KR102224500B1 - System and method for providing interactive customer response service using virtual host character based on artificial intelligence - Google Patents

Abstract

The present invention relates to a system and method for providing an interactive customer service service using a virtual host character based on artificial intelligence.
The system for providing an interactive customer response service using an artificial intelligence-based virtual host character according to the present invention comprises: a general telephone for receiving an incoming call from a customer; When an incoming call is made to a regular phone, it automatically receives the caller's phone number, extracts the caller's phone number, analyzes the caller's voice sound data, generates the customer's language type information, and sends the generated language type information together with the caller's phone number. A call detection device that transmits to an artificial intelligence server and provides a conversation service after a call is received using a virtual host character based on the generated language type information; And receiving the caller's phone number and the customer's language type information transmitted from the call detection device, registering new customer information based on this, and reflecting the learning result to the call detection device by performing mechanical learning based on the given language model. Includes an intelligent server.
According to the present invention, the virtual host character naturally communicates with the customer through the virtual voice generated by artificial intelligence, so that the customer can get a sense of satisfaction with the friendly response, and the business host can efficiently respond to the customer's phone call. It has the advantage of being able to perform other tasks.

Description

System and method for providing interactive customer response service using virtual host character based on artificial intelligence

The present invention relates to a system and method for providing an interactive customer service service, and in more detail, a virtual host character naturally communicates with a customer through a virtual voice generated by artificial intelligence, thereby obtaining a satisfaction from the customer's point of view. From the standpoint of a business host, it relates to a system and method for providing an interactive customer service service using an artificial intelligence-based virtual host character that can efficiently respond to a customer's phone while performing other tasks.

Today, with the development of information and communication technology, society is becoming more complex, and it is not an exaggeration to say that everyday life itself is a communication life. Modern people are living their daily lives by constantly communicating under the environment of a communication society.

In particular, it is very difficult for operators (business owners) who operate businesses such as financial institutions, restaurants, hospitals, beauty salons, hotels, pensions, and airlines to properly respond to the constantly incoming calls from customers.

Therefore, if the owner can respond to calls from customers through other means of response instead of responding to the calls from customers individually, the business owner can perform other natural tasks without being constrained by the phone response from customers, resulting in less stress and less stress. This will also help increase sales (increase profits).

Meanwhile, Korean Patent Laid-Open Publication No. 10-2002-0032257 (Patent Document 1) discloses a "real-time customer response system using characters on the Internet and its operation method", and accordingly, real-time customer response using characters on the Internet. The system is characterized in that a web service operator responds in real time to a service user who visits a web site using a character in an Internet environment to provide services such as counseling and companion shopping.

In the case of Patent Document 1 as described above, by using a character in a client/server environment, two or more people (the first or second user, or the user and the service operator) can go together to shop on the Internet, so that the user purchases a product. I don't know if there will be an effect that makes decision-making easier, but this only provides a service that helps users purchase goods (or services) by using characters in an internet shopping environment. , There is a problem that cannot be responded at all in the case of non-internet environment such as when making a reservation or making a phone call to a beauty salon, hotel, pension, airline company, etc.

Korean Patent Application Publication No. 10-2002-0032257 (published on May 3, 2002)

The present invention was created in consideration of the above, and a virtual host character naturally communicates with the customer through a virtual voice generated by artificial intelligence, so that the customer can get a feeling of satisfaction with the friendly response, and the business host can get a sense of satisfaction. Its purpose is to provide a system and method for providing an interactive customer response service using artificial intelligence-based virtual host characters that can efficiently respond to customer calls and perform other tasks.

In order to achieve the above object, the system for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention,

A general telephone installed in a business and receiving an incoming call from a customer;

It is installed in the business and is electrically connected to the general telephone of the business. When a call is received from the general telephone, it automatically receives and extracts the caller's phone number, and automatically analyzes the caller's voice and sound data to generate the customer's language form information. Thus, a call detection device that transmits the generated language type information together with the caller's phone number to an external artificial intelligence server, and provides a conversation service after a call is received using a virtual host character based on the generated language type information; And

The call detection device is connected to the wired/wireless Internet, receives the caller's phone number and the customer's language type information transmitted from the call detection device, registers customer information based on this, and performs mechanical learning based on a given language model. It is characterized in that it includes an artificial intelligence server that performs and reflects the learning result to the call detection device.

Here, preferably, greeting, announcement, out of office setting, language model setting of the business where the service is running, customer inquiry and modification, frequent customer setting, reservation management, order management, voice recording setting, SMS/MMS sending setting, virtual It may further include an administrator console that performs functions such as host character setting, business WiFi setting, chatbot setting, event/coupon setting, and the like.

At this time, the administrator console can enter and set the business's own words, abbreviations, and special terms at any time.

In addition, it may further include a database in which corpus, chuimsae, silence, reservations, recordings, customer information, and the like are stored.

In addition, the call detection device may generate language type information by dividing it by gender, age group, and language type in generating language type information of a customer by analyzing the caller's voice sound data by itself.

