KR100822170B1 - Database construction method and system for speech recognition ars service - Google Patents

Abstract

A method and a system for constructing a database for a speech recognition ARS(Automatic Response System) service are provided to enable a user to receive a desired service immediately or select a service within a narrow range, thereby reducing burden of searching the desired service in comparison with a DTMF(Dual Tone Multi Frequency) method. A database constructing method for a speech recognition ARS service comprises the following steps of: collecting service inquiry speech from plural users(S403); converting the collected service inquiry speech of each user into a text(S405); grouping the converted texts according to similar service items, and setting a service tag to the grouped group(S409); and mapping the set service tag with a destination(S411).

Description

Database Construction Method and System for Speech Recognition Ars Service TECHNICAL FIELD

1 is a diagram illustrating an example of an ARS service using a conventional DTMF.

2 is a block diagram of a system for constructing a speech recognition database for a speech recognition ARS service according to an exemplary embodiment of the present invention.

3A is a diagram illustrating a process of converting a user's speech into text in a transcription unit according to an embodiment of the present invention.

3B is a diagram illustrating a process of grouping semantic tags into services having similar tendencies.

4 is a flowchart illustrating a method for establishing a database for a voice recognition ARS service according to an embodiment of the present invention.

5 is a diagram illustrating a configuration of an ARS system using speech recognition according to an embodiment of the present invention.

<Explanation of symbols for the main parts of the drawings>

210: data collection unit 230: transcription unit

250: tagging unit 270: destination setting unit

250: tag DB 271: tag-destination DB

The present invention relates to an ARS service, and more particularly, to a method and system for constructing a database for an ARS service.

ARS (Automatic Response System) is a system that stores various information by voice and tells the user how to search the necessary information by voice when the user connects to the system by phone. .

Currently, these ARS services are provided in the DTMF tree structure menu. That is, when a user accesses the system by using a telephone, the user guides the menu by voice so as to search for necessary information, and when the user inputs a desired menu number, the user finally approaches the desired information. However, the user is not familiar with the DTMF tree structure menu. It is not possible to determine exactly which path to go to in order to move to the desired service, and in many cases, the user goes through several trials and errors to move to the desired service.

1 is a diagram illustrating an example of an ARS service using a conventional DTMF, as shown in FIG. 1, a user accesses an ARS system by telephone in order to obtain service information of 'BI 2 R', and displays an initial DTMF service menu; That is, select "3. Plan / Additional Service" from "1. Usage fee, 2. Lost / Pause, 3. Plan / Additional Service, 4. Member's Card" and select "3.1 Plan" from the DTMF service menu. Choose. From the DTMF service menu provided again, select "3.1.1 Biggie" and then "3.1.1 Biggie 2200", and finally "3.1.1.2 Detailed Service Information" to arrive at the final route. That is, the desired information can be obtained only through a total of five steps.

As such, in the conventional ARS service, the ARS system presents a service example that should be selected to the user at each step as “1 usage fee, 2 lost and pauses.”, And the user selects the service that he / she wants. It waits and takes a lot of time to move to the final information.

As such, the user does not know the complicated menu structure and it takes a long time to move between services. Therefore, the user has to spend considerable time to move to a desired service. This inconvenience becomes a burden on the user, and for that reason, the user attempts to connect to an agent from the time of entry into the DTMF. Many agent connections lead to an increase in the number of agents, which incurs costs.

The present invention has been proposed to solve the above problems, and an object thereof is to provide a method and system for constructing a speech recognition database for providing an ARS service using speech recognition.

Other objects and advantages of the present invention can be understood by the following description, and will be more clearly understood by the embodiments of the present invention. Also, it will be readily appreciated that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the claims.

According to a first aspect of the present invention for achieving the above object, a method for constructing a database for a voice recognition ARS service, the method comprising: (a) collecting a service question voice from a plurality of users; (b) converting each collected service question voice into text; (c) grouping the converted text into similar service items and setting a service tag in the grouped group; And (e) mapping a destination to the set service tag.

Preferably, step (c) comprises: (c-1) setting a semantic tag in the converted text; And (c-2) grouping each set semantic tag for each similar service item and setting a service tag in the grouped group.

Also, preferably, the step (c-1) may include extracting a keyword from the converted text; Generating a semantic tag based on the extracted keyword; And mapping the extracted keyword to the generated semantic tag.

