KR102604277B1 - Complex sentiment analysis method using speaker separation STT of multi-party call and system for executing the same - Google Patents

Abstract

A complex emotion analysis server using STT according to an embodiment of the present invention includes an STT conversion unit that converts call voice data between users into call text data in real time when a plurality of user terminals are connected, based on a predetermined database. A correction data generator for generating correction data by modifying call voice data and the call text data, and generating a first reference database for storing the number of times for each reference data referenced when generating the correction text data in a first reference database for each speaker. and an emotion determination unit that determines an emotion for each speaker based on an emotion determination keyword depending on whether a predetermined emotion keyword exists among reference data for each speaker stored in the first reference database.

Description

Complex sentiment analysis method using speaker separation STT of multi-party call and system for executing the same}

The present invention relates to a complex emotion analysis method using speaker separation STT in a multi-party call and a system for executing the same. More specifically, it relates to a multi-party call analysis method that can analyze multi-party call voice data to separate speakers and then provide the emotions between the speakers. This relates to a complex emotion analysis method using speaker separation STT in a call and a system for executing the same.

Recently, with the development of computing technology, the form of computers has become increasingly smaller, and various wearable devices that can be carried around at all times have been created. As the form of computers changes, the types of human interaction required have also become more diverse, and various intelligent services are required. , As research on artificial intelligence for intelligent services is actively progressing, it is also becoming important for devices to recognize people's emotional information and interact appropriately with people.

Because humans express their emotions to others in a variety of ways through facial expressions, voices, and gestures, research in various fields is being conducted to recognize and determine human emotional information through media such as video, voice, and biological signals. It is actively underway.

General voice data to text conversion (STT: Speech To Text) technology generates words and sentences by analyzing the pronunciation, intonation, and length of the input voice data when recorded voice or voice data played in real time is input. It speaks of technology.

However, in the case of voice data, the pronunciation and intonation of the speaker are different, and there is noise in the voice depending on the speaker's surrounding environment, making it difficult to recognize and analyze the correct pronunciation.

Korean Patent Publication No. 10-2022-0121455 relates to a speaker classification system using STT, which uses STT to store the conversation between speakers as text, a voice signal input unit through which voice signals are input from multiple speakers, and STT. It includes a structure that provides conversation text between speakers to users' mobile terminals through the change unit, but does not provide a function to analyze and classify data by analyzing the emotions of speakers' voices, and a structure for the recognition rate of voice data. There is this shortcoming.

In order to increase the recognition rate of such voice data, a database storing numerous sample data for various pronunciations and intonations is needed, and a correction system that can provide feedback for misrecognized voice data is needed.

Korea Patent Publication No. 10-2022-0121455 (published on September 1, 2022)

The purpose of the present invention is to provide a complex emotion analysis method using speaker separation STT in a multi-party call, which can analyze multi-party call voice data, separate speakers, and then provide emotions between speakers, and a system for executing the same.

In addition, the present invention provides that if a predetermined keyword is present in each call text data converted from multi-party call voice data to call text, the text data can be modified using reference data corresponding to the text keyword to generate modified text data. The purpose is to provide a complex emotion analysis method using speaker separation STT for multi-party calls and a system for executing the same.

The objects of the present invention are not limited to the objects mentioned above, and other objects and advantages of the present invention that are not mentioned can be understood by the following description and will be more clearly understood by the examples of the present invention. Additionally, it will be readily apparent that the objects and advantages of the present invention can be realized by the means and combinations thereof indicated in the patent claims.

To achieve this purpose, a complex emotion analysis server using STT includes an STT conversion unit that converts call voice data between users into call text data in real time when a plurality of user terminals are connected, and the call voice based on a predetermined database. A correction data generator for generating correction data by modifying data and the call text data, a first reference database generator for storing the number of times for each reference data referenced when generating the correction text data in a first reference database for each speaker, and and an emotion determination unit that determines an emotion for each speaker based on the emotion determination keyword depending on whether a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database.

In one embodiment, the emotion determination unit extracts the reference count for each speaker when a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database, and extracts the corresponding emotion according to the reference number of the reference data for each speaker. It may include an emotion decision unit that decides with an emotion corresponding to the decision keyword.