In addition, the call detection device includes: a language discriminator for automatically discriminating information of a speaker by analyzing sound signal data of an input language; A silence processor for discriminating and processing whether the input voice data is silent data or not as a "sound signal pattern"; A built-in speech-to-text (STT) that converts input voice data into text; And a chuimsae processor that discriminates whether or not filler words of the input language are "language model patterns" and adjusts (extends) a waiting time or automatically responds with a chuimsae when a customer response is delayed.

In addition, the call detection device is involved in dynamically changing (extending) a threshold of end-of-speech (EOS) set in advance by discrimination of words of the pendulum by the pendulum processor. With a timer; It may further include an encoder for generating a voice packet by encoding the voice data passed through the silence processor, and transmitting the generated voice packet to the artificial intelligence server.

Here, the silent processor may include a silent phoneme estimation acoustic model for discriminating and processing a silent sound signal from the caller's conversation content, and a noise estimation acoustic model for estimating and processing a noise factor in the conversation content. .

In addition, the chuimsae processor receives the customer's conversation in real time and analyzes whether or not the chuimsae word is included, and if the specified chuimsae word is present, the threshold of end-of-speech (EOS) set in advance is determined. It dynamically changes (extends) and can be configured to respond to customers by creating a response reminder and wait for the next conversation.

In addition, the call detection device includes a conversation manager who considers the natural language conversation flow and situation and analyzes the intention to determine the best conversation strategy to respond to customers in natural language; and receives the caller's conversation contents in real time from the conversation manager. An intention reasoning unit for finally determining a speaker's intention for a specific dialogue phrase, and transmitting a result to the dialogue manager after executing an additional reasoning task; The personalization manager may further include a personalization manager that receives a request from the conversation manager, comprehensively infers the propensity of a specific customer at a current point in time, generates the content of a product or service, etc. to be suggested to the customer, and delivers the content to the conversation manager. .

In this case, the conversation manager may have an information correction function for correcting information that may be misrecognized among the conversation contents through a conversation, and a confirmation function for generating and proceeding with a conversation for correction if the information needs to be confirmed.

In addition, the conversation service provided by the artificial intelligence server may be in the form of a service in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence.

In addition, in order to achieve the above object, the method of providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention,

As a method of providing an interactive customer response service based on a system for providing an interactive customer response service using an artificial intelligence-based virtual host character including a general telephone, a call detection device, and an artificial intelligence server,

a) receiving an incoming call from a customer by the general telephone in the establishment;

b) when a call is received from the general telephone, automatically receiving it by the call detection device and extracting the caller's telephone number;

c) automatically analyzing the caller's voice sound data by the call detection device to generate the customer's language type information, and transmitting the generated language type information together with the caller's phone number to the artificial intelligence server;

d) registering customer information by receiving the caller's phone number and the language type information of the customer transmitted from the call detection device by the artificial intelligence server;

e) performing mechanical learning based on the language model given by the artificial intelligence server and reflecting the learning result to the call detection device; And

f) It is characterized in that it includes the step of providing a conversation service after a call is received using a virtual host character based on the language type information generated by the call detection device.

Here, preferably, the system for providing interactive customer service using the artificial intelligence-based virtual host character further includes an administrator console, and a greeting, announcement, absence setting, business language model setting, customer inquiry by the administrator console And performing correction, frequent customer setting, reservation management, order management, voice recording setting, SMS/MMS sending setting, and the like.

In addition, in step c), the call detection device may generate language type information by dividing it by gender, age group, and language type in generating the customer's language type information by analyzing the caller's voice sound data by itself. .

In addition, in step c), the process of generating the customer's language form information by analyzing the caller's voice sound data by the call detection device itself is performed by analyzing the sound signal data of the language inputted by the language discriminator. Automatically discriminating information; Discriminating and processing whether the voice data input by the silence processor is silent data or not as a "sound signal pattern"; Converting voice data input by built-in speech-to-text (STT) into text; And discriminating whether or not filler words of the language input by the chuimsae processor are determined by the "language model pattern", and adjusting (extending) the waiting time or automatically responding with the chuimsae in case the customer response is delayed. I can.

At this time, the call detection device automatically analyzes the caller's voice and sound data to generate the customer's language type information. Dynamically changing (extending) a threshold of end-of-speech (EOS); It may further include the step of generating a voice packet by encoding the voice data passed through the silencer by an encoder, and transmitting the generated voice packet to the artificial intelligence server.

Here, the silent processor may include a silent phoneme estimation acoustic model for discriminating and processing a silent sound signal from the caller's conversation content, and a noise estimation acoustic model for estimating and processing a noise factor in the conversation content. .

In addition, the chuimsae processor receives the customer's conversation in real time and analyzes whether or not the chuimsae word is included, and if the specified chuimsae word is present, the threshold of end-of-speech (EOS) set in advance is determined. It dynamically changes (extends) and can be configured to respond to customers by creating a response reminder and wait for the next conversation.

In addition, in step c), the call detection device automatically analyzes the caller's voice and sound data to generate the customer's language form information. Determining a conversation strategy and responding to customers in natural language; Receiving conversation contents of a sender from the conversation manager in real time by an intention inferring device, finally determining the speaker's intention for a specific conversation phrase, and delivering the result to the conversation manager after executing an additional reasoning task; Further comprising the step of receiving a request from the conversation manager by the personalization manager, comprehensively inferring the propensity of a specific customer at the current point of time, generating the content of the product or service to be proposed to the customer, and transmitting the content to the conversation manager. can do.

