Multi-Modal Emotion Recognizer Using Segmentation and Dynamics

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing emotion recognition systems often output a single emotion value, which is insufficient for providing accurate and varied emotion-based services, and are influenced by initial weight settings, limiting their ability to differentiate emotions across multiple input data sources.

Innovation Solution

A robot system equipped with an emotion recognizer that uses a multi-modal approach, incorporating uni-modal and multi-modal recognizers, including text, speech, and image analysis, to recognize emotions through deep learning models, allowing for the integration of multiple data sources and providing a comprehensive emotion recognition result.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single emotion value is output from the analyzer module, then the system complexity is reduced, but the measurement precision and information completeness of emotion recognition deteriorates

Engineering Contradiction:
Improvesystem complexityVSAvoidemotion recognition precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the emotion recognition output into multiple independent emotion values (e.g., joy, sadness, anger, fear, surprise, disgust) instead of selecting a single emotion. Each emotion is recognized and output separately with its own confidence score, allowing the system to maintain low complexity while achieving high precision through parallel independent recognition channels.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent adds a new dimension to the output by including both emotion type and confidence score for multiple emotions simultaneously. This transforms the output from a single-dimensional emotion label to a multi-dimensional structure containing emotion category, intensity, and reliability metrics, thereby improving information completeness without significantly increasing system complexity.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If multiple data sources are integrated for emotion recognition, then the measurement precision improves, but the device complexity increases due to multiple recognizers and data integration mechanisms

Engineering Contradiction:
Improveemotion recognition accuracyVSAvoidsystem structure complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges multiple uni-modal recognizers (text, speech, image) into a unified multi-modal recognition framework. The recognizers process different data sources independently and their results are combined through a fusion mechanism that outputs integrated emotion values, achieving high precision through comprehensive data integration while managing complexity through modular architecture.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal emotion recognition system that handles multiple data sources (text, speech, image) through a common multi-modal framework. The same recognition architecture processes different input types, allowing the system to achieve high precision across diverse data sources without requiring separate specialized systems for each modality.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Ease of operation

If initial weight settings are used for each data source, then the ease of operation is improved, but the adaptability deteriorates as the system cannot differentiate emotions across various input data effectively

Engineering Contradiction:
Improvesystem configuration easeVSAvoidemotion differentiation capability
Core Design Contradiction:
Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic weight adjustment for different data sources based on the specific recognition context and input data quality. The system automatically adapts the influence of each modality (text, speech, image) according to the situation, allowing it to effectively differentiate emotions across various input data types while maintaining ease of operation through automatic adaptation rather than manual configuration.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of data source weighting dynamically based on the recognition task and input characteristics. By adjusting the weight parameters of different modalities according to the specific situation, the system achieves high adaptability in emotion differentiation while maintaining simple operation through automated parameter adjustment rather than fixed initial weights.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3623118B1Emotion recognizer, robot including the same, and server including the same
Publication Date: 2022.12.07 LG ELECTRONICS INC
  • EP3623118B1 patent drawingFigure 1~2
  • EP3623118B1 patent drawingFigure 3~4
  • EP3623118B1 patent drawingFigure 5

AI summary

An emotion recognizer according to an aspect of the present invention includes: an uni-modal preprocessor configured to include a plurality of recognizers for each modal learned to recognize emotion information of a user contained in uni-modal input data; and a multi-modal recognizer configured to merge output data of the plurality of recognizers for each modal, and be learned to recognize the emotion information of the user contained in the merged data, wherein the emotion recognizer outputs a complex emotion recognition result including an emotion recognition result of each of the plurality of recognizers for each modal and an emotion recognition result of the multi-modal recognizer.