In addition, a complex emotion analysis method using STT to achieve this purpose includes converting call voice data between users into call text data in real time when a plurality of user terminals are connected, and the call voice data based on a predetermined database. and generating correction data by modifying the call text data, storing the number of times for each reference data referenced when generating the correction text data in a first reference database for each speaker, and for each speaker stored in the first reference database. It includes the step of determining an emotion for each speaker based on the emotion determination keyword depending on whether a predetermined emotion keyword exists in the reference data.

In one embodiment, the step of determining an emotion for each speaker based on an emotion determination keyword depending on whether the predetermined emotion keyword exists is performed when the predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database. It may include extracting the reference count for each speaker and determining the emotion corresponding to the emotion determination keyword according to the reference count of reference data for each speaker.

According to the present invention as described above, there is an advantage in that it is possible to analyze multi-party call voice data, separate speakers, and then provide emotions between speakers.

In addition, the present invention provides that if a predetermined keyword is present in each call text data converted from multi-party call voice data to call text, the text data can be modified using reference data corresponding to the text keyword to generate modified text data. The purpose is to provide a complex emotion analysis method using speaker separation STT for multi-party calls and a system for executing the same.

Figure 1 is a network configuration diagram for explaining a complex emotion analysis system using STT according to an embodiment of the present invention.
Figure 2 is a block diagram for explaining an embodiment of a complex emotion analysis server according to an embodiment of the present invention.
Figure 3 is a flowchart illustrating an embodiment of a complex emotion analysis method using STT according to the present invention.

The above-mentioned objects, features, and advantages will be described in detail later with reference to the attached drawings, so that those skilled in the art will be able to easily implement the technical idea of the present invention. In describing the present invention, if it is determined that a detailed description of known technologies related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings. In the drawings, identical reference numerals are used to indicate identical or similar components.

Figure 1 is a network configuration diagram for explaining a complex emotion analysis system using STT according to an embodiment of the present invention.

Referring to FIG. 1, a complex emotion analysis system using STT includes a first user terminal 100, a second user terminal 200, and a complex emotion analysis server 300.

The first user terminal 100 is a terminal owned by the first user who makes a call with the user who owns the second user terminal 200 when a call is made to the second user terminal 200.

The first user terminal 100 may be a smart phone, a portable terminal, a mobile terminal, a personal digital assistant (PDA), a portable multimedia player (PMP) terminal, or a personal computer. It includes various terminals such as personal computers, laptop computers, and tablet PCs.

The second user terminal 200 is a terminal owned by a second user who makes a call with the user who owns the first user terminal 100 when a call is made to the first user terminal 100 .

The second user terminal 200 may be a smart phone, a portable terminal, a mobile terminal, a personal digital assistant (PDA), a portable multimedia player (PMP) terminal, or a personal computer. It includes various terminals such as personal computers, laptop computers, and tablet PCs.

When a call is connected between the first user terminal 100 and the second user terminal 200, the complex emotion analysis server 300 records call voice data between the first user terminal 100 and the second user terminal 200. It is a server that analyzes emotions after analyzing them to separate the first and second users.

For this purpose, the complex emotion analysis server 300 includes a keyword database in which reference data for each keyword is stored. Keywords can be voice or text, and reference data can be implemented as text.

For example, the complex emotion analysis server 300 may implement keywords as text such as “Siul, Soul, Soul, Siiul, Soul,” and reference data as text such as “Seoul.”

For another example, when the complex emotion analysis server 300 includes a keyword database in which keywords are implemented as text and reference data is implemented as text, the text keyword is text such as “Siul, Soul, Siiul, Soul.” It is implemented as, and the reference data can be implemented as text such as “Seoul”.

As described above, the complex emotion analysis server 300 generates a voice keyword by implementing each word with different pitch, length, intonation, and pronunciation using a word dictionary database, and then generates a text keyword based on the voice keyword. do. Then, the complex emotion analysis server 300 creates a keyword database by matching and storing each of the voice keywords and the text keywords with the corresponding words as reference data.

That is, if the word is text, the complex emotion analysis server 300 creates a keyword database by matching and storing text keywords and reference data, and matches and stores voice keywords and reference data generated by converting the word into a standard voice. to create a keyword database.

The reason for creating a keyword database through this process is that even the same word can be uttered in a different voice with different pitch, length, intonation, and pronunciation for each user who speaks it.