In this case, the conversation manager may have an information correction function for correcting information that may be misrecognized among the conversation contents through a conversation, and a confirmation function for generating and proceeding with a conversation for correction if the information needs to be confirmed.

In addition, the conversation service provided by the call detection device in step f) may be in the form of a service in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence.

According to the present invention, the virtual host character naturally communicates with the customer through the virtual voice generated by artificial intelligence, so that the customer can get a sense of satisfaction with the friendly response, and the business host can efficiently respond to the customer's phone call. It has the advantage of being able to perform other tasks.

1 is a diagram schematically showing the configuration of a system for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention.
FIG. 2 is a diagram illustrating an internal configuration of a call detection device of the system for providing an interactive customer service service using an artificial intelligence-based virtual host character shown in FIG. 1.
3 is a flowchart showing an execution process of a method for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention.
4 is a diagram schematically showing a dynamic change of an end point of an utterance through a chuimsae processor.
5 is a view showing a comparison of utterance end processing methods.
6 is a diagram showing an existing processing method and a dynamic change method at the end of an utterance in a system configuration.
7 is a diagram schematically showing a response function of a chuimsae processor.
8 is a diagram schematically illustrating a silent processing function of a silent processor.
9 is a diagram schematically showing an intention reasoning function by a conversation manager and an intention reasoning machine.
10 is a diagram schematically showing information correction and confirmation functions of a conversation manager.
11 is a diagram showing an information correction dialogue process of a dialogue manager.
12 is a diagram schematically illustrating a customer-customized proposal function of a conversation manager.
13 is a diagram showing a system configuration of a language model learned by an artificial intelligence server.
14 is a diagram showing a method of learning and refining language model data by an artificial intelligence server.

Terms or words used in this specification and claims are limited to their usual or dictionary meanings and should not be interpreted, and that the inventor can appropriately define the concept of terms in order to describe his own invention in the best way. Based on the principle, it should be interpreted as a meaning and concept consistent with the technical idea of the present invention.

Throughout the specification, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. In addition, terms such as "...unit", "...group", "module", and "device" described in the specification mean units that process at least one function or operation, which is hardware or software, or a combination of hardware and software. It can be implemented as

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically showing the configuration of a system for providing an interactive customer service service using an artificial intelligence-based virtual host character according to an embodiment of the present invention.

Referring to FIG. 1, an interactive customer service providing system 100 using an artificial intelligence-based virtual host character according to the present invention includes a general phone 110, a call detection device 120, and an artificial intelligence server 130. It consists of including.

The general telephone 110 is installed in the business and receives an incoming call from a customer. Here, the business refers to a business that has subscribed to the interactive artificial intelligence service provided by the present invention. In addition, the general telephone 110 is a telephone used by a business and may include all of a landline telephone, an Internet telephone, and a mobile telephone.

The call detection device 120 is installed in the business and is electrically connected to the general telephone 110 of the business, and when a call is received from the general telephone 110, it automatically receives and extracts the caller's telephone number, and the caller's voice and sound data By self-analyzing and generating customer's language type information, the generated language type information is transmitted to the external artificial intelligence server 130 along with the caller's phone number, and a virtual host character is used based on the generated language type information. Thus, it provides a conversation service after receiving a call. Here, the virtual host character means an artificial intelligence virtual host that responds to a customer over the phone.

The artificial intelligence server 130 is connected to the call detection device 120 through wired/wireless Internet, receives the caller's phone number and the language type information of the customer transmitted from the call detection device 120, and based on this, customer information Is registered, and the learning result is reflected in the call detection device 120 by performing mechanical learning based on the given language model.

Here, the interactive customer service providing system 100 using the artificial intelligence-based virtual host character of the present invention as described above is preferably a greeting, an announcement, an absence setting, a language model setting of the business where the service is running, Administrator console that performs functions such as customer inquiry and modification, frequent customer setting, reservation management, order management, voice recording setting, SMS/MMS sending setting, virtual host character setting, business WiFi setting, chatbot setting, event/coupon setting, etc. 140) may be further included. In this case, the administrator console 140 may input and set the unique word, abbreviation, and special term of the business at any time.

Here, a description will be added to the administrator console 140 and the business language model setting and the virtual host character setting by the administrator console 140 as described above.

First, when explaining access to the manager console 140, a web browser (Internet Explorer, Chrome, etc.) of a PC or Point of Sales (POS) in the business is executed. Then, access the URL of the administrator's homepage of the service provider (eg, http://smartoffice.posicube.com). Then, log in by entering the business's phone number and password. Then select the "Administrator Console" function. After that, it is possible to set various functions as described above through the administrator console.

