Cough Detection via PCA Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cough detection methods, especially audio-based systems, face challenges in accuracy and privacy protection, as they often require human annotation and expose personal information, while thoracic pressure-based systems are expensive and cumbersome.
Innovation Solution
An audio-based cough detection system that uses a cough model to convert audio signals into a frequency-based matrix, transforms it into a lesser dimensional representation using basis vectors, and classifies coughs using principal component analysis and classifiers like random forest, while preserving privacy by preventing the reconstruction of speech sounds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If audio-based cough detection systems are used, then cough detection accuracy is improved, but privacy protection deteriorates as personal information is exposed
Solution Approach 1:
The patent extracts only the essential cough characteristics from the audio signal through feature extraction, transforming the complete audio into a reduced feature set that captures cough properties while excluding personal identifying information. This extraction process separates the diagnostic information (cough features) from the privacy-sensitive information (speech and personal characteristics).
Solution Approach 2:
The system transforms the audio signal from its original parameter space into a different parameter space through mathematical transformations (Fourier transform, Mel-frequency cepstral coefficients). This parameter transformation maintains the essential cough detection capabilities while fundamentally altering the representation to prevent reconstruction of original speech and personal information.
2Measurement precision
If thoracic pressure based systems are used for cough detection, then measurement accuracy is improved, but device complexity and cost worsen
Solution Approach 1:
The patent replaces the mechanical/thoracic pressure measurement system with an acoustic field-based system. Instead of using pressure sensors and specialized equipment to measure thoracic pressure changes, the system uses acoustic sensors (microphones) to capture cough sounds and processes them through signal processing algorithms, thereby substituting a complex mechanical measurement system with a simpler acoustic-based system.
Solution Approach 2:
The system creates a computational model of cough characteristics that replicates the diagnostic information obtainable from thoracic pressure measurements. Through feature extraction and pattern recognition, the system generates a digital representation of cough properties that serves as a substitute for direct pressure measurement, achieving similar diagnostic accuracy without the complexity of pressure sensing equipment.
3Ease of manufacture
If conventional audio-based detection systems are used, then implementation simplicity is improved, but automation level worsens due to requirement of human annotators
Solution Approach 1:
The system implements self-service automation through unsupervised machine learning algorithms that automatically learn cough patterns from training data without requiring human annotators to manually label each audio segment. The algorithm autonomously performs feature extraction, pattern recognition, and classification, enabling the system to automatically distinguish coughs from other sounds without human intervention in the detection process.
Solution Approach 2:
The system incorporates feedback mechanisms where the classification results are continuously refined through comparison with training data and performance metrics. The algorithm learns from its own performance and adjusts its parameters to improve accuracy over time, creating an automated feedback loop that eliminates the need for continuous human annotation while maintaining or improving detection accuracy.
Data Source
AI summary
Examples of the present invention utilize principal component analysis (PCA) to detect cough sounds in an audio stream. Comparison of all or portions of the audio stream with a cough model may be conducted. The cough model may include a number of basis vectors may be based on initial portions of known coughs. The initial portions may be non-user specific, and accordingly the cough model may be used to detect coughs across individuals. Moreover, examples of the present invention may reconstruct the cough sounds from stored features such that the cough sounds are reconstructed but the reconstruction techniques used may be insufficient to reconstruct speech sounds that may also have been recorded, which may increase user privacy.