Hereinafter, a process in which the complex emotion analysis server 300 generates modified text data using call voice data will be described.

In one embodiment, the complex emotion analysis server 300 converts call voice data into call text data based on a keyword database and then checks whether a text keyword exists in the call text data.

If a text keyword exists in the call text data based on the keyword database, the complex emotion analysis server 300 extracts reference data corresponding to the text keyword from the keyword database and then modifies the text data using the reference data to create a modified text. Data can be generated.

For example, when the text keyword is “What do you do in Siul?”, the complex emotion analysis server 300 modifies the call text data based on the reference data “Seoul” corresponding to the text keyword “Seoul” in the keyword database. You can create modified text data “What do you do in Seoul?”

In another embodiment, the complex emotion analysis server 300 determines whether a voice keyword exists in call voice data based on a keyword database.

If a voice keyword exists in the call voice data, the complex emotion analysis server 300 extracts reference data corresponding to the voice keyword from the keyword database and then modifies the text data based on the reference data to generate modified text data. You can. At this time, the complex emotion analysis server 300 stores the frequency characteristics of the voice data corresponding to the voice keyword among the call voice data in the second reference database, and later, when the emotion corresponding to the call voice data is determined, the voice data Frequency characteristics and emotions are stored in correspondence.

For example, when the call voice data received from the customer company server 200 is “Soul, right?”, the complex emotion analysis server 300 uses reference data corresponding to the voice keyword “Soul” in the keyword database, “Seoul.” ” Based on this, the call text data is modified to generate modified text data “Seoul, right?”, and the frequency characteristics corresponding to “Soul” are stored in the second reference database.

In the above embodiment, when there is no text keyword or voice keyword pre-stored in the keyword database in the call voice data, the complex emotion analysis server 300 refers to the keyword database and creates a similar message on the call voice data or call text data. Check whether the keyword exists.

For example, if the call voice data is “How do I get there?” and there is no voice keyword pre-stored in the keyword database, the complex emotion analysis server 300 uses the voice keyword “Seaul” as a similar keyword in the keyword database. It is judged that

If a similar keyword exists in the call voice data or call text data with reference to the keyword database, the complex emotion analysis server 300 modifies the call text data using the similar keyword and generates modified text data.

Then, the complex emotion analysis server 300 updates the keyword database using data matching similar keywords in call voice data or call text data.

For example, if the call text data generated based on call voice data is “How do I get to Siuuuul?” and the similar keyword is “Siuul,” the complex emotion analysis server 300 uses the similar keyword “Siuul” and Update the keyword database using the matching call text data “Siuuuul.”

Therefore, if the voice keyword “Seoul, Soul, Siul, Soul” and the reference data “Seoul” are stored in the original database, the voice keyword “Seoul, Siul, Soul, Siuul, Soul” and the reference data “Seoul” are stored in the original database. The keyword database is updated with the data “Seoul”.

As described above, the complex emotion analysis server 300 converts the call voice data into call text data and then modifies it based on the keyword database to calculate the number of times for each reference data referenced when generating the modified text data in the first reference database for each speaker. Save it to

The reason for this is to determine the modified text data as an emotion corresponding to the predetermined emotion determination keyword according to the number of references to the reference data when the reference data referenced is a predetermined emotion determination keyword.

The complex emotion analysis server 300 compares reference data for each speaker stored in the first reference database and predetermined emotion keywords, and if a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database, a reference for each speaker. The reference count of the data is extracted, and the emotion corresponding to the emotion determination keyword is determined according to the reference count of the reference data for each speaker.

In one embodiment, the complex emotion analysis server 300 extracts the reference count for each speaker when a predetermined emotion keyword is present among the reference data for each speaker stored in the first reference database, and adds the reference count for each speaker to the reference count of the reference data for each speaker. Accordingly, the emotion corresponding to the relevant emotion decision keyword can be determined.

In another embodiment, when the same reference data exists among the speaker-specific reference data stored in the first reference database, the complex emotion analysis server 300 determines the emotion corresponding to the emotion determination keyword according to the reference number of the same reference data. It can be decided based on the feelings between the speakers.

