Voice Biomarker Screening for Blood Glucose Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voice signal analysis systems face challenges in efficiently determining blood glucose levels in healthy individuals and as a potential biomarker for diabetes, requiring advanced methods to process voice data in real-time and reduce processing overhead.
Innovation Solution
A predictive model using a multi-class random forest classifier is generated with selected voice biomarkers that demonstrate significant differences, intra-stability, and decision-making ability, allowing for efficient discrimination between different blood glucose levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If voice signal analysis is performed to determine blood glucose levels, then non-invasive monitoring capability is improved, but processing overhead and computational complexity increase
Solution Approach 1:
The patent extracts and isolates specific voice biomarkers (pitch, loudness, spectral features) from the complex voice signal that are most strongly correlated with blood glucose levels. By focusing only on these relevant features rather than analyzing the entire voice signal spectrum, the system achieves non-invasive glucose monitoring while significantly reducing computational overhead and processing requirements
Solution Approach 2:
The voice signal analysis is segmented into distinct processing stages: voice signal acquisition, biomarker extraction (pitch, loudness, spectral features), and classification/prediction. This segmentation allows each stage to be optimized independently, reducing overall system complexity while maintaining the non-invasive monitoring capability
2Measurement precision
If comprehensive voice biomarker analysis is performed to improve prediction accuracy, then measurement precision is improved, but processing time increases
Solution Approach 1:
The system performs preliminary action by pre-identifying and extracting only the most discriminative voice biomarkers (pitch, loudness, specific spectral features) that have been shown to correlate with blood glucose levels. This preliminary feature selection occurs before the main classification process, ensuring high prediction accuracy while minimizing processing time by avoiding analysis of irrelevant voice signal components
Solution Approach 2:
The patent applies partial action by selecting a specific subset of voice biomarkers that are most strongly correlated with blood glucose levels rather than analyzing all possible voice features. This selective approach achieves sufficient prediction accuracy (as demonstrated by the model's performance) while significantly reducing processing time compared to comprehensive analysis of all voice signal characteristics
3Reliability
If multiple voice biomarkers are analyzed to improve model reliability, then prediction reliability is improved, but device complexity increases
Solution Approach 1:
The patent introduces voice biomarkers as intermediary features that mediate between the raw voice signal and the blood glucose level prediction. These biomarkers (pitch, loudness, spectral features) serve as simplified representations that capture the relevant physiological information while reducing the complexity of the relationship between voice signals and glucose levels, thereby improving model reliability without proportionally increasing device complexity
Data Source
AI summary
Provided are methods and systems for generating a model for determining blood glucose levels using voice samples and associated embodiments. Different criteria for selecting voice features as biomarkers for determining blood glucose levels were investigated. Models generated using voice features selected using multiple criteria including a univariate measure, a measure of intra-stability of the voice feature and a measure of the decision-making ability of the voice feature were found to discriminate between subjects with different blood glucose levels and perform better than models generated using fewer criteria. The described embodiments can be used to generate models useful for integration into various applications for predicting blood glucose levels using voice.