JP2016012216A - Congress analysis device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To calculate a degree of livening up in a congress as a relative evaluation value, not as an absolute evaluation value.SOLUTION: A liven-up degree calculation unit 12 extracts a feature amount as time series along progress of a congress on the basis of congress data recorded as voice and/or video, calculates a liven-up degree as the time series along progress of the congress, as an absolute evaluation value on the basis of the feature amount, and then corrects it to a relative evaluation value at respective units 31, 32 and 33. By considering a reference liven-up degree as all congress participants, the liven-up degree is corrected over whole congress, in an integral correction unit 31. A sub team correction unit 32 considers a reference liven-up degree of some of participants with a topic which some of the participants participated in discussion, as a target, for correcting the liven-up degree for every section where the topic is discussed. An individual correction unit 33 corrects the liven-up degree on the basis of natures of the participants and utterance contents of the participants, for every utterance section of the respective participant in the congress.

Description

  The present invention relates to a conference analysis apparatus, method, and program for estimating the degree of excitement of a conference by analyzing the states of a plurality of participants in the conference.

  A vital part of the success of the conference is the excitement. By grasping the status of discussion in real time and performing control to get more ideas from participants, it is expected to activate discussion and create effective knowledge. For example, by presenting a value obtained by quantifying the excitement every time, it is possible to expect an effect of prompting the attendance of a conference participant who refers to the value.

  In recent years, techniques have been developed to measure the degree of excitement of meetings. For example, in Patent Document 1, the total amount of utterances of the participants in the conference hall is acquired, the speech action in the conference hall is recognized based on the acquired total utterance amount, and the conference activity is evaluated and calculated based on the recognition result. Presents the activity of the conference. Patent Document 2 provides a conference visualization system that captures the voices of multiple participants during a conference and displays the conversation status of the participants changing in real time, thereby inducing a more active discussion. To do. Patent Document 3 provides a meeting function calculation device that contributes to the activation of a meeting by evaluating the meeting activity in consideration of the relationship between meeting participants, such as the degree of utterance between meeting participants. Patent Document 4 provides a conference analysis system that can automatically analyze conference participants and the entire conference using only a simple sensor device having a volume sensor function and an acceleration sensor function.

  Further, Non-Patent Document 1 reports that there is a correlation between the degree of excitement and the characteristic amount of sound (basic frequency F0, energy). In Non-Patent Document 2, while segmenting speech data with a Hidden Markov Model (HMM), the degree of excitement of each segment is classified into high and low. In Non-Patent Document 3, topics are divided by Latent Dirichlet allocation (LDA), and the degree of excitement is estimated by a support vector machine (SVM).

JP 2006-302047 A JP 2008-262046 A JP 2009-163431 A JP 2012-113442 A

B. Wrede and E. Shriberg. Spotting hotspots in meetings: Human judgments and prosodic cues. In Proc. Eurospeech, Geneva, Sep. 2003. Gatica-Perez, D .; McCowan, I .; Zhang, D .; Bengio, S., "Detecting Group Interest-Level in Meetings," Acoustics, Speech, and Signal Processing, 2005. Proceedings. (ICASSP '05). IEEE International Conference on, vol.1, no., Pp.489,492, March 18-23, 2005. Kishita, Y .; Noguchi, H .; Sanada, H .; Mori, T., "A probabilistic inference of participants interest level in a multi-party conversation based on multi-modal sensing," Multimedia and Expo Workshops (ICMEW), 2013 IEEE International Conference on, vol., No., Pp.1,4, 15-19 July 2013

  The degree of excitement is a subjective indicator. Evaluation changes with people. For example, it cannot be said that Mr. A and Mr. B have the same evaluation. Of course, neither is the correct answer. Therefore, it is necessary to appropriately correct the measured degree of swell.

  However, all the above prior arts are absolute evaluations. Even if the participants, the location of the speech, and the content of the speech change, the evaluation of the excitement does not change. For example, the evaluation of the degree of excitement between bright A and mature B cannot be distinguished. In addition, the privacy story naturally goes down in absolute excitement. Conversely, the excitement of self-assertion is not the original excitement desired at the meeting.