Unlike the above and the embodiment, the complex emotion analysis server 300 performs frequency analysis on the call voice data between the first user terminal 100 and the second user terminal 200 to extract frequency characteristics, and extracts the frequency characteristics of the call voice data. The frequency feature and the frequency feature stored in the second reference database are compared, and according to the comparison result, the speaker's emotion is determined as the emotion corresponding to the predetermined emotion determination keyword.

The above emotion-determining keywords are core elements of emotional experience and refer to the degree to which positive emotions (or pleasure) and negative emotions (or discomfort) are experienced, as well as information about the external world (whether it is beneficial or harmful to life, The process of converting an emotion (or reward or punishment) into an internal emotional signal or state to respond appropriately (approach or avoidance) to an object or situation can be defined as a core affect. In terms of motivation, Pleasure is defined as an emotional experience based on the motivation to approach things that are beneficial to life, and displeasure is defined as an emotional experience based on the motivation to avoid things that are harmful.

The reason why pleasure-displeasure is the core of emotional experience is because it is an emotion that everyone experiences universally and is a primal emotion that does not require special learning, and this experience is consistent through objective indicators such as face, voice, and body in subjective verbal expressions. It has the characteristic of appearing clearly. Looking at the results of previous studies that analyzed everyday vocabulary in two dimensions of polarity and arousal, emotional expression vocabulary tends to be polarized into positive and negative, rather than being evenly distributed along a continuum from negative to positive.

Accordingly, the complex emotion analysis server 300 uses a keyword database derived for vocabulary including adjectives, verbs, and nouns as a measure of representative emotional expression vocabulary as a result of analyzing the dimensions of the Korean emotional vocabulary.

In order to create such a keyword database, emotional vocabularies must be extracted and polarity values for the extracted emotional vocabularies must be derived. Three stages of processing can be performed to extract emotional vocabularies. First, the task of removing stop words such as English, numbers, Korean-character vocabulary, and special characters is performed. For the remaining vocabulary, words with values below a certain standard can be removed based on the TF and TF-IDF values, and finally, identity The vocabulary to be used in the emotional dictionary can be finalized through the task of assigning . The emotional polarity value for the vocabulary extracted in this way can be defined as the positive/negative recall ratio in which the emotional vocabulary appears.

When emotions are analyzed and defined through voice and text, a user adaptation process is performed as described above. No matter how objectively you analyze emotions using big data, individual deviations may occur, so you go through an adaptation process and a learning process. After learning individual deviations through feedback and artificial neural networks, the learning results can be updated in advance to learn and update feature vectors of vocabulary and pitch.

Figure 2 is a block diagram for explaining an embodiment of a complex emotion analysis server according to an embodiment of the present invention.

Referring to FIG. 2, the complex emotion analysis server 300 includes an STT conversion unit 310, a keyword database 320, a modified data generation unit 330, a first reference database 340, and a second reference database 350. , includes an emotion determination unit 360 and a control unit 370.

The STT converter 310 converts call voice data between users into call text data in real time when a plurality of user terminals are connected.

The keyword database 320 stores reference data for each keyword. At this time, the keyword may be voice or text, and the reference data may be implemented as text.

For example, keywords stored in the keyword database 320 may be implemented as text such as “Siul, Soul, Soul, Siiul, Soul,” etc., and reference data may be implemented as text such as “Seoul.”

For another example, if the keyword in the keyword database 320 is implemented as text and the reference data is implemented as text, the text keyword is implemented as text such as “Siul, Soul, Siiul, Soul”, and the reference data is It can be implemented as text such as “Seoul”.

As described above, the present invention uses a word dictionary database to generate voice keywords by implementing each word with different pitch, length, intonation, and pronunciation, and then generates text keywords based on the voice keywords. Then, the present invention creates a keyword database by matching and storing each of the voice keywords and the text keywords as reference data.

That is, if the word is text, the text keywords and reference data are matched and stored in the keyword database 320, and the voice keywords and reference data generated by converting the word into a standard voice are matched and stored in the keyword database 320. .

The reason for creating a keyword database through this process is that even the same word can be uttered in a different voice with different pitch, length, intonation, and pronunciation for each user who speaks it.

The modified data generator 330 modifies the call voice data and the call text data based on the keyword database 320 to generate modified data.

In one embodiment, the modified data generator 330 converts call voice data into call text data based on the keyword database 320 and then checks whether a text keyword exists in the call text data.

