Asthma Detection Research Based on Voice Signal Processing and Machine Learning
July 22nd, 2025
This study employed voice signal analysis and machine learning methods, utilizing MATLAB to extract distinctive voice features for non-invasive early detection of asthma. The Support Vector Machine (SVM) and Random Forest (RF) algorithms demonstrated comparable performance in terms of overall classification accuracy, although SVM may achieve a better balance between sensitivity and specificity.
- I'll discussed the most recent developments and the current experimental challenges in this field next. Current technologies use AI-driven voice analysis, motioning techniques like and SVM, signal processing tools such as MFCC, and variable acoustic sensors to detect disease-related patterns in sound signals. Current challenges in clinical translation of voice-based diagnostics includes scarce data, limited model generalization, privacy ethics, conflicts, and interoperability barriers.
[Instructor] To begin, open the command line tool and clone the GitHub repository to the local machine. Go to the official Python website to install Python on the machine. After installing Python, install the required Python libraries using requirements.txt file. Alternatively, if the text file is not available, install the necessary libraries manually using the given command. Then install PyCharm from its official website. Now, locate the main script files in the cloned repository. Use PyCharm or a text editor to open the Python files and review their contents and structure. To configure the dataset, verify that the data files required by the code are available and saved in the Excel format. Update the file paths in the scripts to match the actual file locations on the system. After identifying the main script, run the code using the appropriate command for that script. During execution, the code will load the data, train the machine learning models, and display results such as accuracy, confusion matrix, and the receiver operating characteristic, or ROC curve. This figure presents the confusion matrices for the support vector machine or SVM model and the random forest or RF model, showing their classification results for asthma and healthy control subjects. The SVM model accurately classified 14 asthma subjects and 12 healthy controls out of 30 samples in the confusion matrix. The RF model also correctly predicted 14 asthma cases and 12 healthy controls, showing comparable classification results. The SVM model achieved a higher area under the curve value of 0.95 in the ROC analysis, reflecting superior classification performance. Both SVM and RF models achieved an identical overall classification accuracy of 87% on the test dataset. In class-wise performance, the SVM model showed higher recall in the asthma group at 0.93, but lower recall in the healthy control group at 0.80, suggesting that while SVM is effective at identifying asthma, it may misclassify some healthy individuals. The RF model achieved the same recall values of 0.93 for asthma and 0.80 for healthy controls, indicating that both models have similar sensitivity and are equally capable of identifying positive cases.
View the full transcript and gain access to thousands of scientific videos
Overview
This study investigates the use of AI-driven voice analysis and machine learning techniques for the non-invasive early detection of asthma. By employing Support Vector Machine (SVM) and Random Forest (RF) algorithms, the research demonstrates comparable classification accuracy in identifying asthma cases.
Key Study Components
Area of Science
- Neuroscience
- Machine Learning
- Voice Signal Analysis
Background
- Current technologies utilize AI-driven voice analysis to detect disease-related patterns.
- Challenges in clinical translation include data scarcity and privacy ethics.
- Machine learning models like SVM and RF are employed for classification tasks.
- Voice-based diagnostics are gaining attention for their non-invasive nature.
Purpose of Study
- To explore the effectiveness of voice signal analysis in detecting asthma.
- To compare the performance of SVM and RF algorithms.
- To address challenges in implementing voice-based diagnostics in clinical settings.
Methods Used
- Voice signal analysis using MATLAB.
- Machine learning algorithms: Support Vector Machine (SVM) and Random Forest (RF).
- Evaluation of classification accuracy and performance metrics.
- Confusion matrix and ROC curve analysis for model comparison.
Main Results
- Both SVM and RF achieved an overall classification accuracy of 87%.
- SVM showed a higher area under the curve value of 0.95 in ROC analysis.
- SVM had higher recall in the asthma group (0.93) compared to healthy controls (0.80).
- RF demonstrated similar recall values for both groups, indicating comparable sensitivity.
Conclusions
- Voice analysis combined with machine learning can effectively detect asthma.
- SVM may provide a better balance between sensitivity and specificity.
- Further research is needed to address clinical translation challenges.
Frequently Asked Questions
Executive Industry Relevance
Voice signal processing combined with machine learning offers a scalable, non-invasive approach for early asthma detection, addressing a critical need for objective, reproducible diagnostics in respiratory disease research. By leveraging quantitative voice features and robust classification models, this workflow enhances predictive confidence at the target validation and assay development stages. The methodology supports portfolio-wide risk reduction by enabling standardized, data-driven decision-making in biomarker discovery and translational research.
Strategic Applications in Biopharma R&D
Early Discovery & Target Validation
- Enables interrogation of disease-relevant voice biomarkers for respiratory target validation.
- Supports mechanistic de-risking by quantifying differential phonetic features between patient and control groups.
- Facilitates predictive confidence in early-stage biomarker selection and triage.
Screening & Assay Development
- Prepares validated voice-based assays for downstream machine learning model integration.
- Standardizes feature extraction and dimensionality reduction for reproducible quantitative outputs.
- Enables scalable screening of candidate biomarkers using SVM and RF classification models.
Translational & Preclinical Research
- Aligns non-invasive voice biomarkers with translational endpoints for respiratory disease models.
- Supports continuity from discovery through preclinical validation by providing objective, quantitative readouts.
- Reduces biological risk in advancing candidate biomarkers to later-stage studies.
Pipeline & Workflow Integration
This workflow integrates from early discovery through lead identification and preclinical validation, leveraging voice signal analytics and machine learning for robust biomarker development.
- Discovery Biology: Quantitative voice feature analysis supports hypothesis testing and pathway clarification in respiratory disease.
- Screening: Machine learning models provide reproducible, quantitative classification outputs for candidate biomarker evaluation.
- Analytics: Outputs include accuracy, confusion matrices, and ROC curves to compare model performance and inform go/no-go decisions.
- Translational Research: Non-invasive voice biomarkers facilitate alignment with clinical endpoints and preclinical model systems.
- Enterprise Reuse: The standardized workflow and feature set enable reuse across respiratory and potentially other disease areas.
Operational & Enterprise Impact
- Scientific Value: Increases predictive confidence and reduces mechanistic ambiguity in respiratory biomarker discovery.
- Operational Value: Delivers standardized, reproducible, and scalable voice-based assays for machine learning integration.
- Strategic Value: Improves go/no-go decision quality and capital efficiency by providing objective, quantitative outputs.
- Portfolio Impact: Enables risk-adjusted prioritization and advancement of candidate biomarkers across the R&D pipeline.
Implementation Considerations
- Requires expertise in signal processing, machine learning, and respiratory disease biology.
- Needs access to high-quality voice recording infrastructure and computational analytics platforms.
- Demands cross-team standardization of data collection, feature extraction, and model evaluation protocols.
- Adaptation across diverse patient populations and disease models may require additional validation.
- Data scarcity and model generalization remain practical limitations for broader clinical translation.