Also, preferably, the step (c-2) may include grouping each set semantic tag by similar service items; Setting a service tag in the grouped group; And mapping keywords of semantic tags belonging to the group to the service tag.

On the other hand, according to a second aspect of the present invention for achieving the above object, a system for building a database for speech recognition ARS service, collecting means for collecting a service question voice from a plurality of users; Conversion means for converting the collected service question voices of each user into text; Tagging means for grouping the converted text into similar service items and setting a service tag in the grouped group; And destination mapping means for mapping a destination to the service tag.

Preferably, the tagging means sets a semantic tag in the converted text, and sets the service tag by grouping each set semantic tag by similar service items.

Further, preferably, the tagging means extracts a keyword from the converted text and generates a semantic tag based thereon.

Also, preferably, the tagging means maps and stores keywords of semantic tags belonging to the group to the service tag.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In addition, in describing the present invention, when it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 is a block diagram of a system for constructing a speech recognition database for a speech recognition ARS service according to an exemplary embodiment of the present invention.

As shown in FIG. 2, the system for constructing a voice recognition database includes a data collector 210 collecting voices of users who use the ARS service, and a transcription unit 230 for converting the collected voices into text. ), A tagging unit 250 for setting tags for each user's voice and a destination setting unit 270 for connecting a service destination for each tag.

The voice of the main user is first collected for the voice recognition ARS service. To this end, the data collection unit 210 when the user is connected to use the ARS service to tell the user "the desired service. For example, how much did I spend? Tell them to say, ”and encourage them to speak their desired service and record it as a voice file. The recorded voice file is stored in the data DB 211.

Next, the transcription unit 230 converts the speech of the user collected by the data collection unit 210 into text according to a predetermined rule. The conversion of the user's voice to text is performed according to a predetermined rule in the process. For example, a voice recognition of the recording file uttered by the user, “Do you want to report the lost phone?” Is recorded as [noise] [side_speech]. Where [noise] is an indication that noise is in the middle and [side_speech] is an indication that someone's voice is in the background. Other notation methods are as follows.

[breath_noise]: Inhale or exhale

[hang_up]: hang up,

[dtmf]: DTMF tones,

[fragment]: voice interrupted,

(()): When voice is spoken but difficult to predict,

((word)): In case of difficult to understand or predicted words

* word *: The speaker pronounced it wrong. * * Notation of the pronunciation notation is written in the correct spelling,

~: The waveform of the vocalization is cut off. If the beginning of the utterance is cut off, write ~ in front of it.

@ hes @: Sounds that sound like "um", "uh", "ah", "eh", "ehn", "hm" before the word. In addition to the pronunciation, 'hey, hello', etc., and words that have meaning should not be written as @ hes @. Hey, hello and so on.

The above-described notation rule is set in the notation dictionary DB 235, and the transcription unit 230 converts the text into speech for each user's utterance stored in the material DB 211 with reference to the notation dictionary DB 235. To perform. The transcription unit 230 stores the result of the conversion into the text in the transcription DB 237. In the speech recognition ARS service, the speech recognition rate is an important factor. As described in the present invention, the data stored in the transmission DB 237 can increase the speech recognition rate in the speech recognition ARS service. When noise or the like is added to the user's voice, it is possible to analyze whether or not the voice form is expressed, and according to the analysis result, more accurate voice recognition is possible when the actual user's voice is recognized.

In more detail, the transcription unit 230 recognizes and converts speech to text (STT) engine 231 and text generated by the STT engine 231 based on a predetermined rule. It includes a refining engine 233. 3A is a diagram illustrating a process of converting a user's speech into text in a transcription unit according to an embodiment of the present invention.

Referring to FIG. 3A, the STT engine 231 converts voices of users stored in the data DB 211 into text using Speech To Text technology. As shown in FIG. 3A, when the voice file of "I want to apply for additional services (cough)" is stored in the data DB 211, the STT engine 231 requests the voice file as text, i. I'm going to do it. " The refining engine 233 refines the text converted by the STT engine 231 with reference to the writing dictionary DB 235. That is, the refining engine 233 writes "side_speech" which refers to the "echo" which is another person's cough sound in the "I want to apply for supplementary service" converted into text. Therefore, the refined text in the refining engine 233 is "I would like to apply for side service [side_speech]". The final purified text file is stored in the transcription DB 237.

