Hybrid ML Model for Cough Detection Using Time-Domain and Spectrogram Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting and characterizing body actions, particularly cough events, face challenges in reliability and accuracy, necessitating an improved approach for effective identification and characterization.
Innovation Solution
A computer-implemented method involving a trained machine learning model that processes time-domain and spectrogram representations of signals from various modalities to generate joint features, determining whether and to what extent an event is related to a body action, such as a cough, by inputting time-domain and spectrogram features into separate inputs of the model and combining them to produce event information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a machine learning model uses only time-domain representation of signals, then the model complexity is lower, but the detection accuracy and reliability of body actions deteriorates
Solution Approach 1:
The patent combines multiple signal representations (time-domain and spectrogram) as parallel inputs to the machine learning model. This merging of different feature spaces allows the model to leverage complementary information from both temporal and frequency domains, thereby improving detection accuracy while maintaining a manageable model architecture through structured input processing.
Solution Approach 2:
The patent transforms the signal analysis from a single time-domain dimension to multiple dimensions by introducing spectrogram representation. This adds a frequency-time dimension to the analysis, enabling the model to capture both temporal patterns and spectral characteristics simultaneously, thus enhancing detection precision without excessive complexity increase.
2Reliability
If a machine learning model uses only spectrogram representation of signals, then the frequency analysis capability is improved, but the temporal resolution and overall detection reliability deteriorates
Solution Approach 1:
The patent merges time-domain representation and spectrogram representation as dual inputs to the machine learning model. This combination ensures that both temporal information (from time-domain) and frequency information (from spectrogram) are preserved and processed simultaneously, preventing information loss and improving overall detection reliability through complementary feature integration.
3Measurement precision
If multiple signal modalities are processed separately, then the processing simplicity is maintained, but the characterization accuracy of body actions deteriorates
Solution Approach 1:
The patent processes multiple signal modalities (time-domain and spectrogram) through a unified machine learning model architecture that accepts both as parallel inputs. This merging approach enables the model to learn joint representations that capture interactions between different modalities, thereby improving characterization accuracy while managing processing complexity through integrated model design.
Data Source
AI summary
Systems, methods, and computer programs disclosed herein relate to training and using a machine learning model to detect, identify and/or characterize body actions, in particular cough events.