JP2010002973A - Voice data subject estimation device, and call center using the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice data subject estimation device which estimates the subject of voice data from the voice data for updating relevance and keywords stored in a database, and to provide a call center using the voice data subject estimation device. <P>SOLUTION: The voice data subject estimation device 1 includes: a subject database 2 in which a subject, a keyword and relevancy are stored; a determination part 10 for acquiring a keyword (extracted keyword) extracted from voice recognition data obtained by recognizing voice of voice data and an estimated subject (subject A) estimated from the voice recognition data, acquiring a keyword (keyword B) related with the subject A by referring to the subject database 2, and determining whether or not the extracted keyword is matched with the keyword B; a subject DB registration part 11 for registering the keyword related with the subject A stored in the subject database 2 based on the determination result; and a subject DB update part 12 for updating the relevancy of the keywords related with the subject A. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech data subject estimation device for estimating a subject of conversation content stored as speech data, and a call center using the speech data subject estimation device.

  When recording call recordings that record the content of conversations during a call, or voice mails of extension telephone exchanges that leave messages on the phone when the other party is absent It is necessary to have information such as themes and keywords that can accurately determine whether or not.

Japanese Patent Application Laid-Open No. 2004-151867 discloses a recorded data subject detection device that allows a user to easily know the subject (business) of recorded data.
Conventionally, in order to obtain information that can be used to determine what kind of business the recorded data is, the recorded data is voice-recognized, converted into document data, sentence data, and pre-registered keywords And the number of appearances of the keyword is detected. Based on the detected number of occurrences and the importance of the keyword registered in advance, it is determined whether or not the keyword is suitable for the subject. All keywords judged to be output as the subject of the recording data.

Japanese Patent Laying-Open No. 2004-96502 (paragraphs 0010 to 0014, FIG. 1)

In the conventional subject detection device, “keywords” extracted from document data obtained by converting voice data are registered in the database in advance, and means for adding keywords to the database and the importance (relevance) of keywords are determined. There is no way to update.
Therefore, it has been difficult to increase the extraction accuracy indicating whether or not keywords suitable for the subject have been extracted. Therefore, as the use period of the subject detection apparatus becomes longer, the extraction accuracy decreases, and there is a tendency for more important keywords that are not registered in the database to be able to cope with changes in the value of importance. Therefore, the subject estimated from the extracted keyword may not match the subject of conversation.

  Moreover, even if the importance of the keyword is changed and the extraction accuracy is adjusted, a large amount of audio data is required to confirm whether the adjustment is appropriate.

  In addition, since the output result of the subject estimated by the conventional subject detection device is merely a list of keywords that are keywords and not sentences, the user needs to associate the subject with the output words. It was. For this reason, there are some cases where the subject matter is different depending on the interpretation of the user by estimating a keyword that is not suitable for the subject matter.

  Therefore, the present invention provides an audio data subject estimation device capable of updating the relevance and keywords stored in a database in a device for estimating the subject of the speech data from the speech data, and a call center using the speech data subject estimation device. For the purpose.

  In order to achieve the object, a speech data subject estimation device according to claim 1 includes a data storage unit storing the speech data, and a speech recognition unit that recognizes the speech data and generates speech recognition data. A vocabulary database that stores vocabulary posted in the dictionary, a character string of the speech recognition data, a keyword extraction unit that extracts keywords that match the vocabulary stored in the vocabulary database, and a data subject , A theme database storing a keyword related to the theme, a degree of relevance obtained by quantifying the relationship between the theme and the keyword, and an extracted keyword extracted by the keyword extraction unit with reference to the theme database; A subject estimation unit that assumes the subject having the highest degree of relevance as an estimated subject, and a keyword related to the estimated subject by referring to the subject database. A determination unit that determines whether or not the acquired subject keyword (keyword B) matches the extracted keyword, and a mismatch keyword that is determined to be unmatched by the determination unit is used as the estimated subject. The keyword DB includes a subject DB registration unit that registers in the subject database, and a subject DB update unit that updates the degree of association of the subject keyword determined to match by the determination unit.

  In the speech data theme estimating apparatus according to claim 2, the theme estimating unit determines whether or not the input unit to which the input theme is input and the input theme and the estimated theme match, And a subject determination unit as an estimated subject (theme A).