Next, referring to the setting of the business language model, first, the “business language model setting” function is selected from the administrator console 140. Then, according to the purpose of the new word to be registered, a menu is selected from among "unique nouns (menu, etc.)", "chuimsae word", "special symbol", and "abbreviation/variant word". Then, after entering the new word and conversion result together, select "Save". Then, when the "Language Model Learning" button is pressed, the artificial intelligence server 130 starts machine learning. And when the learning is finished, the artificial intelligence server 130 automatically reflects the result, as shown in FIG. 2.

Next, referring to the virtual host character setting, first, the “virtual host character setting” function is selected from the administrator console 140. Then, select one of the pre-set host characters (for example, "Kim Dae-ri", "Bak Jong-won", "Gourmet Hwang Gyo-ik", "Big Mama Lee Hye-jung", "Three Sis and Three Kids Lee Seo-jin", "Mokbang Ha Jung-woo"). Or, select "Create a new character". Then, select "Available Languages". For example, select one or more of Korean, Chinese, Japanese, and English. Then, enter "Responsive Customer Type". For example, enter the age group, gender, and whether you are a regular member. After that, if there are no more changes, select the "Done" button.

On the other hand, the interactive customer service providing system 100 using an artificial intelligence-based virtual host character of the present invention preferably includes a database 150 in which corpus, chuimsae, silence, reservations, recordings, customer information, etc. are stored. It may contain more. Here, the corpus refers to a text data set classified for a specific purpose for processing natural language.

In addition, the call detection device 120, in generating the language form information of the customer by analyzing the caller's voice sound data by itself, by gender, age group, language type (for example, "female", "20s" , "English") can be used to generate language type information.

In addition, the call detection device 120, as shown in Fig. 2, the language discriminator 121 for automatically discriminating information of the speaker by analyzing the audio signal data of the input language; A silence processor 122 for discriminating and processing whether the input voice data is silent data or not as a "sound signal pattern"; A built-in speech-to-text (STT) 123 for converting the input voice data into text; And a chuimsae processor 124 that discriminates whether or not filler words of the input language by "language model pattern" and adjusts (extends) the waiting time or automatically responds with a chuimsae when customer response is delayed. I can. Here, the language discriminator 121, the silence processor 122, the built-in speech-to-text (STT) 123 that converts the input voice data into text, the chuimsae processor 124, and the like are software or hardware. It can be composed of.

In addition, the call detection device 120 dynamically changes (extends) a threshold of end-of-speech (EOS) set in advance by discrimination of words by the pendulum processor 124. The utterance end timer 126 involved in ); It may further include an encoder for generating a voice packet by encoding the voice data passed through the silence processor 122 and transmitting the generated voice packet to the artificial intelligence server 130.

Here, the silence processor 122 estimates a silent phoneme estimation acoustic model 122a (refer to FIG. 8) for discriminating and processing a silent sound signal from the caller's conversation content, and a noise factor in the conversation content. A noise estimation acoustic model 122b to be processed may be provided.

In addition, the chuimsae processor 124 receives the conversation contents of the customer in real time, analyzes whether or not the words are included, and if a specified word is present, a threshold of end-of-speech set in advance; EOS) dynamically changes (extends) and can be configured to respond to customers by creating a response message and wait for the next conversation.

In addition, the call detection device 120 includes a conversation manager 901 (refer to FIG. 9) that considers a natural language conversation flow and situation and analyzes an intention to determine the best conversation strategy to respond to customers in natural language; An intention inferring device that receives the caller's conversation content from the conversation manager 901 in real time, finally determines the speaker's intention for a specific conversation phrase, and delivers the result to the conversation manager 901 after executing an additional reasoning task (902) and; Personalization manager that receives proposal content from the conversation manager 901, comprehensively infers the propensity of a specific customer at the current point of time, creates content, such as a product or service to be proposed to the customer, and delivers it to the conversation manager 901 It may further include 903 (see FIG. 12). In this case, the conversation manager 901 may include an information correction function for correcting information that may be misrecognized among the conversation contents into a conversation, and a confirmation function for generating and proceeding with a conversation for correction when the information needs to be confirmed. Here, the conversation manager 901, the intention inferring device 902, and the personalization manager 903 as described above may be formed of software or hardware.

In addition, the conversation service provided by the artificial intelligence server 130 may be in the form of a service in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence.

As described above, the interactive customer service providing system 100 using the artificial intelligence-based virtual host character of the present invention uses the artificial intelligence server 130 and the call detection device 120 to enable the virtual host character to Although it has been described as providing a service to respond to customers, it is not necessarily limited to providing services in this form, and in some cases, a service in which the host or employee of the business directly responds to the customer's phone call, like a general call center, is provided. You can also provide. In this case, of course, in a state in which the host or employee of the business cuts off the connection between the general telephone 110 and the call detection device 120 and the artificial intelligence server 130 in the business by a simple switch operation, such a direct response service is available. Done.

Then, hereinafter, a method of providing an interactive customer service service using an artificial intelligence-based virtual host character based on the system for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention having the above configuration. Let's explain.

3 is a flowchart illustrating an execution process of a method for providing an interactive customer service service using an artificial intelligence-based virtual host character according to an embodiment of the present invention.