  Further, when the conference is conducted remotely, further correction is necessary. In other words, not only is there an effect due to the difference in volume between each other, but the frequency of remarks naturally falls due to the remote psychological distance sense, so the absolute evaluation value can be corrected. is necessary.

  As described above, relative evaluation is necessary for various elements, not absolute evaluation. In other words, it is necessary to estimate the relative excitement by considering the participant's personality and place of participation, the content of the remarks, and the conversational relationship. As described above, in the prior art, such an estimation framework is used. Is not provided.

  The present invention has been made in view of the above-described problems of the prior art, and an object of the present invention is to provide a conference analysis apparatus, method, and program capable of calculating the degree of excitement of a conference as a relative evaluation.

In order to achieve the above object, the present invention provides a conference analysis apparatus for calculating a degree of excitement of a conference from conference data recorded including audio and / or video, and is adapted to the progress of the conference from the conference data. A feature amount as a time series is extracted, and based on the feature amount, a climax calculation unit that calculates a climax as a time series along the progress of the meeting as an absolute evaluation value, and a climax as the absolute evaluation value A correction unit that corrects the degree to a degree of swell as a relative evaluation value, and the correction unit includes at least one of the units shown in the following (1) to (3), whereby the relative evaluation value It is characterized by correcting to the degree of swell.
(1) Comprehensive correction unit that corrects the degree of excitement over the entire meeting by considering the standard degree of excitement as the whole conference participant in the degree of excitement calculated by the excitement degree calculating unit. (2) By analyzing the topic, a topic that a part of the conference participants participated in the discussion is taken into account. (3) A personal correction unit that corrects the degree of excitement based on the nature of the participant and the content of the speech for each speech segment of the participant in the conference by analyzing the conference data

Further, the present invention is a conference analysis method for calculating a degree of excitement of a conference from conference data recorded including audio and / or video, and is characterized as a time series along the progress of the conference from the conference data. Extract the amount, and based on the feature amount, as an absolute evaluation value, a climax degree calculation stage for calculating the climax degree as a time series along the progress of the meeting, and the climax degree as the absolute evaluation value as a relative evaluation value And a correction stage for correcting the degree of excitement as the degree of excitement, and the correction stage includes at least one of the stages shown in the following (1) to (3), whereby the degree of excitement as the relative evaluation value It is characterized by correcting to.
(1) Comprehensive correction stage for correcting the degree of excitement over the entire meeting by taking into account the standard degree of excitement for the entire conference participants in the degree of excitement calculated in the step of calculating the degree of excitement (2) By analyzing the topic, a topic that a part of the conference participants participated in the discussion is taken into account. Sub-team correction stage for correction (3) Individual correction stage for correcting the degree of excitement based on the nature of the participant and the content of the speech for each speech section of each participant in the conference by analyzing the conference data

  Furthermore, the present invention is a program for causing a computer to function as the conference analysis apparatus.

  According to the present invention, the degree of excitement of the meeting calculated as an absolute evaluation is (1) the standard degree of excitement as a whole meeting participant, and (2) the topic in which some of the meeting participants participated in the discussion. Relative evaluation because correction is made based on at least one of the standard excitement level of the participant, (3) the nature of the participant for each participant's speech section and the content of the speech in the meeting. The degree of excitement can be obtained.

It is a functional block diagram of the meeting analysis device concerning one embodiment. It is a figure which shows typically the recording of the meeting by a meeting recording part, when a meeting is divided and implemented in two venues. It is a figure which shows the example of the climax degree which a climax calculation part calculates. It is a figure which shows the example of the conversation relationship estimated by the conversation relationship estimation part. It is a figure which shows the example of the adjustment value for correct | amending in a personal correction part as a table | surface.