If a text keyword exists in the call text data based on the keyword database 320, the modified data generator 330 extracts reference data corresponding to the text keyword from the database and then modifies the text data using the reference data. Modified text data can be created.

For example, when the text keyword is “What do you do in Siul?”, the modified data generation unit 330 modifies the call text data based on the reference data “Seoul” corresponding to the text keyword “Seoul” in the database. You can create modified text data “What do you do in Seoul?”

In another embodiment, the modified data generator 330 determines whether a voice keyword exists in the call voice data based on the keyword database 320.

If a voice keyword exists in the call voice data, the modified data generator 330 may extract reference data corresponding to the voice keyword from the database and then modify the text data based on the reference data to generate modified text data. there is.

At this time, the frequency characteristics of the voice data corresponding to the voice keyword among the call voice data are stored in the second reference database 350, and later, when the emotion corresponding to the call voice data is determined, the frequency characteristic and emotion of the voice data correspond to the corresponding voice data. and stored in the second reference database 350.

For example, when the call voice data is “Soul, right?”, the correction data generator 330 modifies the call text data based on the reference data “Seoul” corresponding to the voice keyword “Soul” in the database. Thus, modified text data “Seoul?” is generated, and the frequency characteristic corresponding to “Soul” is stored in the second reference database 350.

In the above embodiment, if the text keyword or voice keyword pre-stored in the database does not exist in the call voice data, the correction data generator 330 refers to the database and creates a similar keyword in the call voice data or call text data. Check whether it exists or not.

For example, if the call voice data is “How do I get to Siwoowooul?” and there is no voice keyword pre-stored in the database, the modified data generation unit 330 determines that the voice keyword “Siwooul” is a similar keyword in the database. do.

If a similar keyword exists in the call voice data or call text data with reference to the keyword database 320, the modified data generator 330 modifies the call text data using the similar keyword and generates modified text data. do.

Then, the modified data generator 330 updates the keyword database 320 using data matching similar keywords in the call voice data or call text data.

For example, if the call text data generated based on the call voice data is “How do I get to Siuuuul?” and the similar keyword is “Siuul,” the modified data generation unit 330 may use the similar keyword “Siuul” and The keyword database 320 is updated using the matching call text data “Siuuuul.”

Therefore, if the voice keyword “Seoul, Soul, Siul, Soul” and the reference data “Seoul” are stored in the original database, the voice keyword “Seoul, Siul, Soul, Siuul, Soul” and the reference data “Seoul” are stored in the original database. The keyword database 320 is updated with the data “Seoul”.

The number of times each reference data referenced when generating the modified text data is stored in the first reference database 340 for each speaker.

That is, when call voice data is converted into call text data and modified based on the database to generate modified text data, the number of times each reference data is referenced is stored in the first reference database 340 for each speaker.

The reason for this is to determine the modified text data as an emotion corresponding to the predetermined emotion determination keyword according to the number of references to the reference data when the reference data referenced is a predetermined emotion determination keyword.

The emotion determination unit 360 compares the reference data for each speaker stored in the first reference database 340 and the predetermined emotion keyword, and determines whether the predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database 340. In this case, the reference count of the reference data for each speaker is extracted, and the emotion corresponding to the emotion determination keyword is determined according to the reference count of the reference data for each speaker.

In one embodiment, the emotion determination unit 360 extracts the reference count for each speaker when a predetermined emotion keyword is present among the reference data for each speaker stored in the first reference database 340, and refers to the reference data for each speaker. Depending on the number of times, the emotion corresponding to the relevant emotion decision keyword can be determined.

In another embodiment, if the same reference data exists among the reference data for each speaker stored in the first reference database, the emotion determination unit 360 determines the emotion corresponding to the corresponding emotion determination keyword according to the number of references to the same reference data. It can be decided based on the feelings between the speakers.

Unlike the above and the embodiment, the emotion determination unit 360 performs frequency analysis on the call voice data between the first user terminal 100 and the second user terminal 200 to extract frequency characteristics and frequency of the call voice data. The feature and the frequency feature stored in the second reference database 350 are compared, and according to the comparison result, the speaker's emotion is determined as the emotion corresponding to the predetermined emotion determination keyword.

Figure 3 is a flowchart illustrating an embodiment of a complex emotion analysis method using STT according to the present invention.

