Voice Emotion Estimation Using Dual Models for Linguistic Content
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing emotion estimation technologies, such as those described in Japanese Unexamined Patent Application Publication No. 2019-28910, do not adequately utilize machine learning to estimate customer emotions during business talks, limiting their effectiveness in improving customer service and support quality.
Innovation Solution
An emotion estimation method that employs two distinct models: a first model that estimates emotions based on linguistic information and a second model that estimates emotions without relying on linguistic information, switching between them based on the presence of linguistic content in voice data, utilizing machine learning algorithms like decision trees, CNN, RNN, or deep learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single emotion estimation model is used for all voice data, then the device complexity is reduced, but the emotion estimation accuracy deteriorates because the model cannot adapt to different types of voice content
Solution Approach 1:
The patent divides the emotion estimation system into multiple specialized models: a first model for voice data with linguistic information and a second model for voice data without linguistic information. Each model is optimized for specific types of input, improving estimation accuracy without requiring an overly complex universal model.
Solution Approach 2:
The system dynamically selects which estimation model to use based on whether the voice data contains linguistic information. This dynamic adaptation allows the system to optimize performance for different input types while maintaining manageable complexity through conditional model selection.
2Measurement precision
If linguistic information is always used for emotion estimation, then the measurement precision improves, but the reliability deteriorates when verbal emotions do not reflect actual emotions
Solution Approach 1:
The patent introduces a determination unit that acts as an intermediary to assess whether voice data contains linguistic information before selecting the estimation model. This intermediary layer ensures that linguistic information is used appropriately only when it reliably indicates actual emotion, improving overall system reliability.
Solution Approach 2:
Instead of always using linguistic information for emotion estimation, the patent inverts the approach by using the second model (without linguistic information) when linguistic content may not reflect true emotions. This inversion allows the system to capture actual emotions even when verbal expressions are misleading.
3Productivity
If the system processes all voice data through the first estimation model, then the productivity is maintained, but the measurement precision deteriorates for voice data without linguistic information
Solution Approach 1:
The system dynamically routes voice data to the appropriate estimation model based on the presence of linguistic information. This dynamic processing maintains high productivity by avoiding unnecessary processing steps while improving measurement precision by selecting the most suitable model for each data type.
Solution Approach 2:
The patent changes the processing parameters (which model to use) based on the characteristics of the input voice data. When linguistic information is present, the first model is used; when absent, the second model is used. This parameter adaptation maintains efficiency while optimizing accuracy for different data types.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An emotion estimation method that is executed by an information processing device (10) includes: acquiring voice data; determining whether the voice data includes linguistic information; and estimating an emotion corresponding to the voice data by inputting the voice data to a first estimation model, when the voice data includes the linguistic information, or estimating the emotion corresponding to the voice data by inputting the voice data to a second estimation model, when the voice data does not include the linguistic information. The first estimation model estimates the emotion based on the linguistic information. The second estimation model estimates the emotion without being based on the linguistic information.