  In the speech data theme estimating apparatus according to claim 3, the theme database stores a number of samples that is a total number of histories in which the input theme matches the estimated theme in the theme determination unit, and the subject determination unit includes: When the determination results match, the subject database is referred to and 1 is added to the number of samples related to the estimated subject, and the subject estimation unit is stored in the subject database and the character string of the speech recognition data. The keywords that match the keyword are extracted, the subject related to the extracted keyword and the number of samples of the subject are obtained, the relevance of the keywords related to the subject is added together, and the number of samples of the subject is further calculated. It is characterized in that the subject having the highest average relevance divided is the estimated subject.

  The speech data theme estimation apparatus according to claim 4, wherein the theme determination unit further refers to the theme database and determines whether or not the estimated theme matches a theme registered in the theme database. A summary keyword extraction unit that extracts a keyword that matches the character string of the summary data input from the input unit and the vocabulary stored in the term database when the term database is not matched The subject DB new registration unit that registers the estimated subject and the summary keyword extracted by the summary keyword extraction unit in the subject database.

  A call center according to a fifth aspect is characterized in that the speech data subject estimation device is provided.

  According to the present invention, since the degree of relevance and keywords stored in the database can be updated in the apparatus for estimating the subject of the voice data from the voice data, the optimum subject for the voice data is estimated. It is possible to provide a speech data subject estimation apparatus capable of performing the above and a call center using the same.

  Next, embodiments of the present invention will be described in detail with reference to the drawings as appropriate.

(First embodiment)
FIG. 1 is a configuration diagram of an audio data subject estimation device according to the first embodiment of the present invention.
As shown in FIG. 1, the speech data theme estimation device 1 according to the first embodiment includes a theme database 2, a term database 3, and a data storage unit 4, and more specifically, a random access memory (RAM). ), ROM (read-only memory), HDD (hard disk drive) and the like.
As shown in FIG. 1, the speech data subject estimation device 1 according to the first embodiment includes a speech recognition unit 5, an input unit 6, a subject estimation unit 7, a subject determination unit 8, a keyword extraction unit 9, a determination unit 10, A subject DB registration unit 11 and a subject DB update unit 12 are provided, and specifically, a control device such as a CPU (central processing unit) is configured.
Furthermore, as shown in FIG. 1, the speech data subject estimation device 1 according to the first embodiment includes an input unit 6, and specifically includes an input device such as a keyboard and a mouse, and an output device such as a monitor. Is done.
The speech data subject estimation device 1 according to the first embodiment includes hardware resources such as the control device, the storage device, the input device, and the output device, and a software program including an operating system (OS) and applications. And realized.

  The subject database 2 stores a subject of data registered in advance, a keyword related to the subject, a degree of association indicating the relationship between the subject and the keyword by a numerical value, and the number of samples (details will be described later). Has been.

  The term database 3 stores vocabularies to be published in national language dictionaries and encyclopedias. Then, a new vocabulary can be added to the term database 3 as necessary.

The data storage unit 4 stores audio data.
The audio data is mainly digital data in which a human voice is recorded by a recording device such as a voice recorder, a cassette tape recorder, or a mini disk (MD) player with a recording function.

  The input unit 6 is a device to which data is input, and an image displayed on a monitor screen is selected with a mouse or the like using a direct input from a PC (Personal Computer) keyboard or a GUI (Graphical User Interface). By doing so, data is input. Alternatively, the monitor may be of a touch panel type, and touch the touch panel to select characters and input data.

  The voice recognition unit 5 acquires voice data from the data storage unit 4, performs voice recognition processing, and generates voice recognition data. In the voice recognition process, the generated voice recognition data is preferably text data.

  The subject estimation unit 7 obtains speech recognition data from the speech recognition unit 5, collates a character string of the speech recognition data with a keyword stored in the subject database 2, and extracts a matching keyword (collation keyword). . Then, referring to the subject database 2, the subject relating to the matching keyword is extracted. Next, the subject estimation unit 7 refers to the subject database 2 and acquires the number of samples of each extracted subject. Then, the extracted degrees of relevance of the extracted keywords are added to each extracted theme, the combined degrees of association are divided by the number of samples, and an estimated value of each extracted theme is calculated. Finally, the subject estimation unit 7 sets the subject having the largest estimated value as the estimated subject.

