Speech Emotion Recognition Using Multi-Utterance Context Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech emotion recognition technologies inaccurately recognize emotions as they primarily analyze single sentences without considering the speech context, leading to inadequate emotion recognition in multi-round interactions.
Innovation Solution
A method utilizing two neural network models to determine emotional state information for current and previous utterances, incorporating statistical operations and context analysis to enhance emotion recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If speech emotion recognition is performed based on single sentence analysis, then the recognition process is simple and fast, but the recognition accuracy is low due to lack of context consideration
Solution Approach 1:
The patent segments the speech recognition task into two independent parts: (1) emotional state information extraction from individual speech frames using a first neural network model, and (2) emotional state determination from statistical results of multiple utterances using a second neural network model. This segmentation allows each model to focus on specific aspects, improving overall accuracy while maintaining manageable system complexity through modular architecture.
Solution Approach 2:
The patent transitions from single-sentence analysis to multi-utterance statistical analysis, adding the dimension of temporal context and historical speech patterns. By aggregating statistical results from n utterances (where n > 1) and feeding them into the second neural network model, the system captures emotional patterns across time, significantly improving recognition accuracy without overwhelming complexity.
2Measurement precision
If statistical operation is performed on multiple utterances to capture context, then emotion recognition accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The first neural network model performs preliminary extraction of emotional state information from individual speech frames before the second model processes the statistical results. This preliminary action prepares and pre-processes the data, reducing the computational burden on the second model and enabling more efficient multi-utterance analysis without excessive processing delays.
Solution Approach 2:
The patent extracts and separates the statistical computation from the emotional state determination. By calculating statistical results (mean, variance, etc.) of emotional state information from multiple utterances as a distinct step, and then feeding only these aggregated statistics into the second neural network model, the system reduces redundant computations and optimizes processing efficiency while maintaining context-aware accuracy.
Data Source
AI summary
A plurality of pieces of emotional state information corresponding to a plurality of speech frames in a current utterance are obtained based on a first neural network model; statistical operation is performed on the plurality of pieces of emotional state information, to obtain a statistical result, and then the emotional state information corresponding to the current utterance is obtained based on a second neural network device, the statistical result corresponding to the current utterance, and statistical results corresponding to a plurality of utterances before the current utterance.