the opinion button has never been pressed, it can be said that the development of the conference is ideal.
As another example, records may be added in the order of (a), (d), and (e). In this case, there are many participants who particularly point out repetitive remarks, and as the time passes, the number of times the opinion button is pressed increases, and the quality of the conference decreases.
The number of opinions, subjective opinion contents, subjective / objective results, and man-hours of the display result 51 are displayed on the display 48 (not shown) with appropriate headings attached thereto. For example, “Number of people who pressed the button: 10 people”, “Please be careful: the same remarks”, “Computer judgment: the same remarks ... "," Our profit or loss: -5800 yen ".

(Contents of various information)
FIG. 5 is a diagram for explaining the contents of various types of information according to the present embodiment.
The information acquired through the microphone 34 is a natural voice 85 (a waveform that changes over time). The detection device 3 can convert the natural voice 85 into speech information 86, voiceprint information 87, and sound pressure information 88. Further, the detection device 3 can extract the extracted word 89 from the utterance information 86 and acquire the synonym 90 based on the extracted word 89. Further, the detection device 3 can extract the extracted voice print information 91 from the voice print information 87. These can all be realized using known techniques.

The comment information 86 is a character string associated with the passage of time.
The voiceprint information 87 is a frequency and an amplitude associated with the passage of time. The amplitude is expressed in color (such as contour lines) on a two-dimensional plane.
The sound pressure information 88 is sound pressure (amplitude) associated with the passage of time.
The extracted words 89 are nouns, verbs, and modifiers for nouns and verbs extracted from the remark information 86. The extracted word 89 may be extracted from the utterance information 86 or may be directly input from the keyboard 24. It may be input from a keyboard (not shown) of the detection device 3.
The synonym 90 is a word having the same or similar meaning as the extracted word 89.
The extracted voiceprint information 91 is a part of the voiceprint information 87 in terms of time.

(Comparison source, comparison destination, switching interval, evaluation interval)
FIG. 6 is a diagram for explaining the comparison source, the comparison destination, the switching interval, and the evaluation interval according to the present embodiment.
The detection device 3 creates speech information 86, voiceprint information 87, and sound pressure information 88 in real time with the passage of time after the start of the conference. That is, a new file (86, 87, 88 arranged in time series) is created every time the switching timing 97 arrives. A time length between a certain switching timing 97 and the next switching timing 97 is referred to as a “switching interval” 98, and a time period divided by the switching timing 97 is referred to as a switching period. The switching periods are sequentially assigned in the order of 1, 2,..., N−1, n, n + 1,.
On the other hand, the detection device 3 creates an extracted word 92 as an NG word, an extracted word 93 as a main subject, a sample voiceprint information database 94, an allowable sound pressure 95, and a scheduled time 96 before starting the conference, and an auxiliary storage device 33 is stored.

Each detection part 39-44 of the detection apparatus 3 compares the said files, or compares the said file and the said information (code | symbol 92-96) produced in advance. Hereinafter, these comparison targets are referred to as “comparison source” and “comparison destination”. In FIG. 6, two symbols (a, b, etc.) written in the balloons on the right side of the respective detection units 39 to 44 are set as “comparison source” (left) and “comparison destination” (right). Information is shown.
For example, the repetitive speech detection unit 39 compares the speech information 86b (also expressed as f) of the switching cycle n-1 with the speech information 86a (also expressed as g) of the switching cycle n, and acquires the quality value.
The NG utterance detection unit 40 compares the extracted word 92 (noted as “a”) created before the start of the conference with the utterance information 86a (also denoted as “g”) of the switching period n, and acquires a quality value. To do. The six quality values created in this way are transmitted to the evaluation device 4 as a part of the evaluation result 37.
The opinion acquisition device 2 creates an opinion result 26 in synchronization with the detection device 3. Then, the opinion result 26 is transmitted to the evaluation device 4.

There is an evaluation timing 99 as a concept different from the switching timing 97. The evaluation timing 99 is a timing for evaluating the quality of the conference (corresponding to “20 minutes later” and “40 minutes later” in the balloons 81 to 84 in FIG. 4). A time length between one evaluation timing 99 and the next evaluation timing 99 is referred to as an “evaluation interval” 100, and a time period divided by the evaluation timing 99 is referred to as an evaluation cycle. The evaluation periods are assigned in order of 1, 2,..., N−1, n, n + 1,.
The operations of the detection units 39 to 44 of the evaluation device 4, the opinion acquisition device 2, and the detection device 3 are performed every evaluation cycle.

  The switching timing 97 and the evaluation timing 99 may or may not match. If they do not match, the detection device 3 is controlled at two timings. That is, when the switching interval 98 is shorter than the evaluation interval 100, not all of the comment information 86 and the like created for each switching interval 98 are used. When the switching interval 98 is longer than the evaluation interval 100, each of the detection units 39 to 44 waits without performing processing until new comment information 86 and the like are created.

(Processing procedure)
Hereinafter, the processing procedure will be described. The processing procedure includes (1) opinion reflection magnification acquisition processing procedure, (2) repeated speech detection processing procedure, (3) NG speech detection processing procedure, (4) derailment speech detection processing procedure, and (5) exclusive speech detection processing procedure. , (6) Emotional utterance detection processing procedure, (7) End detection processing procedure, and (8) Result display processing procedure. In order to execute (8), it is assumed that the execution of (1) to (7) has been completed.
Although details will be described later, step S311 in (2), step S321 in (3), step S331 in (4), step S341 in (5), step S351 in (6), and step S361 in (7) are: Repeated every switching cycle. On the other hand, all the steps of (1), (2) after step S312, (3) after step S322, (4) after step S332, (5) after step S342, (6) after step S352, After step S362 of (7) and all steps of (8) are repeated every evaluation cycle.

(Opinion reflection rate acquisition processing procedure)
FIG. 7 is a flowchart of an opinion reflection magnification acquisition processing procedure according to this embodiment.
In step S301, the opinion reflection magnification setting unit 28 of the opinion acquisition device 2 acquires the number of opinions.
Specifically, the opinion reflection magnification setting unit 28 acquires, for each identification number, the number of times the participant presses the opinion button (the number of opinions) for each evaluation interval 100.

In step S302, the opinion reflection magnification setting unit 28 acquires an opinion reflection magnification.
Specifically, the opinion reflection magnification setting unit 28 searches the opinion reflection magnification table 27 using the number of opinions acquired in step S301 as a search key on the vertical axis and the identification number as a search key on the horizontal axis. Get the opinion reflection ratio of.

