KR20190023749A - Apparatus and method for emotion recognition of user - Google Patents

Abstract

The present invention is to provide an apparatus and a method for recognizing a user emotion, which can accurately determines a user emotional state by adjusting the center and width of each input membership function of an adaptive neuro-fuzzy inference unit and a rule base of the input and output through backpropagation learning of a variety of input information for determining the user emotional state. The apparatus of the present invention comprises: a fuzzy logic classifying unit for providing user emotional state information analyzed based on a facial expression of a user as a membership function for input of an adaptive neuro-fuzzy inference unit; an emotional state analyzing unit for providing the user emotional state information obtained by analyzing a voice and an image of the user with a predetermined emotional state program as the membership function for input of the adaptive neuro-fuzzy inference unit; and an adaptive neuro-fuzzy inference unit for calculating an output for input information provided by the fuzzy logic classifying unit and the emotional state analyzing unit, performing a learning process to allow a root mean square (RMS) error value to be a preset reference value or less by adjusting a parameter to reduce an error of a calculation value and a target output, and recognizing a user emotional state using a learning result. Therefore, according to the present invention, a user emotional state can be accurately determined by adjusting the center and width of each input membership function and a rule base of the input and output through backpropagation learning of a variety of input information for determining the user emotional state.

Description

[0001] APPARATUS AND METHOD FOR EMOTION RECOGNITION OF USER [0002]

The present invention relates to an apparatus and method for recognizing user's emotions, and more particularly, to an apparatus and method for recognizing user's emotions, And more particularly, to a user emotion recognition apparatus and method for accurately determining a user's emotion state by adjusting a rule base of input and output.

2. Description of the Related Art In recent years, in order to facilitate user's convenience in life, development of a robot field for supporting a user's intended work has been actively performed.

In particular, there is a growing interest in the field of intelligent robots that enables a robot to intelligently determine the interaction between a user and a robot and to perform an operation according to the result.

Currently developed intelligent robots are being developed to generate and express emotions that are more similar to humans, such as generating sensations according to predefined conditions or expressing corresponding behaviors using simple external sensor information.

In addition, intelligent robots are actively studying forms that can interact with humans by reflecting emotional states of human beings through images captured through cameras and reflecting them on emotional generation.

As described above, it is difficult to grasp the emotion of the human being by using the recognition information of the image photographed by the camera, or to grasp the intensity, tempo, and intonation of the voice through the voice recognition using the microphone, There is a limit.

Korean Patent Registration No. 10-0850352 (entitled " Apparatus and method for emotion expression of intelligent robot for expressing emotion using state information "

In order to solve such a problem, the present invention adjusts the center and width of each input membership function of the adaptive neuro fuzzy reasoning unit and the rule base of input and output through various kinds of input information for judging the emotional state of the user, And to provide a user emotion recognition apparatus and method that accurately determine the emotion state of a user.

According to an aspect of the present invention, there is provided a fuzzy logic classification unit comprising: a fuzzy logic classifier for providing emotion state information of a user based on facial expressions of a user as a membership function for input of an adaptive neuro fuzzy reasoning unit; An emotion state analyzing unit for providing emotion state information of a user who has analyzed a user's voice and an image with a preset emotion state program as a belonging function to the input of the adaptive neuro fuzzy reasoning unit; And calculating an output of the input information provided by the fuzzy logic classification unit and the emotion state analyzing unit and adjusting a parameter to reduce an error between the calculated value and a target output to obtain a root mean square error (RMS) The learning process is performed so that the value of the learning result is less than a predetermined reference value and the emotion state of the user is recognized using the learning result.

In addition, the user's emotional recognition apparatus according to the present invention detects at least one of a user's atrial pulse rate, a user's touching and touching time, and a user's operation speed and outputs the adaptive neuro fuzzy inference function as a membership function for the user's emotional state. And a sensor unit for outputting a signal.

In addition, the facial expression of the fuzzy logic classification unit according to the present invention is characterized by including a position of an eyebrow, a shape of a mouth, a position and a direction of a head, and a position and direction of an eye.

The fuzzy logic classification unit according to the present invention is characterized by extracting an emotional state by fuzzification according to a rule base of a preset input and output facial expressions.