  FIG. 1 is a functional block diagram of a conference analysis apparatus 10 according to an embodiment. The conference analysis apparatus 10 includes a conference recording unit 11, a climax calculation unit 12, and a correction unit 30.

  In the conference recording unit 11, the conference is recorded as audio and / or video with a microphone and a camera, and the excitement degree calculation unit 12 and the correction unit 30 (the total correction unit 31, the personal correction unit 33, and the conversation relation estimation unit 321 as conference data) ) To provide. If the conference is held separately at each remote venue, the conference recording unit 11 records the conference by recording audio and video at each venue in parallel.

  FIG. 2 is a diagram schematically showing the recording of the conference by the conference recording unit 11 when the conference is divided into two venues (location L1 and location L2). In the place L1, there are three people, Mr. A, Mr. B and Mr. C, as participants in the conference, and the situation is photographed by the camera C1 and recorded by the microphone M1. In the place L2, there are five people, D, E, F, G, and H, as participants in the conference. The situation is recorded by the camera C2 and recorded by the microphone M2.

  Communication between the locations L1 and L2 is possible through a speaker or a display (not shown in FIG. 2), and a total of eight people at both locations L1 and L2 can be communicated. As a whole, the conference proceeds. As shown in FIG. 2, the conference recording unit 11 records the progress of the entire conference by acquiring the video and audio of the cameras C1, C2 and microphones M1, M2 arranged at the respective locations L1, L2. .

  In the example of FIG. 2, a camera and a microphone are prepared for each meeting place, but may be prepared for each meeting participant. Even in such a case, the conference recording unit 11 acquires a record for each participant who participates in the same conference, and records the progress of the conference as a whole.

  As for audio, it is only necessary to acquire one data in which the entire conference is recorded. For example, if the microphone M1 records only the speech of the participant at the location L1, and the microphone M2 records only the speech of the participant at the location L2, the data obtained by mixing the recordings in time synchronization is acquired. Just do it. If the speech from the location L2 is also reproduced by the speaker at the location L1, the data recorded at the location L1 may be acquired. Regarding the video, the entire conference may be recorded separately for each venue or for each individual.

  Although not shown in FIG. 2, the conference recording unit 11 may further acquire biometric data for each participant as information constituting the conference data. In this case, each participant wears a sensor or device for acquiring biological data such as an electroencephalogram and a pulse, and information on the worn sensor or device is stored in the conference recording unit 11 (video or audio). Just like the case), it can be recorded in association with the conference time.

  Similarly, information from a device such as Kinect (registered trademark) or Google glass may be acquired as a sensor. These can handle depth data as output data in addition to audio and video, and can detect facial expressions, laughter and whispering from the output data.

  The excitement degree calculation unit 12 extracts each feature amount of audio and / or video (and / or biometric data, etc.) in which the conference is recorded on the time series, and uses the time series in accordance with the progress of the conference as the time series. Calculate the degree of excitement.

  Here, the excitement degree calculation unit 12 calculates the excitement degree of the meeting by a conventional technique. The calculated degree of excitement belongs to the absolute evaluation, as is clear from the fact that the degree of excitement is extracted from the feature quantities of the audio and video. Accordingly, the correction unit 30 adds corrections from various viewpoints to the degree of climax as the absolute evaluation, thereby obtaining the degree of climax as the corrected relative evaluation.

  As a configuration for the correction, the correction unit 30 includes a general correction unit 31, a subteam correction unit 32, a conversation relationship estimation unit 321, an individual correction unit 33, and a participant profile DB (database) 331. The respective parts are classified into those that perform correction from three viewpoints as follows (1) to (3).

  (1) The general correction unit 31 corrects the degree of excitement by considering the character of the entire team.

  (2) The subteam correction unit 32 (and the conversation relationship estimation unit 321) corrects the degree of excitement based on the activity of the subteam on the topic discussed in a certain time period.

  (3) The personal correction unit 33 (and the participant profile DB 331) corrects the degree of excitement according to the personality, role, or participation location of the participant as a personal factor.