  The subject determination unit 8 determines whether or not the data input from the input unit 6 (input subject) matches the data input from the subject estimation unit 7 (estimated subject). The matched data (matched subject = input subject = estimated subject) is sent to the determination unit 10. If they match, the subject database 2 is referenced to search for the subject registered in the database 2 that matches the matching subject, and “1” is added to the number of samples related to the subject. That is, the number of samples is the total number of histories in which the input subject and the estimated subject match so far in the subject determination unit 8.

  The keyword extraction unit 9 acquires the speech recognition data from the speech recognition unit 5 and collates the character string of the speech recognition data with the vocabulary stored in the term database 3 to extract a matching vocabulary (extracted keyword). Process.

  The determination unit 10 acquires data (matching theme) sent from the theme determination unit 8, refers to the subject database 2, searches for a subject that matches the matching subject, and extracts a subject keyword related to the subject (matching subject). (Keyword B) is acquired, and it is determined whether or not the subject keyword matches the extracted keyword extracted by the keyword extraction unit 9, and the determination result (a matching keyword or a keyword that does not match) is transmitted.

  The subject DB registration unit 11 acquires the extracted keyword (mismatch keyword) sent from the determination unit 10. Then, with reference to the subject database 2, the mismatch keyword is additionally registered as a keyword related to the subject (matched subject) of the estimated voice data.

  The subject DB update unit 12 acquires a matching extracted keyword (matching keyword) sent from the determination unit 10. Then, the subject DB update unit 12 refers to the subject database 2 and updates the relevance level of the matching keyword related to the subject (matching subject) of the estimated voice data.

  Hereinafter, with reference to the flowcharts shown in FIG. 2 and FIG. 3, the speech data subject estimation device 1 according to the first embodiment estimates the subject of speech data and updates the subject database 2 used for the estimation. explain.

The user stores the voice data in the data storage unit 4 of the voice data subject estimation device 1 in advance (FIG. 2, S101).
The voice recognition unit 5 acquires voice data from the data storage unit 4, performs voice recognition processing, and generates voice recognition data (S102).
The subject estimation unit 7 acquires speech recognition data from the speech recognition unit 5 (S103). Further, the subject estimation unit 7 estimates the subject of the audio data based on the subject, the keyword, and the degree of association stored in the subject database 2 (S104).

The details of the process of estimating the subject of the speech data shown in S104 will be described using an example of the subject database 2 and speech data (speech recognition data) shown in FIG. 4 and the flowchart shown in FIG.
The subject database 2 shown in FIG. 4 stores information on the subject, the keyword, the degree of association, and the number of samples. "Submission questions", "Transfer confirmation", "Account creation questions", "Address change questions", etc. are registered in the subject, and keywords related to these subjects are "cancellation", "procedure", "branch", "Account" etc. are registered. Each keyword has a numerical value indicating the degree of association with the subject. For example, the degree of association between the theme “cancellation question” and the keyword “cancellation” is “30”. The relevance “-” indicates that the registration is not registered. For example, the keyword “cancel” is not registered in the subject “confirmation of transfer”. If the relevance value is “0” or more, it indicates that the keyword is registered in the subject.

The subject estimation unit 7 compares the character string of the speech recognition data (“I want to cancel the account at Toranomon branch”) with the keyword registered in the subject database 2 and extracts a matching keyword (matching keyword). (S201). In this case, collation keywords “cancellation”, “procedure”, “branch”, and “account” registered in the subject database 2 shown in FIG. 4 are extracted.
Then, the subject estimation unit 7 extracts the subject related to the matching keyword from the subject database 2 (S202). In this case, the subjects of “cancellation question”, “transfer confirmation” and “account creation question” are extracted.

Next, the subject estimation unit 7 refers to the subject database 2 and acquires the number of samples of each extracted subject (S203). In this case, the number of samples “30” for “cancellation question”, “22” for “transfer confirmation”, and “20” for “account creation question” is acquired.
The subject estimation unit 7 adds the relevance levels of the matching keywords to each extracted subject, divides by the number of samples, and calculates an estimated value of each extracted subject (S204).