Further, the fuzzy logic classifier according to the present invention is characterized in that the emotion state deduced through fuzzy logic is defuzzified and output as a general numerical value.

Further, the fuzzy logic classification unit according to the present invention may be configured to classify the user's emotional state into Ang, Fear, Sad, Nor, Sha, Sup, Joy, As shown in FIG.

Further, the adaptive neuro fuzzy inference unit according to the present invention performs a learning process of rearranging the input and output rule bases by adjusting the center and width of each input belonging function.

In addition, the adaptive neurofuzzy inference unit according to the present invention stores the center value, the width value, and the weight value of the forehead part of the forehead part belonging function at the end of the learning, and uses the initialization data in the process of performing the restart or re- .

Further, the adaptive neurofuzzy reasoning unit according to the present invention performs a backpropagation learning process.

According to another aspect of the present invention, there is provided a method of analyzing an input / output data, comprising the steps of: (a) defining a fuzzy logic classification unit and an emotion state analysis unit by defining input / output data pairs; b) initializing an adaptive neurofuzzy inference membership function and detecting an average root mean square error (RMSE) value by receiving the input / output data defined in the step a); c) performing the learning using the adaptive neurofuzzy reasoning part backpropagation learning algorithm; d) Upon completion of the learning, the adaptive neurofuzzy reasoning unit adjusts the center and width of the forehead part belonging function and the linear coefficient value of the backward part belonging function, and determines whether or not the mean square root error value is less than a preset reference value step; And e) storing the center and the width of the adjusted forehead part belonging function and the linear coefficient values of the backward part belonging function in the database if the mean square error value is less than the reference value as a result of the determination.

The end of learning in step d) according to the present invention is completed when the generated mean square error value is less than a preset convergence condition value.

In the step e) of the present invention, if the mean square error value is equal to or greater than a reference value, the step c) and the step d) are performed again by adjusting the learning rate.

In addition, the center and the width of the forehead part belonging function stored in step (e) according to the present invention and the linear coefficient value of the backward part belonging function may be set such that, when the adaptive neurofuzzy reasoning part restarts or re- Is used.

Further, the step (e) according to the present invention may further include the step of storing input / output data inputted in the emotion recognition service process after learning is completed in real time, and the stored input / output data may be further learned in a pauser of the emotion recognition service .

The present invention adjusts the center and the width of each input membership function of the adaptive neuro fuzzy reasoning unit and the rule base of input and output through various kinds of input information for determining the emotional state of the user through back propagation learning, There is an advantage to be able to judge.

1 is a block diagram showing a user emotion recognition apparatus according to the present invention.
FIG. 2 is a block diagram showing a fuzzy logic classification unit of the user emotion recognition apparatus according to FIG. 1. FIG.
3 is a view showing an example of a membership function of an eyebrow state input fuzzy variable inputted to the user's emotional recognition apparatus according to the present invention.
FIG. 4 illustrates an example of a membership function of an output fuzzy variable of the user emotion recognition apparatus according to the present invention. FIG.
5 is a view illustrating a screen for defining a fuzzy rule base for an eyebrow state of a user's emotional recognition apparatus according to the present invention.
FIG. 6 is an exemplary view showing a three-dimensional space of the fuzzy rule base with respect to the eyebrow state according to FIG. 5;
FIG. 7 is an exemplary view showing an example of a result of non-fuzzification of the eyebrow state according to FIG. 5;
8 is a diagram illustrating a screen for defining a fuzzy rule base with respect to a mouth state of the user emotion recognition apparatus according to the present invention.
FIG. 9 is a diagram illustrating a screen for defining a fuzzy rule base for a head state of a user emotion recognition apparatus according to the present invention; FIG.
10 is a diagram illustrating a screen for defining a fuzzy rule base for an eye state of a user emotion recognition apparatus according to the present invention;
11 is a view showing an example of a membership function of a fuzzy logic classification unit used as an input of the adaptive neuro fuzzy reasoning unit of the user emotion recognition apparatus according to the present invention.
FIG. 12 is an exemplary view showing a membership function of a speech recognition result used as an input of the adaptive neuro fuzzy reasoning unit of the user emotion recognition apparatus according to the present invention; FIG.
FIG. 13 illustrates an example of a membership function of an image recognition result used as an input of the adaptive neuro fuzzy inference unit of the user emotion recognition apparatus according to the present invention. FIG.
FIG. 14 is an exemplary view showing a membership function of the heart rate used as an input of the adaptive neuro fuzzy reasoning unit of the user emotion recognition apparatus according to the present invention. FIG.
15 is a flowchart illustrating an operation process of the user emotion recognition apparatus according to the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings.