  Each configuration classified into the above (1) to (3) can independently correct the degree of swell calculated by the swell degree calculation unit 12, but in the functional block diagram of FIG. A configuration in which all the corrections (1) to (3) are applied and applied in this order is shown. The order of application may be arbitrary. Moreover, the correction | amendment by only arbitrary parts may be applied among (1)-(3).

  Hereinafter, the details of each part of FIG. 1 after the climax calculation part 12 will be described. As described above, any part of (1) to (3) can be applied in any order. Hereinafter, as shown in the configuration of FIG. 1, (1) to (3) A case where correction is performed by applying all of the above in this order will be described as an example.

  In addition, each part (1)-(3) of the correction | amendment part 30 calculates | requires the coefficient for correct | amending the excitement degree calculated by the excitement degree calculation part 12 by analyzing meeting data, respectively. Here, for the overall correction unit 31 of (1), the coefficient for correction is calculated by the same embodiment as the degree of swell calculation unit 12 calculates the degree of swell. That is, if the climax calculation unit 12 calculates the climax using a specific type of feature amount, the general correction unit 31 also calculates the climax with the same type of feature amount, Calculate the coefficient. Accordingly, as will be described below, if the overall correction unit 31 directly corrects the climax degree calculated by the climax degree calculation unit 12, it is not necessary to calculate the climax degree again.

  On the other hand, the conversation relationship estimation unit 321 in (2) analyzes the conference data to estimate the conversation relationship, and the subteam correction unit 32 obtains information for correcting the degree of excitement from the conversation relationship. In this case, the embodiment of the conversation relation estimation unit 321 can use an item independent from the excitement degree calculation unit 12. That is, the feature amount extracted from the conference data may be independent from the excitement degree calculation unit 12.

  Similarly, the personal correction unit 33 of (3) analyzes the conference data and obtains information for correcting the degree of excitement. The embodiment at this time can use an independent one from the excitement degree calculation unit 12 (and the conversation relation estimation unit 321). That is, the feature amount extracted from the conference data may be independent from the excitement degree calculation unit 12 (and the conversation relation estimation unit 321).

  The excitement degree calculation unit 12 analyzes the recording (conference data) of the conference recording unit 11, and for each participant in the conference, utterance contents, facial expressions, voice features, laughing timing and duration, laughing sound strength, nodding Feature quantities related to video and audio such as timing and line of sight are extracted, and the degree of excitement of the meeting is calculated in time series using the feature quantities.

  When calculating the degree of excitement from the feature amount related to speech, for example, as shown in Non-Patent Document 1, feature amounts such as fundamental frequency F0 and energy are extracted from the speech data, and the extent of excitement is estimated in five stages for each fixed time window. be able to. As shown in the example in Fig. 2, when a remote conference of about 2 hours is held with 3 people at location L1 and 5 people at location L2, for example, video and audio data at location L1 and location L2 are recorded and excited. The degree can be calculated.

  FIG. 3 shows an example of the calculated time series of swells in a part of the conference in which there was an average of 2.14 at place L1 and then an average of 2.5 at place L2.

  In addition, the method of calculating the degree of excitement by analyzing video, the method of calculating the degree of excitement by analyzing both video and audio, and the method of calculating the degree of excitement by biometric data such as brain waves and pulse, The calculation unit 12 may be realized. For example, in the above-mentioned Non-Patent Document 2, three cases are compared with only audio, only video, and both audio and video. In the comparison, the speech features are speech energy, speech frequency, and speech pitch (speech energy, speaking rate, speech pitch). person motion, head angle) and two.

  The general correction unit 31 performs standardization to absorb the character of the entire team (the entire conference participants).

  That is, when the whole team is originally active and tends to speak loudly, the value of the degree of excitement calculated by the degree of excitement calculation unit 12 tends to increase overall regardless of the actual excitement. On the other hand, when it tends to be inactive, it tends to be smaller as a whole regardless of the actual excitement. The general correction unit 31 performs standardization to absorb such differences between teams.