Meanwhile, referring to FIG. 2, the tagging unit 250 sets which service the text file corresponds to for each text file stored in the transcription DB 237. That is, a candidate service tag (hereinafter, semantic tag) is set by selecting main keywords for each text file. For example, for a text file of "I would like to apply for additional services", select "Additional services + application" as the main keyword, and set the "extra_service_apply" semantic tag for the text file of "I want to apply for additional services". . Also, for the text file of "I would like to apply for additional service ringtones", select "Additional Service + Request + Ringtone" as the main keyword, and accordingly, add the "extra_service_bell_apply" semantic tag to the text file of "I would like to apply for additional service ringtones". Set it. As described above, the tagging unit 250 sets which service the user's voice corresponds to for each text file stored in the transcription DB 237, that is, for each user's voice. The key keyword recognition in the tagging unit 250 is preferably performed by natural language speech recognition technology. The semantic tag for each user voice set by the tagging unit 250 is stored in the tag DB 251. In this case, the tagging unit 250 stores the semantic tag in the tag DB 251 and stores the main keywords as described above in the semantic tag.

After completing the first tagging as described above, the developer clusters the semantic tags for each user voice stored in the tag DB 251 into services having similar tendencies. For example, the "additional service ringtone application" may be integrated into the "additional service application" as a submenu of the "additional service application", so the "extra_service_apply" semantic tag and the "extra_service_bell_apply" semantic tag are enclosed with "extra_service_apply". When you group semantic tags, you also cluster the key keywords that were assigned to each cement tag. As a result, "extra_service_apply" semantic tag stores "additional service + request + ringtone" as the main keyword. 3B is a diagram illustrating a process of grouping semantic tags into services having similar tendencies. As shown in FIG. 3B, the semantic tag of the "additional service ringtone request" and the semantic tag of the "additional service application" are "additional service". Incorporates into "application".

As described above, candidate semantic tags are clustered into ones having similar tendencies, and clustered and configured tags are referred to herein as application tags. The application tag is a service item provided by the voice recognition ARS service, and the destination setting unit 270 connects the actual service item for each application tag stored in the tag DB 251 according to a developer's input. That is, each application tag is connected with an automatic response, a menu item, or a counselor. For example, the application tag "extra_service_apply" related to "additional service application" preferably provides a menu item. Since there are various types of supplementary services, when a customer using the ARS service inquires the supplementary service, it is preferable to provide a selection menu such as "Please press 1 for Ring to You and 2 for catch call". Therefore, as described above, the application tag "extra_service_apply" related to "additional service request" is connected to a menu item. Destination information for the application tag is stored in the tag-destination DB 271.

[Table 1] shows an example of the design document for the final destination of [account_bill_Detail] among application tags. [account_bill_Detail] is an application tag for payment fee. The final destination of the application tag [account_bill_Detail] is fd_account_bill_details as shown in [Table 1] below. Destination information for this application tag is stored in the tag-destination DB 271.

account_bill_details prompts conditions name wording Tag Confirmation account_bill_details_conf_init1_exp_r1 For details of your bill ... right? Tag Confirmation account_bill_details_conf_init1_exp_r1 Want to see detailed billing history for each item? options next step always -> Final Destination fd_account_bill_details

As such, the data stored in the tag DB 251 and the tag-destination DB 271 are used to provide a voice recognition ARS service in the customer. When the user requests confirmation of the details of the payment fee, the ARS system recognizes the user's voice and selects the main keywords, and uses the selected keywords to query the application tag in the tag DB 251. Then, the destination for the queried application tag is queried in the tag-destination DB 271. Confirmation text designed for the inquired application tag, for example, in [Table 1], the message 'Do you want to check the detailed billing history for each item?' Is sent to the user. If the user answers 'yes', it is determined that the service item desired by the user is the final destination set in the inquired application tag, and the final service is connected. Immediately send the charge details in an automatic response, guide a sub menu, or connect an agent.

4 is a flowchart illustrating a method for establishing a database for a voice recognition ARS service according to an embodiment of the present invention.

As shown in FIG. 4, first, when a user accesses a database building system according to the present invention, the database building system sends a guide message to guide the user to speak a desired service (S401). For example, "Please tell us what service you want. For example, how much is my fee?" In this case, the user can access a database building system using a wired or wireless terminal, but is not limited thereto.

Subsequently, the database building system records and stores the voice spoken by the user according to the announcement (S403). Steps S401 and S403 are repeated as necessary for the database construction. Preferably about 50,000 user voices are collected.