Referring to FIG. 3, when a plurality of user terminals are connected to a call, the complex emotion analysis server 300 converts call voice data between users into call text data in real time (step S310).

The complex emotion analysis server 300 generates modified data by modifying the call voice data and the call text data based on a predetermined database (step S320).

The complex emotion analysis server 300 stores the number of times each reference data referenced when generating modified text data in the first reference database for each speaker (step S330).

The complex emotion analysis server 300 determines the emotion for each speaker based on the emotion determination keyword depending on whether a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database 340 (step S340).

Although the present invention has been described with reference to limited embodiments and drawings, the present invention is not limited to the above embodiments, and various modifications and variations can be made by those skilled in the art from these descriptions. Accordingly, the spirit of the present invention should be understood only by the scope of the claims set forth below, and all equivalent or equivalent modifications thereof shall fall within the scope of the spirit of the present invention.

100: first user terminal,
200: second user terminal,
300: Complex emotion analysis server
310: STT conversion unit,
320: keyword database,
330: correction data generation unit,
340: first reference database,
350: secondary reference database,
360: emotional decision part,
370: control unit

Claims (5)

An STT conversion unit that converts call voice data between users into call text data in real time when a plurality of user terminals are connected to a call;
a correction data generator that generates correction data by modifying the call voice data and the call text data based on a predetermined database;
If a text keyword exists in the call text data based on the keyword database, the modified data generator extracts reference data corresponding to the text keyword from the database and then modifies the text data using the reference data to generate modified text data. If a voice keyword exists in the voice data, reference data corresponding to the voice keyword is extracted from the database, and the text data is modified based on the reference data to generate modified text data.
The correction data generator is configured to check whether a similar keyword exists in the call voice data or call text data by referring to the database when the text keyword or voice keyword pre-stored in the database does not exist in the call voice data. Contains more,
a first reference database generator that stores the number of times for each reference data referenced when generating the correction data in a first reference database for each speaker; and
an emotion determination unit that determines an emotion for each speaker based on an emotion determination keyword depending on whether a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database;
The emotion determination unit extracts frequency characteristics by analyzing the frequency of call voice data between the first user terminal and the second user terminal, and compares the frequency features of the call voice data with the frequency features stored in the second reference database according to the comparison result. A complex emotion analysis server using speaker separation STT for a multi-party call, characterized in that the emotion corresponding to a predetermined emotion determination keyword is determined by the speaker's emotion.
According to claim 1,
The appraisal decision department
Compare reference data for each speaker stored in the first reference database and predetermined emotion keywords, and if a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database, extract the reference number of the reference data for each speaker; , A complex emotion analysis server using speaker separation STT of a multi-party call, comprising determining the emotion corresponding to the emotion determination keyword according to the reference number of reference data for each speaker.
According to paragraph 1 or 2,
The complex emotion analysis method of the complex emotion analysis server using STT is:
When a plurality of user terminals are connected to a call, converting call voice data between users into call text data in real time;
generating modified text data by modifying the call voice data and the call text data based on a predetermined database;
storing the number of times for each reference data referenced when generating the correction data in a first reference database for each speaker;
A step of determining an emotion for each speaker based on an emotion determination keyword depending on whether a predetermined emotion keyword exists among reference data for each speaker stored in the first reference database;
The step of determining an emotion for each speaker based on the emotion decision keyword depending on whether the predetermined emotion keyword exists is
A complex emotion analysis method using speaker separation STT for a multi-party call, further comprising the step of extracting the reference count for each speaker when a predetermined emotion keyword exists among the reference data for each speaker stored in the first reference database.
According to claim 3,
The step of determining an emotion for each speaker based on the emotion decision keyword depending on whether the predetermined emotion keyword exists is
A complex emotion analysis method using speaker separation STT for a multi-party call, further comprising the step of determining the emotion corresponding to the emotion determination keyword according to the reference number of reference data for each speaker.
According to claim 3,
The step of determining an emotion for each speaker based on the emotion decision keyword depending on whether the predetermined emotion keyword exists is
If the same reference data exists among the reference data for each speaker stored in the first reference database, determining the emotion corresponding to the emotion determination keyword as the emotion between speakers according to the number of references to the same reference data. A complex emotion analysis method using speaker separation STT in a multi-party call.