  Here, in order to extract a state where the team is in an average activity, a scene of greeting in the meeting is used. Many people greet at the beginning of the meeting. Therefore, the general correction unit 31 cuts out the greeting time by voice or gesture.

  Specifically, as the target conference data (voice), the first predetermined time (e.g., 10 minutes) "Hello", the search for "cheers for good work." That is, a predetermined greeting word is searched using a well-known voice analysis technique. Or, search for bows and nods from conference data (video). That is, a predetermined greeting operation is searched using a well-known gesture analysis technique.

  When the greeting time is detected as described above, a predetermined time range such as one minute is defined as the greeting time zone centering on the detection time. And let the average excitement degree of the said greeting time zone be a reference excitement degree. If a plurality of greeting time zones are detected, the average of all of them may be used.

  If no greeting time can be detected, the reference excitement degree may be obtained by averaging the excitement degrees of the entire meeting (the entire time from the start to the end of the meeting).

  In the configuration of FIG. 1, the climax degree calculated by the climax degree calculation unit 12 is normalized as shown in the following formula (1) by the reference climax degree obtained by the general correction unit 31 as described above. This is called the total excitement degree.

  As described above, the reference excitement degree is obtained by applying the same embodiment as the excitement degree calculation unit 12 to the greeting time zone. Therefore, when the information of I (t) has already been obtained by the degree of excitement calculation unit 12, and this is corrected directly by the general correction unit 31, the I (t) is averaged as it is in the greeting time zone. It is only necessary to obtain the reference excitement level, and it is not necessary to analyze the meeting data again in the general correction unit 12.

  The sub team correction unit 32 performs team correction according to the conversational relationship. The conversation relation estimation unit 321 estimates the conversation relation in advance. Therefore, first, the conversation relationship estimation unit 321 will be described below.

  The conversation relation estimation unit 321 identifies a speaker by using a known speaker recognition technique for conference data (voice) or a face recognition technique for conference data (video), and estimates the conversation relation. FIG. 4 shows an example of the conversation relationship estimated by the conversation relationship estimation unit 321 with respect to the example of the meeting in the team as shown in FIG.

  FIG. 4 shows the following conversation relationship. Mr. B, shown by a bold frame, started a topic and interacted with Mr. D twice, as indicated by the line and the number given in the figure. Furthermore, Mr. D and Mr. E exchanged three times on the topic. Regarding the topic, Mr. E and Mr. H exchanged six times, and Mr. H and Mr. D performed one interaction. Other people shown in gray in the figure do not speak on the topic.

  As described above, the conversation relation estimation unit 321 estimates the conversation relation in the topic that has been started, and also obtains the time zone {t | t1 ≦ t ≦ t2} in which the conversation about the topic was made. The conversation relationship estimation unit 321 can be realized by the technique disclosed in Non-Patent Document 4 below.

  [Non-Patent Document 4] Michel Galley, Kathleen McKeown, Eric Fosler-Lussier, and Hongyan Jing. 2003. Discourse segmentation of multi-party conversation. In Proceedings of the 41st Annual Meeting on Association for Computational Linguistics-Volume 1 (ACL '03 ), Vol. 1. Association for Computational Linguistics, Stroudsburg, PA, USA, 562-569.

  As described above, the subteam correction unit 32 performs correction based on the conversation relationship estimated by the conversation relationship estimation unit 321 together with the time zone {t | t1 ≦ t ≦ t2}. Specifically, in a conversational relationship during the time period, the more the team participants (participants who participated in the discussion of the topic) and the number of conversations, the greater the value of the degree of climax naturally even if it is not actually swelled On the contrary, the degree of excitement in the time zone is normalized and lowered. At the same time, the more people who actually speak in the conversation during the time zone, the more likely it is that the activity will be lost due to the fact that it is divided into remote areas naturally. Try to make it bigger.