In step S <b> 303, the opinion reflection magnification setting unit 28 creates the opinion result 26.
Specifically, the opinion reflection magnification setting unit 28 stores the number of opinions acquired in step S301 and the opinion reflection magnification acquired in step S302 in the auxiliary storage device 23 as the opinion result 26 in association with the identification number. To do.
Here, the opinion reflection magnification acquisition processing procedure ends.

  As a modified example of step S301, the opinion reflection magnification setting unit 28 may reset to “0” when the number of acquired opinions exceeds a predetermined threshold, or may not accept more opinions. Also good.

(Repeated speech detection processing procedure)
FIG. 8 is a flowchart of the repeated speech detection processing procedure according to this embodiment.
FIG. 9 is a diagram for explaining functions and the like of the repeated speech detection unit 39 according to the present embodiment.
FIGS. 10A, 10B, and 10C are diagrams illustrating information temporarily stored in the main storage device 32 in the process in which the repetitive speech detection unit 39 according to the present embodiment functions. FIG. 10D is a diagram illustrating an example of a quality value correspondence table used by the repetitive speech detection unit 39 according to the present embodiment. The quality value correspondence table 124 is stored in the auxiliary storage device 33. Hereinafter, the repetitive speech processing procedure will be described along the flowchart of FIG. 8 with reference to FIGS. 9 and 10 as appropriate.
As a premise, the repeated speech detection unit 39 includes a writing unit 101, a writing control unit 102, a comparison information extraction unit 103, and a comparison unit 104, which are also programs (FIG. 9). This also applies to the other detection units 40 to 44 (FIGS. 12, 16, 19, and 21).

In step S <b> 311, the writing unit 101 acquires natural voice 85 via the microphone 34 and time information from the timer 35, and creates speech information 86 as a physical quantity.
At this time, the writing control unit 102 switches the file to which the remark information 86 created by the writing unit 101 is written every time the switching timing 97 arrives. In this way, a plurality of time-series message information files (reference numerals 86a and 86b) are created. You may overwrite the oldest file with the latest message information.

In step S312, the comparison information extraction unit 103 selects two pieces of speech information 86.
Specifically, the comparison information extraction unit 103 acquires the latest file (symbol 86a) and the immediately preceding file (symbol 86b) among the files in which the statement information 86 is written, and obtains the latest file. Is the file for comparison, and the file immediately before is the file for comparison.
If two or more files do not exist, wait until two or more files are created. It is assumed that information indicating that the file has been selected once is attached to each file, and a file to which this information is attached is not selected as a file for comparison. In a later-described NG speech detection processing procedure, derailment speech detection processing procedure, exclusive speech detection processing procedure, emotion speech detection processing procedure, and end detection processing procedure, the comparison destination (comparison source in the exclusive speech detection processing procedure) to be acquired The same applies when the file does not exist.

In step S313, the comparison information extraction unit 103 extracts the extracted words 111 and 114.
Specifically, the comparison information extraction unit 103 obtains nouns and verbs from the two message information files 86 a and 86 b (hereinafter, “file” is simply abbreviated as “message information”) acquired in step S312. , And modifiers for them are extracted as extracted words.
In FIG. 10A, the comparison information extraction unit 103 extracts “array” (noun), “noun” as the extracted word 111 from the utterance information 86b of the comparison source 100 “... check the array ...”. "Check" (verb) is extracted. In FIG. 10B, the comparison information extraction unit 103 extracts “processing” (noun) and “examine” as the extracted word 114 from the utterance information 86a “... "(Verb).

As a modified example of step S313, when the extracted words 111 and 114 cannot be extracted from at least one of the two utterance information 86a and 86b, the process may return to step S312.
For example, the comparison information extraction unit 103 may not be able to extract the extracted words 111 and 114 from the utterance information 86a and 86b, for example, when all participants are reading materials during the meeting.

In step S314, the comparison unit 104 acquires the synonyms 112 and 115.
Specifically, the comparison unit 104 acquires the synonyms 112 and 115 by referring to the thesaurus information 105 for each of the extracted words 111 and 114 extracted as described above. The number of synonyms 112 and 115 to be acquired is not particularly limited (it may be zero). The thesaurus information 105 is lexicographic information in which groups of words having the same or similar meaning are recorded.
In FIG. 10A, the comparison unit 104 “arranges”, “arrangement” and “property” from the extracted word 111 “array” as synonyms 112 and “blocks” as the synonym 112 from the extracted words 111 “check”. Have acquired each. In FIG. 10B, the comparison unit 104 performs “research” and “check” as the synonym 115 from the extracted word 114 “process” as the synonym 115 and “study” from the extracted word 114 “examine”. Have acquired each.

In step S315, the comparison unit 104 creates the word of the comparison source 116 and the word of the comparison destination 119.
Specifically, the comparison unit 104 first arranges the extracted word 111 and the synonym 112 of FIG. 10A from the top to the bottom in the extracted word and synonym column 117 of the comparison source 116 of FIG. .
At this time, the comparison unit 104 determines a combination of the extracted words 111 and 114 and the synonyms 112 and 115 that are nouns and the extracted words 111 and 114 and the synonyms 112 and 115 that are verbs, as in “placement and check”. Create and line up under it. Although not shown here as an example, the comparison unit 104 may arrange a combination of a modifier and a noun, or a combination of a modifier and a verb.
Furthermore, a combination that is semantically unnatural from such combinations may be deleted by a known technique, or the same word is duplicated in the extracted word 111 and the synonym 112. May delete duplicates (the same applies to the comparison destination).
Second, the comparison unit 104 arranges the extracted words 114 and the synonyms 115 of FIG. 10B from the top to the bottom in the extracted words and synonyms column 120 of the comparison destination 119 of FIG.
At this time, the comparison unit 104 creates combinations and arranges them under the same as the comparison source.

In step S316, the comparison unit 104 performs comparison. That is, the extracted word extracted from the remark information 86 as a physical quantity acquired during the meeting and its synonyms are compared between two different periods.
The comparison unit 104 uses the same word or combination in the extracted word and synonym column 120 of the comparison destination 119 in FIG. 10C for each extracted word or combination in the comparing word 116 and the synonym column 117 in FIG. Is stored in the hit number columns 118 and 121 of the comparison source (destination), if not, “0” is stored.
For example, since the “array” in the second row of the comparison destination 119 matches the “array” in the first row of the comparison source 116, “1” is set in the hit number column 118 in the first row of the comparison source 116. Then, “1” is stored in the hit number column 121 on the first line of the comparison destination 119. The hit number column 121 of the comparison destination may be omitted.
In this example, as a result of deleting the overlap, the word of a certain comparison source 116 does not match the words of a plurality of comparison destinations 119, and either “1” or “0” is determined. Assume that it is stored as the number of hits.