FIG. 1 is a block diagram showing a user emotion recognition apparatus according to the present invention, and FIG. 2 is a block diagram showing a fuzzy logic classification unit of the user emotion recognition apparatus according to FIG.

1 and 2, the user emotion recognition apparatus according to the present invention includes a fuzzy logic classification unit 100, an emotion state analysis unit 200, an adaptive neurofuzzy reasoning unit 300, (400), a database (500), and a re-learning unit (700).

The fuzzy logic classifier 100 provides the user's emotion state information analyzed based on the user's facial expression as a membership function for the input of the adaptive neurofuzzy inference unit 300, The user's emotional state using the facial expression including the user's eyebrow position, mouth shape and position, head position and direction, and eye position and direction detected from the three-dimensional image sensor, etc. So that the information can be provided.

In addition, the fuzzy logic classifier 100 fuzzifies the facial expression according to a preset rule base of input and output so that the emotional state of the user can be extracted.

In addition, the fuzzy logic classifier 100 performs defuzzification of the inferred emotion state through fuzzification, and outputs it as a general numerical value, so that different input data spaces can be processed in the same space.

In addition, the fuzzy logic classifier 100 may classify the user's emotional state into Ang, fear (Fer) using the information of eyebrow position, mouth shape and position, head position and direction, ), Sad, Nor, Sha, Sur, and Joy.

The above-mentioned emotion states may be classified into more specific emotions by adding fuzzy rules and membership functions so as to distinguish subtle emotion states in addition to the seven emotion states.

The fuzzy logic classifier 100 extracts a face of a user using a three-dimensional image sensor or the like, extracts a user's eyebrow state (Expression Brow), a mouth state Expression Mouth, Expression Head, Expression Eye, and determines the state of emotion based on the obtained user facial expression, and the result is calculated by the following equation 1 to provide an input to the adaptive neuro fuzzy reasoning unit 300. The eyebrow state classifying unit 110, the mouth state classifying unit 120, the head state classifying unit 130, And a state classifying unit 140.

Where Y is the output vector of the fuzzy inference engine, W is the weight vector, B is the eyebrow condition, M is the mouth condition, H is the head condition, and E is the eye condition.

The eyebrow condition classifying unit 110 operates to extract the emotion state using the state of the eyebrows in the facial expression.

3 (a) and 3 (b) illustrate membership functions of the eyebrow state input fuzzy variables input to the user's emotional recognition apparatus. FIG. 3 (a) FIG. 3 (c) is a membership function according to a state where the left eyebrow is descending, and FIG. 3 (d) is a membership function according to a state where the right eyebrow is descending.

3 (a) to 3 (d) can be defined by dividing the belonging function into three types through the eyebrow state classifying unit 110, and S (Small), M (Medium) L (Large) notation.

The user facial expression information is defined as shown in Table 1 below.

Face area Variable name definition meaning
Eyebrow condition
(EXP_BROW)
BRL (Brow Raiser Left) Left Eyebrow Up BRR (Brow Raiser Right) Right eyebrow up BLL (Brow Lowerer Left) Left Eyebrow Down BLR (Brow Lowerer Right) Right eyebrows down
Mouth condition
(EXT_MOUTH)
Mouth (Smooth) Smile of face MTK (Mouth Kiss) Kiss action MTO (Mouth Open) Open state TGO (Tongue Out) State of tongue

Head condition
(EXP_HEAD)

HTL (Head Turn Left) Head left direction HTR (Head Turn Right) Head in right-handed orientation HUP (Head Up) The head is facing upward HDN (Head Down) Head downward HIL (Head tilt Left) Head tilted left HIR (Head tlt Right) Head tilted to the right

Eye condition
(EXP_EYE)

ECL (Eye Close Left) Left eye closed Eye Close Right (ECR) Right eye closed Eye Turn Left (ETL) Both eyes moved to the left Eye Turn Right (ETR) Both eyes moved to the right EUP (Eye UP) Both pupils moved up EDN (Eye Down) Both eyes moved down

FIG. 4 shows an example of the belonging function of the eyebrow state output. The emotion states deduced through the fuzzy processing are shown as Ang, F, Sad, Nor, Sha, (Sup), and joy (Joy), so that each emotion state can be output.