  Based on the above policy, the conversation relationship estimation unit 321 corrects the degree of excitement of the corresponding time zone for the estimated conversation time zone {t | t1 ≦ t ≦ t2} by the following equation (2), for example. Can do.

Here, I ^t (t) is the total excitement level at time t, M is the number of participants, R is the number of conversations, and P is the number of participants' participation places. In this description, since correction is performed with the configuration of FIG. 1, the total swell degree I ^t (t) is a correction target, and is a value I ^S (t) that has been corrected.

  For example, in the case of FIG. 3, the sub-teams from time t1 to t2 are B, D, E, and H. The number M of sub-team participants is 4. The number of conversations R is 12. Furthermore, the number P of participants' participation places is two places.

  The personal correction unit 33 corrects the degree of excitement by absorbing personal elements.

  In order to perform the correction, the role, personality, and participation place of the participant are acquired in advance and stored in the participant profile DB 331. In addition, the personal correction unit 33 determines the privacy of the speech content by analyzing the speech content analyzed by the excitement degree calculation unit 12 and / or the conversation relation estimation unit 321 in advance using a personal information detection technique. A well-known personal information detection technique can be used, and the speech content text is analyzed to determine whether there is a predetermined privacy-related term.

Based on the above preparation, when the participant x from the time t3 to t4 speaks, the personal correction unit 33 concerned the time range {t | t3 ≦ t ≦ t4} (all the relevant ranges that exist in plural) The adjustment value Mod (x) is added to the degree of swell as shown in the following formula (3), and the degree of swell I ^S (t) is further corrected to I ^P (t). .

  Here, the adjustment value Mod (x) is a value (both positive, negative, and zero) determined by the role, personality, participation location of the participant x and the privacy of the content of the participant x in the time range. The value is determined from the information prepared in advance.

FIG. 5 is an example of a table for calculating the adjustment value. For example, if Mr. A is “personality: outgoing”, “role: leader”, and participation place “remote”, and the content of the statement in that time zone is “privacy: yes”, then Mr. A ’s time The adjustment value in the band is calculated as follows.
Mod (A) = (-0.5) + (-0.5) + 0.5 + 0.5 = 0

  In addition, although the specific value of the adjustment value illustrated in the table of FIG. 5 is determined in advance, the value may be given according to the following policy.

  As for the personality, since the remarks are originally active if they are extroverted, the degree of excitement is adjusted, and vice versa if they are introverted. As for the role, since leaders are forced to speak frequently, the leaders are adjusted to reduce the level of excitement. As for the place of participation, if it is remote, it tends to be passive from the beginning, so adjust it to increase the excitement. As for the privacy of the content of speech, since talks with privacy tend to be passive from the beginning, adjustments are made to increase the excitement when there is privacy.

  Note that the remote location means a sub venue for the main venue. It is assumed that there are main members who lead the conference at the main venue, and that auxiliary participants are participating at the remote sub-site. If there is no such distinction and each venue is equal in terms of the aggressiveness of participating in the conference, the remote / main venue adjustment may be omitted.

  The participant profile DB 331 may be prepared in advance by manual input or the contents thereof may be automatically estimated. For automatic estimation, the following various conventional techniques can be used.

  In the automatic determination of the role of the participant, as a technique for determining by voice information, as shown in [Non-Patent Document 5] below, the role of a leader or the like is automatically estimated by a classifier such as AdaBoost as shown in [Non-Patent Document 5] It is possible.

  [Non-Patent Document 5] Neha P. Garg, Sarah Favre, Hugues Salamin, Dilek Hakkani Tur, and Alessandro Vinciarelli. 2008. Role recognition for meeting participants: an approach based on lexical information and social network analysis. In Proceedings of the 16th ACM international conference on Multimedia (MM '08). ACM, New York, NY, USA, 693-696.

  Moreover, there is the following [Non-Patent Document 6] as an example of a technique for automatically estimating the personality of a participant only with audio and video of a meeting. The technology uses a support vector machine.