Next, the voices of the collected and stored users are converted into text according to a predetermined rule and stored in operation S405. The conversion of the user's voice to text is performed according to a predetermined rule in the process. For example, a voice recognition of the recording file uttered by the user, “Do you want to report the lost phone?” Is recorded as [noise] [side_speech]. Where [noise] is an indication that noise is in the middle and [side_speech] is an indication that someone's voice is in the background. Other notation methods are as follows.

[breath_noise]: Inhale or exhale

[hang_up]: hang up,

[dtmf]: DTMF tones,

[fragment]: voice interrupted,

(()): When voice is spoken but difficult to predict,

((word)): In case of difficult to understand or predicted words

* word *: The speaker pronounced it wrong. * * Notation of the pronunciation notation is written in the correct spelling,

~: The waveform of the vocalization is cut off. If the beginning of the utterance is cut off, write ~ in front of it.

@ hes @: Sounds that sound like "um", "uh", "ah", "eh", "ehn", "hm" before the word. In addition to the pronunciation, 'hey, hello', etc., and words that have meaning should not be written as @ hes @. Hey, hello and so on.

Next, for each of the converted text files, which service the text file corresponds to is set. In other words, a candidate key tag is selected for each text file, and a candidate service tag (hereinafter, referred to as a semantic tag) is set and the corresponding service tag and the key keyword are stored correspondingly (S407).

For example, for a text file of "I want to apply for supplementary services", select "Additional services + application" as the main keyword, and set the "extra_service_apply" semantic tag for the text file of "I want to apply for supplementary services." Corresponds to the keyword "additional service + application" to store. Also, for the text file of "I would like to apply for additional service ringtones", select "Additional Service + Request + Ringtone" as the main keyword, and accordingly, add the "extra_service_bell_apply" semantic tag to the text file of "I would like to apply for additional service ringtones". Set and save corresponding keyword "Additional Service + Application + Ringtone". At this time, the main keyword recognition is preferably made by natural language speech recognition technology.

After completing the first tagging as described above, the database building system clusters the semantic tags for each user voice stored in the tag DB 251 into services having similar tendencies according to the developer's input (S409). For example, the "additional service ringtone application" may be integrated into the "additional service application" as a submenu of the "additional service application", so the "extra_service_apply" semantic tag and the "extra_service_bell_apply" semantic tag are enclosed with "extra_service_apply". When grouping semantic tags, they cluster together with the key keywords that were assigned to each cement tag. As a result, "extra_service_apply" semantic tag stores "additional service + application + ringtone" as the main keyword. As described above, candidate semantic tags are clustered into those having similar tendencies, and each tag grouped and configured is referred to as an application tag. The application tag becomes a service item provided by the voice recognition ARS service.

Next, for each application tag grouped and sorted as described above, an actual service, that is, a destination is connected and stored (S411). That is, each application tag connects to an automatic response providing desired information, connects a selection menu item, or connects an agent. For example, the application tag "extra_service_apply" related to "additional service application" preferably provides a menu item. Since there are various types of supplementary services, when a customer using the ARS service inquires the supplementary service, it is preferable to provide a selection menu such as "Please press 1 for Ring to You and 2 for catch call". Therefore, as described above, the application tag "extra_service_apply" related to "additional service request" is connected to a menu item.

5 is a diagram illustrating a configuration of an ARS system using speech recognition according to an embodiment of the present invention.

As shown in FIG. 5, an ARS system using voice recognition includes a mobile communication terminal 510, a network 520, and a voice recognition ARS server 530. Here, the network 520 is preferably a mobile communication network including a base station, a base station controller, a switch, and the like, and if the user uses a wired terminal, the network may be an internet network and / or a public network (eg, PSTN). . In addition, the tag DB 535 provided in the speech recognition ARS server 530 uses the tag DB 251 of FIG. 2, and the tag-destination DB 237 uses the tag-destination DB 271 of FIG. 2. will be. In this embodiment, the tag DB 535 and the tag-destination DB 535 are described in separate configurations, but may be implemented as one DB according to the embodiment. In this case, the keyword and the destination information are stored in the application tag correspondingly.

The speech recognition unit 531 of the speech recognition ARS server 530 according to the present invention recognizes the user's voice as a statistical language model, so that when the user connects using the mobile communication terminal 510, a desired item is first spoken. Send a guidance message to receive the user's voice.