In step S317, the comparison unit 104 calculates the repetition rate. That is, the number of times of matching is obtained as a result of comparing the extracted word extracted from the utterance information 86 as a physical quantity acquired during the meeting and its synonyms between two different periods.
Specifically, the comparison unit 104 totals the hit numbers in the hit number column 118 of the comparison source 116 and sets the total value as the repetition count 122. Then, a percentage (referred to as “repeat degree”) obtained by dividing the repeat count 122 by the product of the record number of the comparison source 116 and the record count of the comparison destination 119 (referred to as “total comparison count” 123) is calculated.
In the example of FIG. 10C, the number of repetitions 122 is “4”, the total number of comparisons is 64 (8 × 8), and the repetition rate is 6.3% (4 ÷ 64 × 100).
This calculation method is merely an example, and the present invention is not limited to this. The same applies to the following processing procedures.

In step S318, the comparison unit 104 acquires a quality value. That is, the quality value is acquired based on the number of times of matching and the total number of the extracted words and their synonyms or the number of times the extracted words and their synonyms are compared.
Specifically, the comparison unit 104 acquires a quality value from the quality value correspondence table 124 (detailed later in FIG. 10D) based on the repetition rate calculated in step S317. In this example, the quality value “17.5” is acquired based on the repetition rate of 6.3%.

In step S319, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 in FIG. 3A, and acquires “1” in the identification number column 72 and the quality value column 73 in step S318. Each quality value is stored. In the reference information column 74, the total number of repetitions in step S317, the repetition rate calculated in step S317, and the extracted word, synonym, or combination in which “1” is entered in the hit number column 118 in step S316, The headings “repetition count”, “repetition degree” and “repetition content” are added and stored.
Here, the repeated speech detection processing procedure ends.

<Quality value correspondence table>
The quality value correspondence table 124 in FIG. 10D is a table created by the user in advance, and is stored in the auxiliary storage device 33.
Assume that the quality value correspondence table 124 exists for the total number of comparisons. The quality value correspondence table 124 in FIG. 10D is used when the total number of comparisons is “64” (the actual total number of comparisons is a numerical value much larger than 64, but here it is easy to understand. For this reason, it is 64).
The degree of repetition and the quality value are stored in association with the number of repetitions.
In the number of repetitions row 125, the number of repetitions is stored as 64, 63,..., 1, 0 from left to right starting from “64” which is the total number of comparisons.
The repeat degree row 126 stores the repeat degree. For example, in the first column, the number of repetitions 64 ÷ total number of comparisons 64 = 100% is stored. In the second column, the number of repetitions 63 ÷ the total number of comparisons 64 = 98.4% is stored.
The quality value row 127 stores quality values. The quality value is a numerical sequence in which the range from the minimum value −20 to the maximum value 20 is cut at equal intervals.
In this example, the difference between the maximum value and the minimum value is divided by the total number of comparisons, and the step width is ((20 − (− 20)) ÷ (64) = 0.6.

(NG speech detection processing procedure)
FIG. 11 is a flowchart of the NG speech detection processing procedure according to the present embodiment.
FIG. 12 is a diagram for explaining functions and the like of the NG message detection unit 40 according to the present embodiment.
FIGS. 13A, 13B, and 13C are diagrams for explaining information temporarily held in the main storage device 32 in the process in which the NG speech detection unit 40 according to the present embodiment functions. FIG. 13D is a diagram showing an example of a quality value correspondence table used by the NG statement detection unit 40 according to the present embodiment. The quality value correspondence table 141 is stored in the auxiliary storage device 33. Hereinafter, the NG speech processing procedure will be described along the flowchart of FIG. 11 with reference to FIGS. 12 and 13 as appropriate.
Note that FIGS. 11 and 12 are also commonly referred to in the description of the derailment speech detection processing procedure described later. Accordingly, the parentheses in FIGS. 11 and 12 are ignored here.
As a premise, the auxiliary storage device 33 stores extracted words 92 as NG words (FIG. 12). This is a word that is input from an input device such as the keyboard 24 before the conference starts.

  Step S321 is the same as step S311 of the repeated speech detection processing procedure. As a result, a plurality of time-series message information 86 files are created.

In step S322, the comparison information extraction unit 103 selects one piece of speech information 86.
Specifically, the comparison information extraction unit 103 acquires the latest file (symbol 86a) in time among the files in which the statement information 86 is written, and sets this as a comparison destination file.

In step S323, the comparison information extraction unit 103 extracts the extracted word 134.
Specifically, the comparison information extraction unit 103 extracts nouns, verbs, and modifiers for the nouns, verbs, and modifiers as extracted words 134 from the single comment information file 86a acquired in step S322.
In FIG. 13B, the comparison information extraction unit 103 uses “speaking” (modifier) as the extracted word 134 from the utterance information 86 a of the comparison destination 133 “. ), “Previous” (noun), and “perform” (verb). In FIG. 13A, an extracted word 92 as an NG word is stored, which has already been decomposed into parts of speech. Therefore, the comparison information extraction unit 103 does not process the extracted word 92 as an NG word.

  As a modified example of step S323, when the extracted word 134 cannot be extracted from the utterance information 86a at all, the process may return to step S322.

Step S324 is substantially the same as step S314 of the repeated speech processing procedure.
The difference from step S314 is that the source from which synonyms are extracted is a group of two extracted words 111 and 114 in step S314, whereas in step S324, a group of one extracted word 134 and a group of 1 This is a group of extracted words 92 as two NG words.
In FIG. 13A, the comparison unit 104 “executes” “previous” as the synonym 132 from the extracted word 92 “previous” as the NG word, and “executes” as the synonym 132 from the extracted word 92 “execute” as the NG word. ”And“ Implement ”. In FIG. 13B, the comparison unit 104 sets “previous” and “destination” as synonyms 135 from the extracted word 134 “previous”, and “enforces” as the synonym 135 from the extracted word 134 “execute”, respectively. Have acquired. No synonyms have been acquired for the extracted word 92 “speaking” and the extracted word 134 “speaking” as NG words.

  Step S325 is the same as step S315 of the repeated speech detection processing procedure.

  Step S326 is the same as step S316 in the repeated speech detection processing procedure.

In step S327, the comparison unit 104 calculates an NG speech level. That is, the number of times of coincidence obtained as a result of comparing the extracted word extracted from the utterance information as the physical quantity acquired during the meeting and its synonyms with the preset prohibited utterance and its synonyms is acquired.
Specifically, the comparison unit 104 totals the hit numbers in the hit number column 138 of the comparison source 136 and sets the total value as the number of NG utterances 139. Then, a percentage (referred to as “NG speech level”) obtained by dividing the number of NG speeches 139 by the “total number of comparisons” 140 is calculated.
In the example of FIG. 13C, the number of NG utterances is “4”, the total number of comparisons is 48 (6 × 8), and the degree of NG utterance is 8.3% (4 ÷ 48 × 100).

