A Wearable Breath Sound Monitoring Device Using the DWT-based MFCC Optimization Algorithm

Respiratory sounds are the most important feature in determining whether a person with asthma has an attack. The aim of this work is to find the most suitable recognition algorithm for identifying respiratory sounds in daily life environments and to port it to a stand-alone circuit system with a microcontroller as the controller. In this paper, the DWT-MFCC, i.e., the Mel frequency cepstral coefficients (MFCC) algorithm optimized using the discrete wavelet transform(DWT), is proposed as the feature extraction algorithm and the support vector machine (SVM) is used as the classifier. And all the feature combinations that can be obtained from the optimized algorithm, including the optimized MFCC and the optimized MFCC + statistical features, are compared with the feature combinations used in related studies, such as linear predictive cepstral coefficients (LPCC), statistical features, LPCC + statistical features, or the original MFCC. The classifier was cross-validated using MATLAB software. The combination of optimized MFCC + statistical features could achieve an accuracy of up to 96%, with an average accuracy of 92.56%. This is the highest accuracy and most consistent performance of all possible combinations. After porting the recognition algorithm to a stand-alone circuit with ESP32-WROOM-32 as the controller, the desired result was achieved.