In addition, the voice recognition unit 531 recognizes the voice spoken by the user, extracts a main keyword, and searches the application tag in the tag DB 251 using the extracted keyword. In other words, it is to check whether or not the service item desired by the user.

The call connection unit 533 receives an application tag from the voice recognition unit 531, and retrieves a destination for the received application tag from the tag-destination DB 273. For example, when the application tag received from the voice recognition unit 531 is "extra_service_apply", the destination set in the application tag "extra_service_apply" in the tag-destination DB 273, that is, an automatic response or a selection menu, or Search for a destination such as an agent connection. The call connection unit 533 transmits the retrieved destination information to the result guide unit 532.

The result guide unit 532 provides the mobile communication terminal 510 with a result according to the destination information transmitted from the call connection unit 533. If the destination information is an automatic response to the ringtone request, the result guide 532 retrieves the announcement for the ringtone request from the menu DB 539 and provides it to the mobile communication terminal 510. Alternatively, when the destination information transmitted from the call connection unit 533 is a sub menu, for example, the user inquires about 'bigi', 'bigial 2200', 'bigial 2900' The result guide unit 532 guides the submenu to receive the DTMF input as in the related art and provide the final service. Although not shown, the result guide unit 532 provides final service information in association with a database storing a customer information system (WISE), product guide information, and the like.

As described above, when a database is provided according to the present invention and a voice recognition ARS service is provided, a user directly receives a desired service or selects a service that has already been narrowed, so the burden for finding a desired service is small compared to the DTMF method.

In addition, since the search starts from a tree structure that is directly connected to the service or narrowed according to the service requested by the user, the menu step is not deeper than that of the DTMF method. That is, the moving time for moving to the desired service is shorter than that of the DTMF method.

In addition, in the case of the DTMF method, due to the complicated menu structure, the user finds a counselor without attempting from the initial entry. In the case of voice recognition ARS, the number of agent connection calls is reduced compared to the DTMF method due to the reduction of service waiting time due to the relatively low menu structure and low menu level.

As described above, the method of the present invention may be implemented as a program and stored in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.) in a computer-readable form. Since this process can be easily implemented by those skilled in the art will not be described in more detail.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

According to the database construction method of the present invention as described above, it is possible to significantly reduce the time required to develop a speech recognition ARS system, and ultimately to provide a desired service in comparison with the conventional DTMF method in terms of the provision of the speech recognition ARS service. The burden to find is reduced and travel time is reduced. In addition, the number of counseling calls is reduced, which in turn reduces the cost of hiring agents in terms of employment.

Claims (11)

A database construction method for speech recognition ARS service, (a) collecting service question voices from multiple users; (b) converting each collected service question voice into text; (c) grouping the converted text into similar service items and setting a service tag in the grouped group; And (e) mapping a destination to a set service tag. The method of claim 1, Step (c) is, (c-1) setting a semantic tag in the converted text; And (c-2) grouping each set semantic tag by a similar service item, and setting a service tag in the grouped group. The method of claim 2, Step (c-1) is, Extracting a keyword from the converted text; Generating a semantic tag based on the extracted keyword; And And mapping the extracted keyword to the generated semantic tag. The method of claim 3, wherein Step (c-2) is, Grouping each set semantic tag by similar service items; Setting a service tag in the grouped group; And Mapping keywords of semantic tags belonging to the group to the service tag. The method of claim 1, Step (c) is, And extracting a key keyword from the text belonging to the group and mapping the key keyword to the service tag. The method according to any one of claims 1 to 5, The destination is Method of building a database, characterized in that any of the information provided, the selection menu provided, the agent connection. A database construction system for speech recognition ARS service, Collecting means for collecting service question voices from multiple users; Conversion means for converting the collected service question voices of each user into text; Tagging means for grouping the converted text into similar service items and setting a service tag in the grouped group; And Destination mapping means for mapping a destination to the service tag. The method of claim 7, wherein The tagging means, And setting a semantic tag in the converted text, and then setting a service tag by grouping each set semantic tag by similar service items. The method of claim 8, The tagging means, Extracting a keyword from the converted text and generating a semantic tag based on the keyword. The method of claim 9, The tagging means, And a keyword of a semantic tag belonging to the group to the service tag and storing the same. The method of claim 7, wherein The tagging means, And a main keyword is extracted from the text belonging to the group and mapped to the service tag and stored.

Images