Referring to FIG. 3, a method for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention includes a general phone 110, a call detection device 120, and an artificial intelligence server 130 as described above. A method for providing an interactive customer service service based on a system for providing an interactive customer service service using an artificial intelligence-based virtual host character including, first, an incoming call from a customer through the general phone 110 in a business establishment. Receive (step S301).

Then, when a call comes in to the general telephone 110, the call detection device 120 automatically receives the call and extracts the caller's telephone number (step S302).

Then, the caller's voice sound data is analyzed by the caller's voice and sound data by itself to generate the customer's language type information, and the generated language type information is sent to the artificial intelligence server 130 together with the caller's phone number. Transfer (step S303).

Then, the artificial intelligence server 130 receives the caller's phone number and the language type information of the customer transmitted from the call detection device 120, and registers new customer information based on this (step S304). That is, the artificial intelligence server 130 inquires the customer's past history information based on the caller's phone number, and transmits the result to the call detection device 120. At this time, if the caller's past history information does not exist, the call detection device 120 recognizes as a new customer, generates new history information, and transmits it to the artificial intelligence server 130, and the artificial intelligence server 130 It is to register new customer information based on new history information.

Further, mechanical learning is performed based on the language model given by the artificial intelligence server 130 and the learning result is reflected in the call detection device 120 (step S305).

Thereafter, a conversation service after receiving a call is provided using a virtual host character based on the language type information generated by the call detection device 120 (step S306). At this time, when the initial conversation with the caller starts, the call detection device 120 provides a customized conversation response service by using various functions (applications) such as the conversation manager 901. In addition, the conversation manager 901 continuously observes and analyzes the conversation contents.

Here, in the series of processes as described above, preferably, the interactive customer service providing system 100 using the artificial intelligence-based virtual host character further includes a manager console 140, and the manager console 140 ) By: greeting, announcement, absent setting, business language model setting, customer inquiry and modification, frequent customer setting, reservation management, order management, voice recording setting, SMS/MMS sending setting, etc. have.

In addition, in step S303, the call detection device 120 may generate language type information by dividing it by gender, age group, and language type in generating the customer's language type information by analyzing the caller's voice sound data by itself. I can.

In addition, the process of generating the customer's language form information by analyzing the caller's voice sound data by the call detection device 120 in step S303, the sound signal of the language input by the language discriminator 121 Analyzing the data and automatically discriminating information of a speaker; Discriminating and processing whether the voice data input by the silence processor 122 is silent data or not as a "sound signal pattern"; Converting voice data input by the built-in speech-to-text (STT) 123 into text; And discriminating whether or not filler words of the language input by the chuimsae processor 124 by "language model pattern", and adjusting (extending) the waiting time or automatically responding to the chuimsae when the customer response is delayed. It can be configured.

At this time, the call detection device 120, in generating the language form information of the customer by analyzing the caller's voice sound data by itself, using the utterance end timer 126, the chuimsae word discrimination by the chuimsae processor 124 Dynamically changing (extending) a threshold of end-of-speech (EOS) set in advance by; The encoder 125 may further include encoding the voice data passed through the silent processor 122 to generate a voice packet, and transmitting the generated voice packet to the artificial intelligence server 130.

Here, the silent processor 122 includes a silent phoneme estimation acoustic model 122a for discriminating and processing a silent sound signal from the caller's conversation content, and a noise estimation acoustic for estimating and processing a noise element in the conversation content. A model 122b may be provided.

In addition, the chuimsae processor 124 receives the conversation contents of the customer in real time, analyzes whether or not the words are included, and if a specified word is present, a threshold of end-of-speech set in advance; EOS) dynamically changes (extends) and can be configured to respond to customers by creating a response message and wait for the next conversation.

In addition, in the step S303, the call detection device 120 automatically analyzes the caller's voice and sound data to generate the customer's language form information, taking into account the natural language conversation flow and situation by the conversation manager 901, and Analyzing and determining the best conversation strategy to respond to customers in natural language; The intention inferring device 902 receives the caller's conversation content from the conversation manager 901 in real time, finally determines the speaker's intention for a specific conversation phrase, executes an additional reasoning task, and returns the result to the conversation manager ( 901); The personalization manager 903 receives a request for proposal content from the conversation manager 901, comprehensively infers the propensity of a specific customer at the current point of time, and creates the content of the product or service to be offered to the customer, and then the conversation manager ( 901).

In this case, the conversation manager 901 may include an information correction function for correcting information that may be misrecognized among the conversation contents into a conversation, and a confirmation function for generating and proceeding with a conversation for correction when the information needs to be confirmed.

In addition, the conversation service provided by the call detection device 120 in step S306 may be in the form of a service in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence.

Hereinafter, a further description will be given in relation to the system and method for providing an interactive customer service service using an artificial intelligence-based virtual host character according to the present invention as described above.

4 is a diagram schematically showing a dynamic change of an end point of an utterance through a chuimsae processor.

Referring to FIG. 4, when a call is made to the phone 110 in the business that is connected to the call detection device 120, the call detection device 120 automatically receives the call and then starts a conversation service with the caller. At this time, the chuimsae processor 124 receives the customer's conversation content in real time and analyzes whether Filler words are included in the conversation content. As a result of the analysis, if there are specified words (eg, "Uh...", "Let's stay still...", "Well...", etc.), the preset threshold for the end of the utterance (EOS: Threshold of end-of-speech) is dynamically changed (extended). Then, the chuimsae processor 124 responds to the customer by generating a response chuimsae (eg, "Yes", "I will wait", etc.), and waits for the next conversation.

