Cough detection using a non-contact microphone: A nocturnal cough study

Abstract

An automatic non-contact cough detector designed especially for night audio recordings that can distinguish coughs from snores and other sounds is presented. Two different classifiers were implemented and tested: a Gaussian Mixture Model (GMM) and a Deep Neural Network (DNN). The detected coughs were analyzed and compared in different sleep stages and in terms of severity of Obstructive Sleep Apnea (OSA), along with age, Body Mass Index (BMI), and gender. The database was composed of nocturnal audio signals from 89 subjects recorded during a polysomnography study. The DNN-based system outperformed the GMM-based system, at 99.8% accuracy, with a sensitivity and specificity of 86.1% and 99.9%, respectively (Positive Predictive Value (PPV) of 78.4%). Cough events were significantly more frequent during wakefulness than in the sleep stages (p < 0.0001) and were significantly less frequent during deep sleep than in other sleep stages (p < 0.0001). A positive correlation was found between BMI and the number of nocturnal coughs (R = 0.232, p < 0.05), and between the number of nocturnal coughs and OSA severity in men (R = 0.278, p < 0.05). This non-contact cough detection system may thus be implemented to track the progression of respiratory illnesses and test reactions to different medications even at night when a contact sensor is uncomfortable or infeasible.

Leveraging AI and Machine Learning to Develop and Evaluate a Contextualized User-Friendly Cough Audio Classifier for Detecting Respiratory Diseases: Protocol for a Diagnostic Study in Rural Tanzania

“…The recording room will not be fitted with additional soundproof materials to mimic the clinical environment where the diagnostic procedures for patients with TB, asthma, and COPD are performed. A comprehensive summary of the audio sound capturing process is detailed in Multimedia Appendix 1 [ 11 , 12 , 31 - 36 ].…”

Identification of Respiratory Diseases using Peak Energy Analysis of Acoustic Cough

“…Exploring the acoustic properties of voluntary cough sounds have revealed notable distinctions among coughs associated with various diseases (24). Analyzing the waveforms which capture time and frequency content, can help identify characteristics in cough sounds that are indicative of the presence of mucus in the airways (25,26).…”

“…Apart from the identification of pathological cough, the peak energy analysis along with the other time domain and frequency domain features was used to clearly distinguish between coughs and other noises in a sleep environment (25).…”

“…The importance of peak energy analysis is well articulated in the study, which implemented it as a part of mixed-domain feature vector for the quick identification or early screening of COVID-19 (24). Apart from the identification of pathological cough, the peak energy analysis along with the other time domain and frequency domain features was used to clearly distinguish between coughs and other noises in a sleep environment (25).…”

Visualization and Quantification of Facemask Leakage Flows and Interpersonal Transmission with Varying Face Coverings

“…The team was able to identify sites of leakage in the masks and determine how far the 'cough' traveled after passing through the masks [30]. For normal breathing, the foot pump delivers 500 mL of air to the model per breath, while for coughing, the foot pump delivers 1500 mL, mimicking the average amount of air expelled during a cough [31].…”