<1. Cancellation question>
Sum of relevance = Cancellation: 30+ Procedure: 22+ Branch: 0+ Account: 11 = 63
Estimated value = 63 ÷ number of samples: 30 = 2.1
<2. Confirmation of transfer>
Sum of relevance = Cancellation: None + Procedure: 16 + Branch: 8 + Account: 2 = 26
Estimated value = 26 ÷ number of samples: 22 = 1.2
<3. Account creation questions>
Total relevance = Cancellation: None + Procedure: 12 + Branch: 6 + Account: 18 = 36
Estimated value = 36 ÷ number of samples: 20 = 1.8

Then, the subject estimation unit 7 sets the “cancellation question” having the largest estimated value as the estimated subject (S205).
Again, following S104 in FIG. 2, the processing will be described using the flowchart shown in FIG.
The subject determination unit 8 acquires data (input subject) input by the user via the input unit 6 and the subject (estimated subject) estimated from the subject estimation unit 7 (FIG. 3, S105). For example, the input subject “cancellation question” and the estimated subject “cancellation question” are acquired.

Next, the subject determination unit 8 determines whether or not the input subject matches the estimated subject (S106).
If they do not match (S106, No), the speech data subject estimation device 1 ends the series of processes.
On the other hand, if they match (S106, Yes), the subject determination unit 8 refers to the subject database 2 for the subject that matches the subject A (input subject = estimated subject = matched subject (subject A)). The subject registered in the database 2 is searched, and “1” is added to the number of samples related to the subject (S107). For example, “1” is added to the sample number “30” of the subject A “cancellation question” to make “31”.
Then, the subject determination unit 8 sends the subject A to the determination unit 10 (S108).

The keyword extraction unit 9 acquires voice recognition data from the voice recognition unit 5 (S109). For example, the voice recognition data “Account Cancellation Procedure at Toranomon Branch” is acquired.
Then, the keyword extraction unit 9 compares the character string of the speech recognition data with the vocabulary stored in the term database 3 and extracts a matching vocabulary (extracted keyword) (S110). For example, “branch”, “account”, “cancellation”, and “procedure” are extracted.

  The determination unit 10 includes the subject A (“cancellation question”) sent from the subject determination unit 8 in S108 and the vocabulary (extracted keyword group: “branch”, “account”) extracted by the keyword extraction unit 9 in S110. , “Cancellation”, “procedure”) (S111). Then, the determination unit 10 refers to the subject database 2 and acquires the keyword group B (“account”, “cancellation”, “procedure”, “change”) related to the subject A (S112).

Next, the determination unit 10 acquires the acquired keyword group B (“account”, “cancellation”, “procedure”, “change”) and the extracted keyword group (extracted in S110) (“branch”, “account”, It is determined whether or not there is an extracted keyword (mismatch keyword X) that does not match the keyword included in the keyword group B in the extracted keyword group (S113).
<Judgment result>
Unmatched extracted keyword (mismatched keyword X) = “branch”
Matching extracted keyword (matching keyword Y) = "account", "cancellation", "procedure"

If there is a mismatch keyword X (S113, Yes), the mismatch keyword X is sent to the subject DB registration unit 11 (S114). The subject DB registration unit 11 registers the acquired mismatch keyword X in the subject database 2 as a keyword related to the subject A with a relevance “0” (S115). As a result, “branch” is registered with the relevance “0” as a keyword related to the subject A (“cancellation question”) in the subject database 2 (FIG. 4 shows this state).
After registration, the process returns to S113, and it is determined whether or not there is a mismatch keyword X that does not match the keyword B.

On the other hand, when there is no mismatch keyword X (S113, No), the match keyword Y is sent to the subject DB update unit 12 (S116). The subject DB update unit 12 adds “1” to the relevance level of the keyword B related to the subject A that matches the acquired matching keyword Y (S117). As described above, the relevance level of each keyword of the theme (theme A “cancellation question”) registered in the theme database shown in FIG. 4 is “30 → 31” for “cancellation” and “22 → 23” for “procedure”. ”And“ Account ”becomes“ 11 → 12 ”.
Then, the theme DB update unit 12 updates the theme, keyword, and degree of association in the theme database 2 (S118). And the audio | voice data theme estimation apparatus 1 complete | finishes a process.

  According to the first embodiment, since the degree of association between a registered subject and a keyword related to the subject can be increased in the subject database 2 included in the speech data subject estimation device 1, the speech data subject estimation device 1 It is possible to estimate a subject having a high degree of relevance to the keyword extracted from the speech data. That is, a subject suitable for audio data is estimated.