(A) is a case where the limit value is short (e.g. 1 second), and if the customer's response stops in the middle of more than 1 second, it is treated as utterance end and natural language misrecognition occurs. (B) is a case where the limit value is long (e.g. 2 seconds), and even if the customer responds without interruption, AI (artificial intelligence) unconditionally waits for 2 seconds, causing inconvenience. (C) is the case of applying the dynamic change function (existing limit value of 1 second). Initially, the utterance end point is maintained at 1 second, but if the conversation does not end normally or there is a word of the customer, the utterance end point is dynamically changed. Automatically responds to memorial service.

5 is a view showing a comparison of utterance end processing methods.

Referring to FIG. 5, when looking at the speech end processing timing and recognition results on the right in relation to the conversation content on the left, there is no additional processing in the existing method, so the reservation is processed as "next week" as a recognition result, but in the dynamic change method The chuimsae processing unit 124 analyzes the chuimsae in the conversation contents, extends the utterance end point, and automatically responds to the chuimsae. As a result, the final recognition result is processed for utterances to end next week, Friday, and 1 o'clock.

6 is a diagram showing an existing processing method and a dynamic change method at the end of an utterance in a system configuration.

Referring to FIG. 6, (A) is an existing processing method, in which the call detection device transmits the utterance end flag and the voice packet to the artificial intelligence server without processing the content of the conversation.

(B) is a dynamic change method, in which the utterance end timer 126 is changed by processing the utterance end timer 126 by processing the utterance in the conversation by the utterance processor 124 of the call detection device 120, and based on the utterance end flag and voice packet Is transmitted to the artificial intelligence server 130.

7 is a diagram schematically showing a response function of a chuimsae processor.

Referring to FIG. 7, when the customer's answer is delayed or the end of the speech is extended, the call detection device 120 automatically responds to the customer with an appropriate sentence, so that the conversation can be continued. To induce. Accordingly, it is possible to proceed with an uninterrupted conversation.

8 is a diagram schematically illustrating a silent processing function of a silent processor.

Referring to FIG. 8, by applying a "Acoustic model" for each silent pattern by the silent processor 122 inside the call detection device, by itself discriminating and processing silent sound signals, faster processing speed and feedback compared to server-based technology Can be provided. In addition, it is possible to reduce the size of voice data transmitted to the encoder 125 and the server and improve the overall response speed.

9 is a diagram schematically showing an intention reasoning function by a conversation manager and an intention reasoning machine.

Referring to FIG. 9, the conversation manager 901 in the call detection device 120 transmits the caller's conversation content to the intention inferring device 902 in real time. When the sender's intention is unclear and additional work is required for inference inference, the conversation manager 901 proposes the inference work to the customer and receives approval. The intention inferring device 902 derives the result after executing the additional inference task and transmits the result to the conversation manager 901. Then, the conversation manager 901 delivers the inference result to the customer.

10 is a diagram schematically showing information correction and confirmation functions of a conversation manager.

Referring to FIG. 10, the conversation manager 901 analyzes the content of the caller's conversation in real time. At this time, it is discriminated whether important information that needs to be confirmed is included in the conversation contents of the customer. In the case of information that needs to be confirmed, a dialog for correction is created and proceeded. When the final calibration result is approved by the customer, the next operation is carried out.

11 is a diagram showing an information correction dialogue process of a dialogue manager.

Referring to FIG. 11, when information correction is not completed at once, the conversation manager 901 repeatedly generates and executes a correction query to generate accurate information and finally receive approval from the customer.

12 is a diagram schematically illustrating a customer-customized proposal function of a conversation manager.

Referring to FIG. 12, the conversation manager 901 may discriminate a "suggest required" situation among conversation contents of a customer. In this regard, the conversation manager 901 requests the personalization manager 903 for suggestions. Personalization manager 903 uses customer history / real-time information / database to create contents to be suggested for goods or services to be purchased by customers, and then delivers them to conversation manager 901. Then, the conversation manager 901 delivers the content to be proposed for the product or service to the customer.

13 is a diagram showing a system configuration of a language model learned by an artificial intelligence server.

13, the language model (business language model) learned by the artificial intelligence server 130 is largely composed of an industry language model (ILM) and a business language model (BLM). . Such language model setting is made by the administrator console 140, and when the language model learning button is pressed in the administrator console 140, the artificial intelligence server 130 starts mechanical learning, and automatically reflects the result when the learning is finished. .

14 is a diagram showing a method of learning and refining language model data by an artificial intelligence server.

Referring to FIG. 14, as described above, when the language model learning button is pressed in the administrator console 140, the artificial intelligence server 130 starts mechanical learning. At this time, the artificial intelligence server 130 receives the language model data, chat log data, and voice log data, and performs the task of classifying chuimsae/special symbols, classifying abbreviations/variants and other texts, Morphology is analyzed based on the language model (BLM) for each business. After that, the recognition unit segmentation, pattern and phoneme information, and pronunciation dictionary are updated to finally derive a learning and refined language model.