5 is a diagram illustrating a screen for defining a fuzzy rule base for the eyebrow state of the user's emotional recognition apparatus. In the rule base setting screen 600, a user defines an input condition definition 610 for defining the state of eyebrows A plurality of conditional statements 620 corresponding to the state of the eyebrows are set and the input condition definition 610 may represent the conditional statements set by the fuzzy rule base for the eyebrow condition as a three dimensional space as shown in FIG.

FIG. 7 illustrates an example of a result of non-fuzzification of eyebrows in a sad emotional state. The emotion state deduced through fuzzification is defuzzified according to each emotional state to obtain a general numerical value 630 Output.

8 is a diagram illustrating an example of a screen for defining a fuzzy rule base for the eyebrow state of the user's emotional recognition apparatus. Referring to FIG. 8, We define them by using S (Small), M (Medium), and L (Large) notation. The emotion states deduced from fuzzy are defined as Ang, , The definition of input condition is defined so that each emotional state can be provided through the output membership function by defining seven kinds of sadness (Sad), normal (Nor), shame (Sha), surprise (Sup), and joy Set conditional statements through.

9 is a diagram illustrating a screen for defining a fuzzy rule base for the head state of the user's emotional recognition apparatus. The head position classification unit 130 divides and defines membership functions according to the head state into three types , S (Small), M (Medium), and L (Large) notation, and the emotion states deduced through fuzzing are defined as Ang, Fear, Sad, , Shame (Sha), surprise (Sup), and joy (Joy), and conditional statements are defined through input condition definition so that each emotion state can be provided through the output membership function.

10 is a diagram illustrating a screen for defining a fuzzy rule base with respect to an eye state of the user's emotional recognition apparatus. The fuzzy rule base according to the eye state is classified into three types and defined through the eye state classifying unit 140 , S (Small), M (Medium), and L (Large) notation, and the emotion states deduced through fuzzing are defined as Ang, Fear, Sad, , Shame (Sha), surprise (Sup), and joy (Joy), and conditional statements are defined through input condition definition so that each emotion state can be provided through the output membership function.

The emotion state analyzing unit 200 provides emotion state information of a user who has analyzed a user's voice and image as a preset emotion state program as a membership function for the input of the adaptive neurofuzzy inference unit 300, For example, when a user's conversation voice is applied to a microphone installed in an interface device (e.g., an interactive robot), it is processed into digital voice streaming data and transmitted to the interface device or a remote server (e.g., And receives the result of analyzing the emotional state of the talker using the voice based on the delivered voice information in the emotional state analysis program installed in the interface device or the server and transmits the analyzed result to the belonging of the adaptive neurofuzzy reasoning unit 300 Function.

When the user's face is photographed by a camera installed in the interface device, the user's face is processed into digital image streaming data and transmitted to the server of the interface device or the remote site, The emotion state analysis program installed in the server receives the result of analyzing the emotional state of the talker using the image based on the delivered image information and provides the result as a belonging function of the adaptive neuro fuzzy reasoning unit 300. [

The adaptive neurofuzzy inference unit 300 may include a fuzzy rule for the input membership function information as shown in FIGS. 11 to 13 provided in the fuzzy logic classification unit 100 and the emotion state analysis unit 200, The output is calculated using a reasoning model composed of a neural network structure and a parameter is adjusted so that the error between the calculated value and the target output is reduced to obtain a mean square root error Root Mean Square (RMS) value is less than a preset reference value, and recognizes the emotion state of the user using the learning result.

The reasoning model is divided into a premise part and a consequence part. The predicate part includes a layer for setting input variables as a whole set, a membership function for each input variable of the predicate part, And outputs the degree of belonging when the input is given, and the latter calculates the final inference value for generating the actual control amount.

The adaptive neuro fuzzy inference unit 300 performs a learning process through a back propagation learning program. The adaptive neuro fuzzy inference unit 300 calculates an actual output of the input data through a forward step, and performs an inverse step Adjusts the adjustment parameters, ie the position and width of the total key belonging function and the coefficients of the post-key line type.