(Second Embodiment)
As shown in FIG. 6, the speech data theme estimating device 1a according to the second embodiment of the present invention is further provided with a summary keyword extraction in addition to the configuration of the speech data theme estimating device 1 according to the first embodiment (FIG. 1). The function is added to the subject determination unit 8a by adding the unit 13 and the subject DB new registration unit 14.
The input unit 6 receives input data and summary data. The input data includes the subject of the audio data (input subject), and the summary data includes information related to the subject.

  The subject determination unit 8a determines whether or not the input subject (subject) acquired from the input unit 6 matches the estimated subject (subject) acquired from the subject estimation unit 7. In the case of a match, the subject database 2 is further referred to to determine whether or not a matching subject (input subject = estimated subject = matched subject) is registered in the subject database 2. If registered, the registered theme (matching theme) is sent to the determination unit 10. If not registered, the unregistered theme (matching theme) is sent to the summary keyword extraction unit 13.

The summary keyword extraction unit 13 acquires an unregistered subject from the subject determination unit 8a. Furthermore, a process of acquiring summary data from the input unit 6 and extracting a vocabulary (summary keyword) that matches the vocabulary included in the summary data by referring to the vocabulary described in the dictionary stored in the term database 3. Do.
The new theme DB registration unit 14 acquires the unregistered subject and the overview keyword from the overview keyword extraction unit 13, and performs processing for registering the unregistered subject in the subject database 2 as the subject and the overview keyword as the keyword related to the subject. .

  As described above, the speech data theme estimation device 1a according to the second embodiment includes the summary keyword extraction unit 13 and the theme DB new registration unit 14, so that the theme database 2 has a theme and its theme. The keyword concerning can be registered.

Below, the process which the audio | voice data theme estimation apparatus 1a which concerns on 2nd Embodiment registers the theme and the keyword which concerns on the theme in the theme database 2 using the flowchart shown in FIG. 7 is demonstrated.
Since the process performed by the subject estimation unit 7 is the same as the process of S101 to S104 (FIG. 2), description thereof is omitted.

  The subject determination unit 8a acquires data (input subject) input by the user via the input unit 6 and the subject (estimated subject) estimated from the subject estimation unit 7 (S301 in FIG. 7). For example, an input subject “ATM location question” and an estimated subject “ATM location question” are acquired.

Next, the subject determination unit 8a determines whether or not the input subject matches the estimated subject (S302).
If they do not match (S302, No), the speech data subject estimation device 1 ends the series of processes.
If they match (S302, Yes), the subject determination unit 8a refers to the subject database 2 to search for a subject that matches the matching subject (input subject = estimated subject), and determines whether the matching subject is registered. Determination is made (S303).

  When the matching subject is registered (S303, Yes), the subject determination unit 8a adds “1” to the number of samples related to the registered subject (matching subject (theme A)) registered in the subject database 2 ( S304). Then, the subject determination unit 8a sends the subject A to the determination unit 10 (S305). Since the subsequent processing (S306) is the same as the processing of S111 to S118 (FIG. 3), description thereof is omitted.

On the other hand, when the matching subject is not registered (S303, No), the subject determination unit 8a sends the unregistered subject (matching subject) to the overview keyword extraction unit 13 (S307).
The summary keyword extraction unit 13 acquires the unregistered subject sent from the subject determination unit 8a and the summary data input by the user via the input unit 6 (S308). For example, unregistered subject "ATM location question" and summary data "Where are the ATMs installed in the branch near Toranomon station?" ”Is acquired.

Then, the summary keyword extraction unit 13 collates the character string of the summary data with the vocabulary stored in the term database 3 and extracts a matching vocabulary (summary keyword) (S309). For example, “station”, “branch”, “installation”, and “ATM” are extracted.
Next, the unregistered subject and the extracted vocabulary (summary keywords: “station”, “branch”, “installation”, “ATM”) are sent to the subject DB new registration unit 14 (S310).

  The new theme DB registration unit 14 acquires an unregistered theme ("ATM location question") and summary keywords ("station", "branch", "installation", "ATM") (S311). Then, the new theme DB registration unit 14 registers the unregistered theme as the theme and the overview keyword as the keyword related to the theme with the relevance “0” in the theme database 2 (S312). After registration, the speech data subject estimation apparatus 1 ends the process.

  According to the second embodiment, an unregistered theme and a keyword related to the theme can be added and registered in the theme database 2 included in the speech data theme estimation apparatus 1a.

(Call center with speech data subject estimation device)
The Example which used the audio | voice data subject estimation apparatus which concerns on embodiment of this invention for the call center of a bank is shown.