  [Non-Patent Document 6] Bruno Lepri, Nadia Mana, Alessandro Cappelletti, Fabio Pianesi, Massimo Zancanaro, Modeling the Personality of Participants During Group Interactions, User Modeling, Adaptation, and Personalization, LNCS, Vol. 5535, pp. 114-125, 2009.

  In addition, the participation location of the participants can be estimated from Wi-Fi (registered trademark), GPS (Global Positioning System), etc., on the assumption that each participant holds a terminal or the like. For example, Google's “Google Location Service” and Kugit ’s “PlaceEngine” have similar databases and provide Wi-Fi positioning services.

  As described above, according to the conference analysis apparatus 10 of the present invention, the degree of excitement of the value as an absolute evaluation estimated by voice, facial expression, gesture, etc., depends on the individuality of the participant, the place of participation, the content of the statement, the speaker relationship, etc. From the sub-team to the individual, it is corrected in three stages, and is used as a relative evaluation value. Thereby, the excitement degree of the meeting can be estimated with higher accuracy.

  11 ... Meeting recording unit, 12 ... Swelling degree calculation unit, 30 ... Correction unit, 31 ... General correction unit, 32 ... Subteam correction unit, 33 ... Personal correction unit

Claims (6)

A conference analyzer that calculates the degree of excitement of a conference from conference data recorded including audio and / or video,
A feature amount as a time series along the progress of the conference is extracted from the conference data, and a degree of excitement is calculated based on the feature amount as a time series along the progress of the conference as an absolute evaluation value And
A correction unit that corrects the climax as the absolute evaluation value to the climax as the relative evaluation value,
The said correction | amendment part correct | amends to the excitement degree as said relative evaluation value by providing at least 1 each part shown to each of the following (1)-(3), The meeting analyzer characterized by the above-mentioned.
(1) Comprehensive correction unit that corrects the degree of excitement over the entire meeting by considering the standard degree of excitement as the whole conference participant in the degree of excitement calculated by the excitement degree calculating unit. (2) By analyzing the topic, a topic that a part of the conference participants participated in the discussion is taken into account. (3) A personal correction unit that corrects the degree of excitement based on the nature of the participant and the content of the speech for each speech segment of the participant in the conference by analyzing the conference data   The said total correction | amendment part extracts the area containing a greeting from the predetermined area after a meeting start, and correct | amends by normalizing by the standard excitement degree in the said extracted area, The correction | amendment of Claim 1 characterized by the above-mentioned. Conference analyzer.   The conference according to claim 1, wherein the sub-team correction unit corrects based on the number of participants, the number of conversations, and the number of places of participation in the section in which the topic is discussed. Analysis equipment.   The personal correction unit corrects for each participant's speech section in the conference based on the personality of the participant, the role in the conference, the place of participation, and the presence or absence of privacy of the speech. Item 4. The meeting analyzer according to any one of Items 1 to 3. A meeting analysis method for calculating a degree of excitement of a meeting from meeting data recorded including audio and / or video,
A feature amount as a time series along the progress of the conference is extracted from the conference data, and a degree of excitement is calculated based on the feature amount as a time series along the progress of the conference as an absolute evaluation value Stages,
A correction stage for correcting the degree of swell as the absolute evaluation value to the degree of swell as a relative evaluation value,
The conference analysis method according to claim 1, wherein the correction stage includes at least one of the stages shown in the following (1) to (3) to correct the degree of excitement as the relative evaluation value.
(1) Comprehensive correction stage for correcting the degree of excitement over the entire meeting by taking into account the standard degree of excitement for the entire conference participants in the degree of excitement calculated in the step of calculating the degree of excitement (2) By analyzing the topic, a topic that a part of the conference participants participated in the discussion is taken into account. Sub-team correction stage for correction (3) Individual correction stage for correcting the degree of excitement based on the nature of the participant and the content of the speech for each speech section of each participant in the conference by analyzing the conference data   A program for causing a computer to function as the conference analyzer according to any one of claims 1 to 4.