As described above, in the system and method for providing an interactive customer service using an artificial intelligence-based virtual host character according to the present invention, the virtual host character naturally communicates with the customer through a virtual voice generated by artificial intelligence, thereby allowing the customer's position. At the same time, you can get a sense of satisfaction with the kind response, and the business host has the advantage of being able to perform other tasks while responding efficiently to the customer's phone call.

As described above, the present invention has been described in detail through preferred embodiments, but the present invention is not limited thereto, and various changes and applications can be made without departing from the technical spirit of the present invention. It is self-explanatory to the technician. Therefore, the true scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present invention.

100: (Invention) Artificial intelligence-based...Interactive customer service providing system
110: desk phone 120: call detection device
121: language differentiator 122: silent processor
123: built-in STT 124: Chuimsae processor
125: encoder 126: utterance end timer
130: artificial intelligence server 140: administrator console
150: Database 122a: Silent Phoneme Estimation Acoustic Model
122b: noise estimation acoustic model 901: dialog manager
902: Intent Reasoner 903: Personalization Manager

Claims (22)

A general telephone installed in a business and receiving an incoming call from a customer;
It is installed in the business and is electrically connected to the general telephone of the business. When a call is received from the general telephone, it automatically receives and extracts the caller's phone number, and automatically analyzes the caller's voice and sound data to generate the customer's language form information. Thus, a call detection device that transmits the generated language type information together with the caller's phone number to an external artificial intelligence server, and provides a conversation service after a call is received using a virtual host character based on the generated language type information; And
The call detection device is connected to the wired/wireless Internet, receives the caller's phone number and the customer's language type information transmitted from the call detection device, registers customer information based on this, and performs mechanical learning based on a given language model. It includes an artificial intelligence server that performs and reflects the learning result to the call detection device,
Greetings, announcements, absent settings, language model setting of the business where the service is running, customer inquiry and modification, frequent customer setting, reservation management, order management, voice recording setting, SMS/MMS sending setting, virtual host character setting, business It further includes an administrator console that performs functions of WiFi setting, chatbot setting, and event/coupon setting,
The administrator console is a system for providing an interactive customer response service using a virtual host character based on artificial intelligence that frequently inputs and sets at least one of a business's own words, abbreviations, and special terms.
delete delete The method of claim 1,
A system for providing interactive customer service using artificial intelligence-based virtual host characters, which further includes a database storing corpus, chuimsae, silence, reservations, recordings, and customer information.
The method of claim 1,
The call detection device uses a virtual host character based on artificial intelligence that generates language type information by analyzing the caller's voice and sound data by itself and generating the language type information of the customer, by gender, age group, and language type. Interactive customer service delivery system.
The method of claim 1,
The call detection device,
A language discriminator for automatically discriminating information of a speaker by analyzing the audio signal data of the input language;
A silence processor for discriminating and processing whether the input voice data is silent data or not as a "sound signal pattern";
A built-in speech-to-text (STT) that converts input voice data into text; And
Artificial intelligence-based AI-based system that includes a filler word processor that determines whether or not filler words of the input language are "language model pattern" and adjusts (extends) the waiting time when customer response is delayed, or automatically responds with a flurry. Interactive customer service providing system using virtual host characters.
The method of claim 6,
The call detection device,
An utterance end timer that dynamically changes (extends) a threshold of end-of-speech (EOS) set in advance by discriminating a vocabulary word by the choo-me-sae processor;
An interactive customer service service using an artificial intelligence-based virtual host character further comprising an encoder for generating a voice packet by encoding the voice data passed through the silent processor, and transmitting the generated voice packet to the artificial intelligence server. Delivery system.
The method of claim 6,
The silent processor has an artificial intelligence-based acoustic model including a silent phoneme estimation acoustic model for discriminating and processing a silent sound signal from the caller's conversation content, and a noise estimation acoustic model for estimating and processing noise elements in the conversation content. Interactive customer service providing system using virtual host characters.
The method of claim 6,
The Chuimsae processor receives the customer's conversation in real time and analyzes whether or not the words are included, and if there is a specified word, the threshold of end-of-speech (EOS) set in advance is dynamically determined. A system for providing interactive customer response services using artificial intelligence-based virtual host characters configured to change (extend), respond to customers by generating a response message, and wait for the next conversation. The method of claim 1,
The call detection device,
A conversation manager who considers the natural language conversation flow and situation, analyzes intentions, determines the best conversation strategy, and responds to customers in natural language;
An intention inference unit for receiving conversation contents of a sender from the conversation manager in real time, finally determining a speaker's intention for a specific conversation phrase, and delivering a result to the conversation manager after executing an additional reasoning task;
Artificial intelligence further comprising a personalization manager that receives proposal contents from the dialogue manager, comprehensively infers the propensity of a specific customer at the present time, creates contents or services, etc. to be proposed to the customer, and delivers the contents to the dialogue manager. An interactive customer service provision system using a virtual host character based.
The method of claim 10,
The conversation manager creates an artificial intelligence-based virtual host character that has an information correction function for correcting information that may be misrecognized among conversation contents into a conversation, and a confirmation function for generating and proceeding with a conversation for correction if information needs to be confirmed. Interactive customer service provision system using.