FIG. 8 is a configuration diagram of a call center including the voice data subject estimation device 1. The bank call center 100 includes a CT server 101, a plurality of operator terminals 102, and a plurality of telephones 103.
The CT server 101 and the operator terminal 102 are connected via a communication line such as a LAN 104, and transmit and receive data via the communication line.
The telephone 103 may be a desk phone or, for example, by using an IP (Internet Protocol) telephone, the operator terminal 102 connected to the headset may have a telephone function as shown in FIG.

  The CT server 101 includes the speech data subject estimation devices 1 and 1a according to the embodiment. In addition, the CT server 101 includes telephone exchange means for distributing calls received from the user 300 to each operator, means for displaying information necessary for monitoring each operator terminal 102, means for storing the information, and each operator terminal. A configuration necessary for operating the call center, such as a means for storing information input from 102, is provided.

The operator terminal 102 includes a PC, an input / output device such as a monitor, a keyboard, and a mouse, and a recording device such as a voice recorder. The input device of the operator terminal 102 corresponds to the input unit 6, and voice data recorded by the recording device is stored in the data storage unit 4 in the CT server 101 via the LAN 104.
When the operator 105 receives a call from the user 300, the operator terminal 102 stores a conversation between the user 300 who made the call and the operator 105 in a recording device provided in the operator terminal 102.

With the above configuration, the speech data subject estimation device 1 (FIG. 1) can be used in a bank call center.
For example, when there is a question from the user 300 “I want to cancel my account at the Toranomon branch” over the phone, the operator will use the content of the phone call as “subject question of cancellation” and the summary data as “Toranomon branch”. Enter “Account Cancellation Procedure”.
Further, the speech data subject estimation device 1 estimates the subject “cancellation question” from the speech data “I would like to cancel the account at Toranomon branch” (FIG. 5, S205).
As a result of the above processing, the speech data subject estimation device 1 adds the keyword “branch” related to the subject “cancellation question” registered in the subject database 2 and the keywords “account”, “cancellation”, “procedure”. ”Can be updated.
As a result, the “cancellation question” is more relevant to the keywords “account”, “cancellation”, and “procedure”, and the probability of estimating the “cancellation question” as the subject for the voice data containing the keyword. Can be increased.
Therefore, the speech data subject estimation device 1 can increase the accuracy of correctly estimating the subject with respect to the question from the user 300 by this processing.

  When inputting a new theme and summary data to the CT server 101 including the speech data theme estimating apparatus 1 via the operator terminal 102, the input unit 6 may use a keyboard, a mouse, or the like. 105 operation logs may be used.

The case where the operation log of the operator 105 is used for the input unit 6 is shown.
All the screens displayed on the monitor of the operator terminal 102 are assigned one title and one or more vocabularies per screen.
Thereby, the operator 105 can set the title given to the screen displayed for the longest time during the dialogue with the customer as the subject (input subject) input to the input unit 6.
Then, the vocabulary (summary keyword) extracted by the outline keyword extraction unit 13 when the operator 105 is the speech data subject estimation device 1a of the second embodiment, which is given to the entire screen displayed during the dialogue with the customer. ). Then, using the operation log of the operator 105, the subject determination unit 8a can acquire an input subject, and the subject DB new registration unit 14 can acquire a summary keyword.
Further, by using the operation log of the operator 105, the relevance of each extracted keyword can be determined according to the total time displayed on the screen during the dialogue.
For example, “0” if the display time is less than 30 seconds, “1” if it is 30 seconds or more and less than 1 minute, “2” if it is 1 minute or more and less than 5 minutes, and “3” if it is 5 minutes or more. Etc. Then, during the process of S114 (FIG. 3), the subject DB update unit 12 can add these relevance values to the relevance level of the keyword B related to the subject A.
As described above, by using the operation log of the operator as the input means, the operator can update the theme database 2 using the input device without inputting the theme.

  Since the present invention can add keywords to the database, change the degree of association between the subject and the keyword, or newly register the subject and the keyword even if the period of use of the subject estimation device becomes longer, A database for estimating the optimal subject can be provided.

  The following effects can be obtained by using the present invention for a system that frequently updates a theme database, a system that frequently adds new functions, and a system that frequently updates data.