In step S328, the comparison unit 104 acquires a quality value. That is, the quality value is acquired based on the number of times of matching and the total number of prohibited words and their synonyms, or the number of times of comparison of prohibited words and their synonyms with extracted words and their synonyms.
Specifically, the comparison unit 104 acquires a quality value from the quality value correspondence table 141 (FIG. 13 (d) details later) based on the NG speech level calculated in step S327. In this example, the quality value “16.7” is acquired based on the repetition rate of 8.3%.

In step S329, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 in FIG. 3A, and acquires “2” in the identification number column 72 and the quality value column 73 in step S328. Each quality value is stored. In the reference information column 74, the total number of NG utterances calculated in step S327, the NG utterance degree calculated in step S327, and the extracted word, synonym, or combination in which “1” is described in the hit number column 138 in step S326 Are stored with headings "NG speech count", "NG speech level" and "NG speech content", respectively.
Here, the repeated speech detection processing procedure ends.

<Quality value correspondence table>
The quality value correspondence table 141 in FIG. 13D is a table created in advance by the user, and is stored in the auxiliary storage device 33.
The quality value correspondence table 141 in FIG. 13D has the same configuration as the quality value correspondence table 124 in FIG. However, a “NG speech count” row 142 is stored instead of the “repeat count” row 125, and an “NG speech count” row 143 is stored instead of the “repeat count” row 126.

(Derailment remark detection processing procedure)
FIG. 11 is a flowchart of the derailment speech detection processing procedure according to this embodiment.
FIG. 12 is a diagram for explaining functions and the like of the derailment message detection unit 41 according to the present embodiment.
FIGS. 14A, 14B, and 14C are diagrams for explaining information temporarily stored in the main storage device 32 in a process in which the derailment speech detection unit 41 according to the present embodiment functions. FIG. 14D is a diagram illustrating an example of a quality value correspondence table used by the derailment speech detection unit 41 according to the present embodiment. The quality value correspondence table 174 is stored in the auxiliary storage device 33. Hereinafter, the derailment speech processing procedure will be described along the flowchart of FIG. 11 with reference to FIGS. 12 and 14 as appropriate.
As a premise, the auxiliary storage device 33 stores an extracted word 93 as the main subject (FIG. 12). This is a word that is input from an input device such as the keyboard 24 before the conference starts.
It should be noted that FIGS. 11 and 12 are also commonly referred to in the description of the NG speech detection processing procedure described above. Therefore, the part in parentheses in FIGS. 11 and 12 is referred to here.

  Step S331 is the same as step S321 of the NG speech detection processing procedure.

  Step S332 is the same as step S322 of the NG speech detection processing procedure.

In step S333, the comparison information extraction unit 103 extracts the extracted word 154.
Specifically, the comparison information extraction unit 103 extracts nouns, verbs, and modifiers as extracted words 154 from the single statement information file 86 a acquired in step S <b> 332.
In FIG. 14B, the comparison information extraction unit 103 extracts “dump analysis” as the extracted word 154 from the utterance information 86a of the comparison destination 153 “... Explains the YYY analysis supporting the dump analysis. “YYY analysis” (noun) and “explain” (verb) are extracted. In FIG. 14A, an extracted word 93 as the main subject is stored, which has already been decomposed into parts of speech. Therefore, the comparison information extraction unit 103 does not process the extracted word 93 as the main subject.

  As a modified example of step S333, when no extracted word can be extracted from the utterance information 86a, the process may return to step S332.

Step S334 is almost the same as step S324 of the NG speech detection processing procedure.
The difference from step S324 is that the source for extracting synonyms is a group of one extracted word 134 and a group of extracted words 92 as one NG word in step S324, whereas in step S334 This is a group of one extracted word 154 and a group of extracted words 93 as one main word.
In FIG. 14A, the comparison unit 104 “explains” from the extracted word 93 “slip” as the main subject as a synonym 152 and “explains” as the synonym 152 from the extracted word 93 “lectures” as the main subject. ", Respectively. In FIG. 14B, the comparison unit 104 acquires “ZZZ analysis” as the synonym 155 from the extracted word 154 “dump analysis”, and “explains” as the synonym 155 from the extracted word 154 “explain”. Yes. Synonyms are not acquired for the extracted word 93 “XXX form creation” and the extracted word 154 “YYY analysis” as the main words.

In step S335, the comparison unit 104 creates the word of the comparison source 156 and the word of the comparison destination 159.
Specifically, the comparison unit 104 first adds the extracted word 93 and the synonym 152 as the main subject in FIG. 14A to the extracted word and synonym column 157 of the comparison source 156 in FIG. Line up from the bottom.
Furthermore, when the same word is duplicated in the extracted word 93 and the synonym 152 as the main subject, the duplicate may be deleted. If the same word is duplicated in the extracted word 154 and the synonym 155, the duplicate may be deleted.
Second, the comparison unit 104 arranges the extracted word 154 and the synonym 155 of FIG. 14B from the top to the bottom in the extracted word and synonym column 160 of the comparison destination 159 of FIG.

<Difference from step S315>
The processing in step S315 “creates a combination of extracted words and synonyms that are nouns and extracted words and synonyms that are verbs and arranges them underneath” is omitted here. This is because here, comparison is mainly made between words such as nouns and verbs. Of course, as in step S315, a combination of nouns and verbs may be created and these may be used as comparison targets.

  Step S336 is the same as step S326 of the NG speech detection processing procedure.

In step S337, the comparison unit 104 calculates a derailment degree. That is, the number of derailments is obtained by subtracting the number of coincidences from either the total number of the main subject utterances and their synonyms or the number of comparisons between the main subject utterances and their synonyms and the extracted words and their synonyms.
Specifically, the comparison unit 104 totals the hit numbers in the hit number column 158 of the comparison source 156 and sets the total value as the main subject utterance count 171. Then, the number obtained by subtracting the number of main subject utterances 171 (referred to as “number of derailment utterances” 173) from the “total number of phrases” 172 that is the number of records of the comparison source 156, and the percentage (“derailment degree”). Calculated).
In the example of FIG. 14C, the main subject utterance count 171 is “1”, the phrase total number 172 is “5”, the derailment utterance count 173 is “4”, and the derailment degree is 80% ((5- 1) ÷ 5 × 100).

<Difference from step S317>
In step S317, it has been described that the denominator of the repeat degree is the total number of comparisons, and here, the denominator of the derailment degree is the total number of words. However, in both step S317 and step S337, either the total number of comparisons or the total number of words may be adopted as the denominator. However, the repeatability and derailment values are larger when the denominator is the total number of phrases.

