Differential Speech Emotion Recognition for User-Aligned Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing emotion recognition systems rely on subjective rules or parameters set by developers or labelers, leading to inaccuracies in emotion recognition due to speaker and environmental characteristics, and fail to accurately reflect actual user impressions.
Innovation Solution
An emotion recognition system that infers differential emotions between two pieces of speech data from the same speaker, using a differential-emotion recognition model to quantify relative emotions, reducing the influence of speaker and environmental variables.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a rule-based method or machine learning method is used to recognize emotions from single speech data, then emotion recognition can be performed, but the recognition accuracy deviates from actual user impression due to subjectivity of developers or labelers
Solution Approach 1:
Instead of recognizing absolute emotions from single speech data (conventional approach), the patent inverts the approach by recognizing differential emotions between two speech data pieces. The differential-emotion recognition model calculates the difference between emotions in first speech data and second speech data, thereby eliminating subjective biases from absolute emotion labeling by developers or labelers and achieving results that better align with actual user impressions.
2Productivity
If absolute emotion recognition is performed from single speech data, then emotion categories can be output, but speaker and environmental characteristics cause inaccuracies
Solution Approach 1:
The patent extracts and eliminates the harmful factors (speaker characteristics and environmental characteristics) by computing the differential emotion between two speech data pieces from the same speaker. By taking the difference, these confounding factors cancel out, leaving only the genuine emotion changes, thereby improving recognition accuracy while maintaining productivity.
3Measurement precision
If differential emotion recognition between two speech data pieces is performed, then accuracy aligning with user impression is achieved, but system complexity increases
Solution Approach 1:
The patent changes the parameter being recognized from absolute emotion values to differential emotion values. By transforming the recognition task from estimating absolute emotion states to calculating emotion differences between two time points, the system achieves higher accuracy aligned with user impressions. The differential-emotion recognition model processes pairs of speech data and outputs emotion transitions, managing complexity through focused differential analysis rather than absolute state estimation.
Data Source
AI summary
An emotion recognition system includes an input unit configured to input first speech data and second speech data, and a processing unit configured to input the first speech data and the second speech data to a differential-emotion recognition model that infers a differential emotion between two pieces of speech data, and acquire, from the differential-emotion recognition model, differential-emotion information indicating a differential emotion between the first speech data and the second speech data.