The method of claim 1,
The conversation service provided by the artificial intelligence server provides an interactive customer response service using an artificial intelligence-based virtual host character, which is a service in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence. system.
As a method of providing an interactive customer response service based on a system for providing an interactive customer response service using an artificial intelligence-based virtual host character including a general telephone, a call detection device, and an artificial intelligence server,
a) receiving an incoming call from a customer by the general telephone in the establishment;
b) when a call is received from the general telephone, automatically receiving it by the call detection device and extracting the caller's telephone number;
c) automatically analyzing the caller's voice sound data by the call detection device to generate the customer's language type information, and transmitting the generated language type information together with the caller's phone number to the artificial intelligence server;
d) registering customer information by receiving the caller's phone number and the language type information of the customer transmitted from the call detection device by the artificial intelligence server;
e) performing mechanical learning based on the language model given by the artificial intelligence server and reflecting the learning result to the call detection device; And
f) providing a conversation service after a call is received using a virtual host character based on the language type information generated by the call detection device,
The system for providing interactive customer service using the artificial intelligence-based virtual host character further includes an administrator console, and a greeting, announcement, absent setting, business language model setting, customer inquiry and modification, regular customers by the administrator console Further comprising the step of performing setting, reservation management, order management, voice recording setting, SMS/MMS sending setting,
A method of providing an interactive customer response service using an artificial intelligence-based virtual host character that frequently inputs and sets at least one of a business's unique word, abbreviation, and special term by the administrator console.
delete The method of claim 13,
In the step c), the call detection device automatically analyzes the voice and sound data of the caller to generate the language type information of the customer.In the case of generating language type information by gender, age group, and language type, the call detection device generates language type information based on artificial intelligence. A method of providing interactive customer service using virtual host characters.
The method of claim 13,
In the step c), the process of generating the customer's language form information by analyzing the caller's voice sound data by the call detection device by itself,
Automatically discriminating information of a speaker by analyzing sound signal data of a language input by a language discriminator;
Discriminating and processing whether the voice data input by the silence processor is silent data or not as a "sound signal pattern";
Converting voice data input by built-in speech-to-text (STT) into text; And
An artificial composition comprising the step of discriminating whether or not Filler words of the language input by the Chuimsae processor are "language model patterns" and adjusting (extending) the waiting time or automatically responding with the Chuimsae in case the customer response is delayed. A method of providing an interactive customer service service using an intelligence-based virtual host character.
The method of claim 16,
The call detection device automatically analyzes the caller's voice and sound data to generate the customer's language form information, using a utterance end timer, the threshold value of the utterance end time set in advance by discrimination of the words of the chuimsae by the chuimsae processor Dynamically changing (extending) (Threshold of end-of-speech; EOS);
An interactive customer service service using an artificial intelligence-based virtual host character further comprising the step of generating a speech packet by encoding the speech data passed through the silencer by an encoder, and transmitting the generated speech packet to the artificial intelligence server. Delivery method.
The method of claim 16,
The silent processor has an artificial intelligence-based acoustic model including a silent phoneme estimation acoustic model for discriminating and processing a silent sound signal from the caller's conversation content, and a noise estimation acoustic model for estimating and processing noise elements in the conversation content. A method of providing interactive customer service using virtual host characters.
The method of claim 16,
The Chuimsae processor receives the customer's conversation in real time, analyzes whether the words are included in the Chuimsae, and if there is a designated Chuimsae word, the threshold of end-of-speech (EOS) set in advance is dynamically determined. A method of providing an interactive customer response service using an artificial intelligence-based virtual host character configured to change (extend), respond to the customer by generating a response message, and wait for the next conversation.
The method of claim 13,
In the step c), the call detection device automatically analyzes the caller's voice and sound data to generate the customer's language form information,
A step of responding to customers in natural language by determining the best conversation strategy by considering the natural language conversation flow and situation and analyzing the intention by the conversation manager;
Receiving conversation contents of a sender from the conversation manager in real time by an intention inferring device, finally determining the speaker's intention for a specific conversation phrase, and delivering the result to the conversation manager after executing an additional reasoning task;
Further comprising the step of receiving a request from the conversation manager by the personalization manager, comprehensively inferring the propensity of a specific customer at the current point of time, generating the content of the product or service to be proposed to the customer, and transmitting the content to the conversation manager. A method of providing an interactive customer service service using a virtual host character based on artificial intelligence.
The method of claim 20,
The conversation manager creates an artificial intelligence-based virtual host character that has an information correction function for correcting information that may be misrecognized among conversation contents into a conversation, and a confirmation function for generating and proceeding with a conversation for correction if information needs to be confirmed. How to provide interactive customer service using.
The method of claim 13,
The conversation service provided by the call detection device in step f) is a conversation using an artificial intelligence-based virtual host character, which is a service form in which a virtual host character communicates with a customer (caller) through a virtual voice generated by artificial intelligence. How to provide customer service.

Images