In step S338, the comparison unit 104 acquires a quality value. That is, the quality value is acquired based on the number of derailments and either the total number of the main subject utterances and their synonyms or the number of times the main subject utterances and their synonyms are compared with the extracted words and their synonyms.
Specifically, the comparison unit 104 acquires a quality value from the quality value correspondence table 174 (FIG. 14 (d) details later) based on the derailment degree calculated in step S337. In this example, the quality value “−12” is acquired based on the derailment degree of 80%.

In step S339, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 of FIG. 3A, and acquires “3” in the identification number column 72 and the quality value column 73 in step S338. Each quality value is stored. In the reference information column 74, the number of derailment utterances 173 calculated in step S337, the derailment degree calculated in step S337, and the extracted word or synonym in which “0” is described in the hit number column 170 in step S336, respectively. The headings “derailment utterance count”, “derailment degree”, and “derailment utterance content” are added and stored.
Here, the repeated speech detection processing procedure ends.

<Quality value correspondence table>
The quality value correspondence table 174 in FIG. 14D is a table created in advance by the user, and is stored in the auxiliary storage device 33.
The quality value correspondence table 174 in FIG. 14D has the same configuration as the quality value correspondence table 124 in FIG. However, a “derailment frequency” row 175 is stored instead of the “repetition frequency” row 125, and a “derailment rate” row 176 is stored instead of the “repetition rate” row 126.

(Exclusive speech detection processing procedure)
FIG. 15 is a flowchart of the exclusive message detection processing procedure according to this embodiment.
FIG. 16 is a diagram for explaining functions and the like of the exclusive message detection unit 42 according to the present embodiment.
FIGS. 17A and 17C are diagrams illustrating information temporarily held in the main storage device 32 in the process in which the exclusive message detection unit 42 according to the present embodiment functions. FIG. 17B is a diagram illustrating an example of a sample voiceprint information database used by the exclusive speech detection unit 42 according to the present embodiment. The sample voiceprint information database 94 and the quality value correspondence table 188 are stored in the auxiliary storage device 33. Hereinafter, the exclusive speech processing procedure will be described along the flowchart of FIG. 15 with reference to FIGS. 16 and 17 as appropriate.

<Sample Voiceprint Information Database>
Prior to the meeting, sample voiceprint information 184 of the participant is acquired. First, the user determines a plurality of “sentences” that the participant should speak. The sentence is not particularly limited, but is preferably a word (greeting word, company name, etc.) that can be spoken by any of the participants during the meeting. Then, the detection device 3 acquires a plurality of “sentences” determined by all participants through the microphone 34.
In the sample voiceprint information database 94 of FIG. 17B, sample voiceprint information is stored in the sample voiceprint information column 184 and sentence is stored in the sentence column 185 in association with the participant name in the participant column 183. Yes.
The participant in the participant column 183 is the name of the participant who participates in the conference.
The sample voice print information in the sample voice print information column 184 is voice print information when the sentence is spoken. Similar to the voiceprint information 87 of FIG. 5, the sample voiceprint information is information having time information on the horizontal axis, frequency on the vertical axis, and amplitude as a color. The length of time on the horizontal axis is not particularly limited (it varies depending on the sentence length and participants).
The sentence in the sentence column 185 is the text information that is spoken.
There are as many records in the sample voiceprint information database 94 as there are combinations of participants and sentences.
Incidentally, the record in the first line indicates that the sample voiceprint information when Mr. A remarked “Thank you” was “SSSS”.

In step S341, the writing unit 101 acquires natural voice 85 through the microphone 34 and time information from the timer 35, and creates voiceprint information 87 as a physical quantity.
At this time, the writing control unit 102 switches the file to which the voiceprint information 87 created by the writing unit 101 is written every time the switching timing 97 arrives. In this way, a plurality of time-series voiceprint information 87 files are created. You may overwrite the oldest file with the latest voiceprint information.

In step S342, the comparison information extraction unit 103 selects one voiceprint information 87.
Specifically, the comparison information extraction unit 103 acquires the latest file in time from among the files in which the voiceprint information 87 is written, and sets this as a comparison source file.

In step S343, the comparison information extraction unit 103 extracts the extracted voiceprint information 91.
Specifically, the comparison information extraction unit 103 first divides the acquired voiceprint information 87 into a speech time zone (column 181) having a predetermined time interval (referred to as “speech interval”), Time-series extracted voiceprint information 91 is created. The speech interval is a concept different from the switching interval 98 and is much shorter than the switching interval.
Second, the comparison information extraction unit 103 acquires a character string corresponding to the divided speech time zone from the speech information 87.
Thirdly, the comparison information extraction unit 103 associates the extracted extracted voiceprint information with the character string for each speech time period and temporarily stores them in the main storage device 32 (FIG. 17A). .

In the table of FIG. 17A, the extracted voiceprint information column 91 stores the extracted voiceprint information and the speech information column 86a stores the speech information in association with the speech timezone stored in the speech timezone column 181. ing.
The speech time zone in the speech time zone column 181 is information indicating the start point and end point of the speech time zone by the elapsed time from the start of the conference, such as “0 to 2 (seconds)”.
The extracted voiceprint information in the extracted voiceprint information column 91 is the voiceprint information 87 that is spoken during the speech time zone.
The speech information in the speech information column 86a is a character string that is spoken during the speech time zone. These character strings are nothing but a character string included in one utterance information file 86a divided for each utterance time zone.
There are as many records in the table of FIG. 17A as the number obtained by dividing the switching interval 98 by the speech interval.
By the way, in the record on the first line, during the period from 0 to 2 seconds after the start of the meeting (this 2 seconds is the “speaking interval”), one of the participants said “Thank you for gathering” The extracted voiceprint information is “SSSZ”.

  As a modified example of step S343, when the extracted voiceprint information 91 cannot be extracted from the voiceprint information 87 at all, the process may return to step S342.