That is, the adaptive neurofuzzy inference unit 300 performs a learning process of rearranging the input and output rule bases by adjusting the center and width of each input membership function, and when learning ends, The width value, and the weight value of the background portion, and is used as the initialization data in the process of restarting or re-learning.

Meanwhile, the user's emotional recognition apparatus according to the present invention detects at least one of a user's atrial heart rate, a user's touch and touch time, and an operation speed of a user, and detects the adaptive neuro fuzzy inference function as a membership function for a user's emotional state. And a sensor unit (400) for outputting the signal to the controller (300).

The sensor unit 400 is a sensor capable of detecting a user's heart rate, contact, etc., and a sensor capable of measuring a heartbeat of a user is installed in an interface device or an interactive robot to measure a user's heartbeat , The heart rate of the user is measured and defined as a membership function so as to be provided as an input membership function of the adaptive neuro fuzzy reasoning unit 300 as shown in FIG.

The sensor unit 400 is a sensor capable of detecting whether or not a user touches the robot. A sensor capable of detecting whether or not a user touches the robot is installed in an interface device or an interactive robot, such as a head, a shoulder, If the user touches the touch panel, it can be provided as an input membership function of the adaptive neurofuzzy inference unit 300 by defining a membership function as input data using touch or touch time.

Also, when two or more sensors are used, the average value of sensor touch time is calculated and used.

The database 500 stores the emotion information of the user. The database 500 stores information about the subject's belonging function (for example, the center and width of the belonging function) and the weight ) Values and the like.

If the re-learning unit 700 stops the use of the user and the re-learning is performed arbitrarily by increasing the learning condition set by the administrator, the preset automatic re-learning condition, or the user's unsatisfactory service level, the adaptive neuro- And requests the unit 300 to proceed with the re-learning.

The operation of the user emotion recognition apparatus according to the present invention will now be described.

The user defines and prepares input / output data pairs of the fuzzy logic classification unit 100 and the emotion state analysis unit 200 (S100), and the adaptive neurofuzzy reasoning unit 300 initializes the belonging function (S110).

After performing step S110, the adaptive neurofuzzy inference unit 300 receives an input / output data pair defined in step S100 (S120) and detects an average square root error (RMSE) value in step S130.

The adaptive neuro fuzzy reasoning unit 300 adjusts the parameter to reduce the error between the value calculated in step S130 and the target output so that the root mean square error (RMS) (S140) using a propagation learning algorithm.

Through the learning process by the back propagation learning algorithm in step S140, the adaptive neuro fuzzy inference unit 300 calculates the actual output of the input data through the forward step and performs an inverse step to minimize the error with respect to the target output (S150), and adjusts the linear coefficient value of the backward key belonging function (S160), and determines whether the average square root error value is equal to or less than a preset reference value (S170).

If it is determined in step S170 that the average square root error value in step S170 is less than the reference value, the central position, width value, and weight value of the forehead part of the forehead part belonging function adjusted in steps S150 and S160 are stored If the average square root error value in step S170 is equal to or greater than the reference value, the preset learning rate is adjusted (S200), and the steps S130 to S160 Allow the step to be run again.

After the step S180, the learned emotion recognition device is installed in the emotion recognition service system such as a robot installed offline, operates in real time, and the input / output data generated in the emotion recognition service process is stored in the database or local And stores it in an arbitrary storage location.

In addition, the stored input / output data is automatically re-learned in a rest period of the emotion recognition service.

The re-learning condition can be set by the administrator, and the robot automatically performs the re-learning, but when the service dissatisfaction of the user also increases, the administrator can stop using the service and arbitrarily re-learn.

Accordingly, it is possible to accurately determine the emotional state of the user by adjusting the center and width of each input membership function and the rule base of input and output through various kinds of input information for judging the user's emotional state through back propagation learning.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims. It can be understood that

In the course of the description of the embodiments of the present invention, the thicknesses of the lines and the sizes of the components shown in the drawings may be exaggerated for clarity and convenience of explanation, , Which may vary depending on the intentions or customs of the user, the operator, and the interpretation of such terms should be based on the contents throughout this specification.

