Speech Emotion Recognition Model with Quantified Intensity Scales
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Automatic Speech Emotion Recognition (ASER) systems face challenges in accurately quantifying and categorizing emotions due to the lack of standardized databases containing spontaneous and dynamic speech data, which affects model updating and performance, especially in differentiating emotional states and intensity levels.
Innovation Solution
A continuous self-updating and re-trainable ASER machine learning model that processes audio and text data from real-life conversations, using a hybridized classifier with Support Vector Machine (SVM) and locality-sensitive hashing (LSH), to quantify and recognize speech emotions by extracting features and assigning measurable scales, allowing for model updates with unlabeled data through deep learning algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing ASER systems use available databases of recorded human speeches for training, then the model can be built and trained, but the data quality is insufficient because the databases contain artificially performed and static speeches rather than spontaneous speeches sampled under natural and dynamic settings
Solution Approach 1:
The system performs self-updating by automatically collecting new speech data, extracting features, and retraining the model without manual intervention. The model serves itself by identifying its own training needs and autonomously acquiring new data from real-world sources, thereby continuously improving data quality while maintaining training reliability.
Solution Approach 2:
The system transitions from static training databases to dynamic, continuously updating data collections. Speech data is continuously sampled from natural settings, and the model adapts to new emotional patterns and speech variations over time, making the training process dynamic rather than static.
2Adaptability or versatility
If emotion recognition models are developed using existing databases, then the models can recognize emotions, but the updating of these models is seldom available due to lack of systematic data collection and model retraining mechanisms
Solution Approach 1:
The system implements a feedback loop where the model's performance is continuously evaluated, and new speech data is collected based on identified gaps or performance degradation. This feedback drives automatic retraining, enabling the model to adapt to new emotional patterns and maintain accuracy over time.
Solution Approach 2:
The model updating process operates continuously rather than periodically. Data collection, feature extraction, and retraining occur as ongoing processes, ensuring the model constantly improves and adapts without interruption, thereby enhancing both adaptability and automation.
3Measurement precision
If existing systems categorize emotions into different categories such as anger, sadness, joy, then the basic emotion recognition works, but there is no systematic and effective way to differentiate emotional states in the orders of magnitude (e.g., grades, levels, degrees, etc.) and scales of intensity or strength
Solution Approach 1:
The system adds a new dimension to emotion classification by introducing intensity scales and magnitude orders alongside traditional emotion categories. This transforms the classification from a single-dimensional category system to a multi-dimensional space that includes both emotion type and intensity level, enabling precise differentiation of emotional states.
Solution Approach 2:
The emotion classification is segmented into multiple hierarchical levels: broad emotion categories (anger, sadness, joy), intermediate intensity levels (low, medium, high), and fine-grained magnitude orders (1-10 scale). This segmentation allows systematic differentiation of emotional states without overwhelming complexity at each level.
4Reliability
If ASER models are trained with labeled data from existing databases, then the classification process can be performed, but the data quality in terms of data schema design, source and collection affects greatly the model build-up and the performance of the recognition process
Solution Approach 1:
The data collection system is designed to universally acquire speech data from multiple sources and settings (natural conversations, recorded speeches, dynamic interactions). This multi-functional collection mechanism ensures high-quality, diverse training data without requiring separate collection processes for each data source, thereby improving recognition performance while maintaining ease of data acquisition.
Data Source
AI summary
A method for training a speech-emotion recognition classifier under a continuously updatable and re-trainable ASER machine learning model. The quantified training data is generated by first processing the utterances of a human speech source and the associated texts in an emotion evaluation and rating process with normalization; then, extracting the features of the utterance; quantifying the feature attributes of the extracted features by labelling, tagging, and weighting the feature attributes, with their values assigned under measurable scales. The quantified training data comprises the normalized results of the emotion evaluation and rating process, the extracted features, the quantified emotional feature attributes, and the hash values of the quantified emotional feature attributes.