In step S344, the comparison unit 104 performs comparison. In other words, the sample voiceprint information database is searched using the speech information corresponding to the extracted voiceprint information and the extracted voiceprint information as search keys, and the name of the participant who made the extracted voiceprint information is acquired for each extracted voiceprint information.
Specifically, the comparison unit 104 performs the following processing.
(1) The speech information “Thank you for gathering” of record # 1 in the table of FIG.
(2) The sentence column 185 of the sample voiceprint information database 94 is searched using the acquired speech information as a search key, and all records having the most similar sentence are acquired (similarity is determined using a known technique). ). Here, it is assumed that the records in the first to third lines including “Thank you” have been acquired.
(3) Using the extracted voiceprint information “SSSZ” of record # 1 in the table of FIG. 17A as a search key, search the sample voiceprint information column 184 of the record of the sample voiceprint information database 94 acquired in (2) above. The participant name of the record having the most similar sample voiceprint information is acquired (similarity is determined using a known technique). Here, it is assumed that the participant name “Mr. A” of the record having the sample voice print information “SSSS” is acquired.
(4) The speech time zone of record # 1 in the table of FIG. 17A and the participant name (speaker name) acquired in (3) above are associated and temporarily stored in the main storage device 32. (# 1 in FIG. 17C).
The speech time zone in the speech time zone column 186 in the table of FIG. 17C is the same as the speech time zone in the table of FIG.
The speaker name in the speaker column 187 is the participant name of the participant who is determined to have made a statement.
(5) The processes (1) to (4) are repeated for all the records # 1, # 2,... In the table of FIG. 17A, and as a result, the table of FIG. .

In step S345, the comparison unit 104 calculates the degree of monopoly. That is, the time lengths of the extracted voiceprint information are totaled for each participant who speaks, and the degree of monopoly is calculated based on the total and the time length of the voiceprint information acquired during the conference.
Specifically, the comparison unit 104 first adds up all the records in the table of FIG. 17C for each speaker name, and creates a group.
Secondly, the number of records belonging to each group is divided by the total number of records in the table of FIG. 17C to obtain the degree of monopoly (percentage) for each speaker. Here, if the number of all records is “3”, the monopoly degree for Mr. A is 2 ÷ 3 × 100 = 66.7%, and the monopoly degree for Mr. B is 1 ÷ 3 = 33.3. %.

In step S346, the comparison unit 104 acquires a quality value. That is, for each participant who made a statement, the time length of the extracted voiceprint information is totaled, and the quality value is obtained from the monopoly calculated based on the total and the time length of the voiceprint information acquired during the meeting. To do.
Specifically, the quality value is obtained from the quality value correspondence table 188 (detailed later) in FIG. 17D based on the largest degree of monopoly obtained in step S345. Here, “−10” with respect to “66.7%” is acquired.

  Note that as a modification of step S346, based on the sum of n (n is an arbitrary number smaller than the number of participants m) counted from the largest degree of monopoly obtained in step S345, FIG. The quality value may be obtained from the quality value correspondence table 188 (detailed later).

In step S347, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 in FIG. 3A, and acquires “4” in the identification number column 72 and the quality value column 73 in step S346. Each quality value is stored. In the reference information column 74, the monopoly degree calculated in step S345 is stored with the heading "Exclusive degree". Furthermore, if the degree of monopoly is a predetermined threshold or more, a character string such as “I have a monopoly on the topic” may be stored.
Here, the exclusive message detection processing procedure ends.

<Quality value correspondence table>
The quality value correspondence table 188 in FIG. 17D is a table created by the user in advance, and is stored in the auxiliary storage device 33.
In the quality value correspondence table 188 in FIG. 17D, the quality value is stored in the quality value row 190 in association with the range of the monopoly degree stored in the monopoly degree row 189.
The range of the monopoly degree of the monopoly degree row 189 is a range obtained by equally dividing 0 to 100% by a width of 20%.
The quality value in the quality value row 190 is a sequence of numbers divided at equal intervals, with the minimum value set to -20 and the maximum value set to 20.

(Emotion remark detection processing procedure)
FIG. 18 is a flowchart of the emotion utterance detection processing procedure according to this embodiment.
FIG. 19 is a diagram for explaining functions and the like of the emotion utterance detection unit 43 according to the present embodiment.
In the following, along with the flowchart of FIG. 18, the emotional speech processing procedure will be described with reference to FIG. 19 as appropriate.
As a premise, an allowable sound pressure 95 is stored in the auxiliary storage device 33 (FIG. 19).

<Allowable sound pressure>
Prior to the meeting, an allowable sound pressure 95 is set. The allowable sound pressure 95 is the sound pressure value (unit dB) itself indicating the allowable limit of the sound pressure information 88 acquired during the conference.

In step S <b> 351, the writing unit 101 acquires natural voice 85 through the microphone 34 and time information from the timer 35, and creates sound pressure information 88 as a physical quantity.
At this time, the writing control unit 102 switches the file to which the sound pressure information 88 created by the writing unit 101 is written every time the switching timing 97 arrives. In this manner, a plurality of time-series sound pressure information 88 files are created. You may overwrite the oldest file with the latest sound pressure information.

In step S352, the comparison information extraction unit 103 selects one sound pressure information 88.
Specifically, the comparison information extraction unit 103 acquires the latest file in time among the files in which the sound pressure information 88 is written, and sets this as a comparison destination file.

In step S353, the comparison information extraction unit 103 extracts the highest sound pressure 191.
Specifically, the comparison information extraction unit 103 acquires the highest value (maximum sound pressure 191) among the sound pressures stored in the sound pressure information 88 acquired in step S352, and stores it in the auxiliary storage device 33.

In step S354, the comparison unit 104 performs comparison. That is, the highest sound pressure extracted from the sound pressure information as a physical quantity acquired during the conference is compared with a preset allowable sound pressure.
Specifically, the comparison unit 104 compares the magnitudes of the allowable sound pressure 95 as a comparison source and the highest sound pressure 191 as a comparison destination. Then, the result of either “highest sound pressure ≧ allowable sound pressure” or “highest sound pressure <allowable sound pressure” is acquired.

In step S355, the comparison unit 104 acquires a quality value.
Specifically, the quality value is obtained from the quality value correspondence table 192 (assumed to be stored in the auxiliary storage device 33) based on the result obtained in step S354. In the quality correspondence table 192, for example, a quality value “−20” is obtained for the result “maximum sound pressure ≧ allowable sound pressure”, and a quality value “20” is obtained for the result “maximum sound pressure <allowable sound pressure”. Each is memorized.

In step S356, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 in FIG. 3A, and acquires “5” in the identification number column 72 and the quality value column 73 in step S355. Each quality value is stored.
In the reference information column 74, if the result obtained in step S354 is “maximum sound pressure ≧ allowable sound pressure”, the allowable sound pressure 95 is set to “above. Record with a heading. If the result obtained in step S354 is “maximum sound pressure <allowable sound pressure”, the sound pressure 191 is stored with the heading “not reached”.
Here, the exclusive message detection processing procedure ends.

(End detection procedure)
FIG. 20 is a flowchart of the end detection processing procedure according to this embodiment.
FIG. 21 is a diagram for explaining functions and the like of the end detection unit 44 according to the present embodiment.
Hereinafter, the end detection processing procedure will be described along the flowchart of FIG. 20 and with reference to FIG. 21 as appropriate.
As a premise, a scheduled time 96 is stored in the auxiliary storage device 33 (FIG. 21).