100: fuzzy logic classification unit
110: eyebrow condition classification unit
120: mouth condition classification unit
130: head state classification section
140: eye state classification unit
200: emotional state analysis unit
300: adaptive neuro fuzzy inference unit
400:
500: database
600: Rule Base Setup Screen
610: Define input conditions
620: Conditional statement
700: re-learning department

Claims (14)

A fuzzy logic classification unit (100) for providing emotion state information of a user analyzed based on a facial expression of a user as a membership function for an input of an adaptive neuro fuzzy reasoning unit (300);
An emotion state analyzer 200 for providing emotion state information of a user analyzing a user's voice and an image with a preset emotion state program as a membership function for an input of the adaptive neurofuzzy inference unit 300; And
Calculates outputs of the input information provided by the fuzzy logic classifier (100) and the emotion state analyzer (200), adjusts the parameters so that the error between the calculated values and the target output is reduced, And an adaptive neurofuzzy inference unit (300) that performs a learning process so that a mean square (RMS) value is less than or equal to a preset reference value, and recognizes a user's emotional state using the learning result. The method according to claim 1,
The user's emotional recognition apparatus detects at least one of a user's atrial heartbeat, a user's touch and touch time, and a user's operation speed and outputs it to the adaptive neuro fuzzy reasoning unit 300 as a membership function for the user's emotional state And a sensor unit (400) for recognizing the user's senses. The method according to claim 1,
Wherein the facial expression of the fuzzy logic classifier (100) includes a position of an eyebrow, a shape of a mouth, a position and a direction of a head, and a position and direction of an eye. The method of claim 3,
Wherein the fuzzy logic classifier (100) extracts an emotional state by fuzzifying the facial expression according to a rule base of a preset input and output. 5. The method of claim 4,
Wherein the fuzzy logic classifier (100) performs defuzzification of the inferred emotion state through fuzzy logic, and outputs the result as a general numerical value. The method according to claim 1,
The fuzzy logic classification unit 100 classifies the user's emotional state into Ang, Fear, Sad, Nor, Sha, Sup, and Joy Wherein the user emotion recognition apparatus comprises: The method according to claim 1,
Wherein the adaptive neurofuzzy inference unit (300) performs a learning process of rearranging input and output rule bases by adjusting the center and width of each input membership function. 8. The method of claim 7,
The adaptive neurofuzzy inference unit 300 stores the center value, the width value, and the weight value of the forehead portion belonging to the forehead function when the learning ends, and uses the initial value as the initialization data in the process of restarting or re-learning And a user emotion recognition device. 8. The method of claim 7,
Wherein the adaptive neurofuzzy inference unit (300) performs a backpropagation learning process. a) defining and preparing input / output data pairs by the fuzzy logic classifier (100) and the emotion state analyzer (200);
b) initializing the belonging function by the adaptive neurofuzzy inference unit 300, receiving input and output data defined in the step a), and detecting a mean square root mean square error (RMSE) value;
c) the adaptive neurofuzzy inference unit 300 performs learning using a backpropagation learning algorithm;
d) When the learning ends, the adaptive neurofuzzy inference unit 300 adjusts the center and width of the forehead part belonging function and the linear coefficient value of the backward part belonging function, and if the mean square root error value is less than a preset reference value ; And
e) if the average root-mean-square error value is less than the reference value, storing the center value and the width of the adjusted forehead-portion belonging function and the linear coefficient value of the rear-part belonging function in the database 500 Recognition method. 11. The method of claim 10,
Wherein the ending of the learning in the step d) is ended when the generated mean square error value is less than a preset convergence condition value. 11. The method of claim 10,
Wherein if the average square root error value is equal to or greater than a reference value, the learning rate is adjusted to re-execute the steps c) and d). 11. The method of claim 10,
The center and the width of the forehead part belonging function stored in step e) and the linear coefficient values of the backward part belonging function are used as initialization data of the belonging function when the adaptive neurofuzzy inferring part 300 performs restart or re-learning Wherein the user emotion recognition method comprises the steps of: 11. The method of claim 10,
Wherein the step (e) further comprises the step of storing input / output data inputted in the emotion recognition service process after learning is completed in real time, and the stored input / output data further learning is performed in a pauser of the emotion recognition service Emotion recognition method.

Images