As mentioned above, although embodiment of this invention was described, this invention is not limited to this, It can implement in the range which does not change the meaning.
For example, by providing the PBX (telephone exchange) with the speech data subject estimation device 1, the system can automatically recognize the contents of telephone questions. Therefore, by adding a database that associates the contents of a question with the department or person who can answer the question, a system that forwards the call to the department or person who can answer the question according to the contents of the caller's telephone question. Can be made. This frees the caller from being traversed.

  For example, by providing the voice data subject estimation device 1 in an ATM (automatic cash dispenser), the system can automatically recognize a desired process simply by an operator speaking into a microphone attached to the ATM. Therefore, it is freed from the troublesome operation of touching the screen until the desired processing is completed by operating the ATM touch panel.

It is a block diagram of the audio | voice data subject estimation apparatus which is the 1st Embodiment of this invention. It is a flowchart which shows the process sequence of the audio | voice data subject estimation apparatus which is the 1st Embodiment of this invention (the first half of a process). It is a flowchart which shows the process sequence of the audio | voice data theme estimation apparatus which is the 1st Embodiment of this invention (the process latter half). It is an example of a theme database and input data. It is a flowchart which shows the process sequence which estimates the theme of audio | voice data. It is a block diagram of the audio | voice data subject estimation apparatus which is the 2nd Embodiment of this invention. It is a flowchart which shows the process sequence of the audio | voice data theme estimation apparatus which is the 2nd Embodiment of this invention. It is the block diagram which used the audio | voice data subject estimation apparatus which concerns on embodiment of this invention for the call center of a bank.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Speech data theme estimation apparatus 2 Theme database 3 Term database 4 Data storage part 5 Speech recognition part 6 Input part 7 Theme estimation part 8 Theme determination part 9 Keyword extraction part 10 Judgment part 11 Theme DB registration part 12 Theme DB update part 13 Outline Keyword extraction unit 14 Subject DB new registration unit 100 Call center 101 CT server 102 Operator terminal 103 Telephone 104 LAN
105 Operator 200 Public network 300 User

Claims (5)

A speech data theme estimation device that recognizes speech data and estimates a theme of the speech data from extracted keywords,
A data storage unit storing the audio data;
A voice recognition unit that recognizes the voice data and generates voice recognition data;
A terminology database that stores vocabulary that appears in the dictionary;
A keyword extraction unit for extracting a keyword in which the character string of the speech recognition data matches the vocabulary stored in the term database;
A theme database storing data themes, keywords related to the themes, and degrees of relevance in which relations between the themes and the keywords are quantified;
A subject estimation unit that refers to the subject database, and uses a subject that is most relevant to the extracted keyword extracted by the keyword extraction unit as an estimated subject;
A determination unit that refers to the subject database, acquires a keyword related to the estimated subject, and determines whether the acquired subject keyword matches the extracted keyword;
A subject DB registration unit for registering inconsistent keywords determined to be inconsistent by the determining unit as keywords related to the estimated subject in the subject database;
A subject matter DB update unit that updates the degree of association of the subject keywords determined to be matched by the determination unit;
A speech data subject estimation device comprising: The subject estimation unit includes:
An input section in which an input subject is input;
Determining whether or not the input subject and the estimated subject match, and a subject determination unit having the matched subject as an estimated subject;
The speech data subject estimation device according to claim 1, comprising: The subject database stores the number of samples that is the total number of histories in which the input subject and the estimated subject match in the subject determination unit,
When the determination result is matched, the subject determination unit refers to the subject database and adds 1 to the number of samples related to the estimated subject.
The subject estimation unit extracts a keyword in which a character string of the speech recognition data matches a keyword stored in the subject database, and acquires a subject related to the extracted keyword and the number of samples of the subject. 3. The speech data subject estimation according to claim 2, wherein the relevance of keywords related to the subject is added together, and the subject having the highest average relevance divided by the number of samples of the subject is taken as the subject of estimation. apparatus. The subject determination unit further refers to the subject database to determine whether the estimated subject matches a subject registered in the subject database.
When there is no match, referring to the term database, a summary keyword extraction unit that extracts a keyword that matches the character string of the summary data input from the input unit and the vocabulary stored in the term database;
A subject DB new registration unit for registering the estimated subject and the summary keyword extracted by the summary keyword extraction unit in the subject database;
The speech data subject estimation apparatus according to claim 2, further comprising:   A call center comprising the speech data subject estimation device according to any one of claims 1 to 4.

Images