<Scheduled time>
Prior to the meeting, a scheduled time 96 is set. The scheduled time 96 is the scheduled value (unit: minutes) of the time from the start to the end of the conference.

In step S361, the writing unit 101 acquires natural voice 85 via the microphone 34, time information from the timer 35, and creates speech information 86 as a physical quantity.
At this time, the writing control unit 102 switches the file to which the remark information 86 created by the writing unit 101 is written every time the switching timing 97 arrives. In this way, a plurality of time-series message information 86 files are created. You may overwrite the oldest file with the latest message information.

In step S362, the comparison information extraction unit 103 acquires one piece of speech information 86.
Specifically, the comparison information extraction unit 103 acquires the latest file (symbol 86a) in time among the files in which the statement information 86 is written, and sets this as a comparison destination file.

In step S363, the comparison information extraction unit 103 extracts the elapsed time 193.
Specifically, the comparison information extraction unit 103 acquires the latest value (elapsed time 193) of the time on the horizontal axis stored in the statement information 86a acquired in step S362, and stores it in the auxiliary storage device 33. To do.

In step S364, the comparison unit 104 performs comparison. That is, the elapsed time as the physical quantity acquired during the meeting is compared with the preset scheduled time.
Specifically, the comparison unit 104 compares the magnitudes of the scheduled time 96 as a comparison source and the elapsed time 193 as a comparison destination. Then, the result of either “elapsed time ≧ scheduled time” or “elapsed time <scheduled time” is acquired.

In step S365, the comparison unit 104 acquires a quality value.
Specifically, the quality value is obtained from the quality value correspondence table 194 (assumed to be stored in the auxiliary storage device 33) based on the result obtained in step S364. In the quality correspondence table 194, for example, a quality value “−20” is stored for the result “elapsed time ≧ scheduled time”, and a quality value “20” is stored for the result “elapsed time <scheduled time”. It shall be.

In step S366, the comparison unit 104 stores the quality value and reference information.
Specifically, the comparison unit 104 creates a new record of the evaluation result 37 in FIG. 3A, and acquires “6” in the identification number column 72 and the quality value column 73 in step S365. Each quality value is stored.
In the reference information column 74, when the result acquired in step S364 is “elapsed time ≧ scheduled time”, the scheduled time 96 is stored with the heading “It is over”. If the result obtained in step S364 is “elapsed time <scheduled time”, the character string “time has not exceeded” is stored.
Here, the end detection processing procedure ends.

(Result display processing procedure)
FIG. 22 is a flowchart of a result display processing procedure according to this embodiment.
In step S <b> 371, the final quality value calculation unit 52 of the evaluation device 4 receives the opinion result 26 and the evaluation result 37.
Specifically, the final quality value calculation unit 52 receives the opinion result 26 created by the opinion acquisition device 2 in step S303 from the opinion acquisition device 2. Further, the detection device 3 receives from the detection device 3 the evaluation result 37 including all the records created in steps S319, S329, S339, S347, S356, and S365.

In step S372, the final quality value calculation unit 52 calculates the final quality value.
Specifically, the final quality value calculation unit 52 first multiplies the opinion reflection magnification of the opinion result 26 received in step S371 and the quality value of the evaluation result 37 with the same associated identification number. Then, the multiplication results are summed and this sum is taken as the final quality value.
Second, the quality border 50 is searched using the final quality value as a search key, and the corresponding comment is acquired.

In step S373, the man-hour calculation unit 53 of the evaluation device 4 calculates the man-hour.
Specifically, the man-hour calculation unit 53 first calculates salary per hour for each participant who pressed the opinion button × elapsed time (minutes) ÷ 60 (man-hours by opinion person).
Second, the total number of man-hours by opinion for all participants who pressed the opinion button is calculated.
Thirdly, if the final quality value is positive, a sign of “+” is added to the total value, and if the final quality value is negative, a sign of “−” is added to the total value.

In step S374, the final quality value calculation unit 52 creates the display result 51.
Specifically, the final quality value calculation unit 52 executes the following processing.
(1) A new record of the display result 51 is created.
(2) The largest number of opinions of the opinion result 26 received in step S371 is acquired and stored in the opinion number column 77 of the new record.
(3) Acquires the identification number of the record having the largest number of opinions of the opinion result 26 received in step S371, searches the opinion button correspondence table 61 using the acquired identification number as a search key, and Get the contents of the opinion of the record to be performed.
At this time, when there are a plurality of the largest opinions in the same rank, the youngest identification number is acquired. If all the opinions are “0”, “no opinion” is acquired as the opinion content.
(4) The opinion content acquired in (3) is stored in the subjective opinion content column 78 of the new record.
(5) The comment acquired in step S372 is stored in the subjective / objective result column 79 of the new record.
(6) Acquire the identification number of the record having the largest number of opinions of the opinion result 26 received in step S371, search the evaluation result 37 using the acquired identification number as a search key, and the corresponding record Get reference information. The acquired reference information is also stored in the subjective / objective result field 79 of the new record.
(7) The man-hour calculated in step S373 is stored in the man-hour column 80 of the new record. In this way, a new record of the display result 51 is completed.

In step S375, the final quality value calculation unit 52 displays the display result 51.
Specifically, the record of the display result 51 completed in step S374 is displayed on the display 48.
Here, the result display processing procedure ends.

It is a figure which shows an example of the whole structure of the meeting assistance system which concerns on this embodiment. (A) is a figure which shows an example of the opinion button corresponding table which concerns on this embodiment. (B) is a figure which shows an example of the opinion reflection magnification table which concerns on this embodiment. (C) is a figure which shows an example of the opinion result which concerns on this embodiment. (A) is a figure showing an example of an evaluation result concerning this embodiment. (B) is a figure which shows an example of the quality border based on this embodiment. It is a figure which shows an example of the display result which concerns on this embodiment. It is a figure for demonstrating the content of the various information which concerns on this embodiment. It is a figure for demonstrating the comparison origin, comparison destination, switching interval, and evaluation interval which concern on this embodiment. It is a flowchart of an opinion reflection magnification acquisition processing procedure according to the present embodiment. It is a flowchart of the repetitive speech detection processing procedure according to the present embodiment. It is a figure explaining the function etc. of the repeated speech detection part which concerns on this embodiment. (A), (b), (c) is a figure explaining the information temporarily hold | maintained in a main memory in the process in which the repeated speech detection part based on this embodiment functions. (D) is a figure which shows an example of the quality value corresponding table which the repeated speech detection part which concerns on this embodiment uses. It is a flowchart of the NG (derailment) utterance detection processing procedure which concerns on this embodiment. It is a figure explaining the function of the NG (derailment) statement detection part etc. which concern on this embodiment. (A), (b), (c) is a figure explaining the information temporarily hold | maintained in a main memory device in the process in which the NG utterance detection part concerning this embodiment functions. (D) is a figure which shows an example of the quality value corresponding table which the NG statement detection part which concerns on this embodiment uses. (A), (b), (c) is a figure explaining the information temporarily hold | maintained in a main memory in the process in which the derailment speech detection part concerning this embodiment functions. (D) is a figure which shows an example of the quality value corresponding table which the derailment speech detection part concerning this embodiment uses. It is a flowchart of the exclusive speech detection processing procedure which concerns on this embodiment. It is a figure explaining the function etc. of the exclusive speech detection part which concerns on this embodiment. (A), (c) is a figure explaining the information temporarily hold | maintained in a main memory in the process in which the exclusive message detection part which concerns on this embodiment functions. (B) is a figure which shows an example of the sample voiceprint information database which the exclusive speech detection part which concerns on this embodiment uses. It is a flowchart of the emotion speech detection processing procedure which concerns on this embodiment. It is a figure explaining the function etc. of the emotion speech detection part which concerns on this embodiment. It is a flowchart of the completion | finish detection processing procedure which concerns on this embodiment. It is a figure explaining the function etc. of the end detection part concerning this embodiment It is a flowchart of a result display processing procedure according to the present embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Meeting support system 2 Opinion acquisition apparatus 3 Detection apparatus 4 Evaluation apparatus 21, 31, 45 Central control apparatus (control part)
22, 32, 46 Main storage (storage unit)
23, 33, 47 Auxiliary storage device (storage unit)
24 Keyboard 26 Opinion Result 27 Opinion Reflection Magnification Table 28 Opinion Reflection Magnification Setting Unit 34 Microphone 35 Timer 37 Evaluation Result 39 Repetitive Speech Detection Unit 40 NG Speech Detection Unit 41 Derailment Speech Detection Unit 42 Exclusive Speech Detection Unit 43 Emotional Speech Detection Unit 44 End Detection unit 48 Display 50 Quality border 51 Display result 52 Final quality value calculation unit 53 Man-hour calculation unit 86 Speech information 87 Voiceprint information 88 Sound pressure information 94 Sample voiceprint information database

Claims (10)

A conference support system that includes an opinion acquisition device, a detection device, and an evaluation device, and evaluates the quality of the conference for one or more evaluation items,
The opinion acquisition device creates a time-series opinion result based on the number of opinions for each evaluation item input by a conference participant, and the detection device acquires a physical quantity during the conference for one or a plurality of the evaluation items. , Create a time-series evaluation result including the quality value obtained based on the physical quantity,
The evaluation device calculates a final quality value of a time series meeting based on the opinion result and the evaluation result for each evaluation item,
A conference support system characterized by that. The evaluation item includes repeated speech detection,
The detection device includes:
The extracted word extracted from the remark information as the physical quantity acquired during the meeting and its synonyms are compared between two different periods to obtain the number of matches.
Obtaining the quality value based on the number of matches and either the total number of the extracted words and their synonyms or the number of times the extracted words and their synonyms were compared;
The conference support system according to claim 1. The evaluation item includes detection of prohibited speech,
The detection device includes:
The extracted word extracted from the utterance information as the physical quantity acquired during the meeting and its synonyms are compared with the preset prohibited utterances and their synonyms to obtain the number of times of matching,
The quality value is acquired based on the number of times of matching and the total number of the prohibited words and their synonyms or the number of times the prohibited words and their synonyms are compared with the extracted word and its synonyms.
The conference support system according to claim 1. The evaluation item includes detection of a statement that does not conform to the main subject of the conference,
The detection device includes:
The extracted word extracted from the utterance information as the physical quantity acquired during the meeting and its synonyms are compared with the preset main subject utterance and its synonyms to obtain the number of matches.
The total number of the main subject utterances and synonyms thereof, or the number of derailments obtained by subtracting the number of coincidence from any number of times the main subject utterances and synonyms thereof and the extracted words and synonyms thereof are compared,
Obtaining the quality value based on the number of derailments and the total number of the subject utterances and their synonyms or the number of comparisons between the subject utterances and their synonyms and the extracted words and their synonyms;
The conference support system according to claim 1. The evaluation item includes detection of exclusive utterances,
The storage unit of the detection device is
A sample voiceprint information database that stores the sentence spoken by the participant and the sample voiceprint information in association with the participant name,
The control unit of the detection device,
Extracting extracted voiceprint information having a predetermined time length from the voiceprint information as the physical quantity acquired during the meeting,
Using the speech information corresponding to the extracted voiceprint information and the extracted voiceprint information as a search key, the sample voiceprint information database is searched, and the name of the participant who has spoken the extracted voiceprint information is obtained for each of the extracted voiceprint information,
Summing up the time length of the extracted voiceprint information for each participant who speaks, and acquiring the quality value based on the total and the time length of the voiceprint information acquired during the meeting,
The conference support system according to claim 1. The evaluation item includes detection of a speech having a sound pressure equal to or higher than a predetermined threshold,
The detection device acquires the quality value by comparing the maximum sound pressure extracted from the sound pressure information as the physical quantity acquired during the meeting with a preset allowable sound pressure.
The conference support system according to claim 1. The evaluation item includes detection of an elapsed time equal to or greater than a predetermined threshold,
The detection device acquires the quality value by comparing an elapsed time as the physical quantity acquired during a meeting with a preset scheduled time.
The conference support system according to claim 1. The opinion result includes an opinion reflection magnification determined based on the number of opinions to be multiplied by the quality value for each evaluation item,
The evaluation device is
For each evaluation item, find the product of the quality value and the opinion reflection magnification, and calculate the final quality value by summing the products.
The conference support system according to claim 1. A conference support method using a conference support system that includes an opinion acquisition device, a detection device, and an evaluation device and evaluates the quality of a conference for one or more evaluation items,
The opinion acquisition device creates a time-series opinion result based on the number of opinions for each evaluation item input by a conference participant, and the detection device acquires a physical quantity during the conference for one or a plurality of the evaluation items. , Create a time-series evaluation result including the quality value obtained based on the physical quantity,
The evaluation device calculates a final quality value of a time series meeting based on the opinion result and the evaluation result for each evaluation item,
A meeting support method characterized by the above. The opinion result includes an opinion reflection magnification determined based on the number of opinions to be multiplied by the quality value for each evaluation item,
The evaluation device is
For each evaluation item, find the product of the quality value and the opinion reflection magnification, and calculate the final quality value by summing the products.
The conference support method according to claim 9.

Images