Robust Detection of Audio-Cough Events Using Local Hu Moments

Abstract

Telehealth has shown potential to improve access to health-care cost-effectively in respiratory illness. However, it has failed to live up to expectation, in part because of poor objective measures of symptoms such as cough events, which could lead to early diagnosis or prevention. Considering the burden that these conditions constitute for national health systems, an effort is needed to foster telehealth potential by developing low cost technology for efficient monitoring and analysis of cough events. This paper, proposes the use of local Hu moments as a robust feature set for automatic cough detection in smartphone-acquired audio signals. The final system feeds a k-Nearest Neighbors classifier with the extracted features. To properly evaluate the system in a diversity of noisy backgrounds, we contaminated real cough audio data with a variety of sounds including noise from both indoor and outdoor environments, and non-cough events (sneeze, laugh, speech, etc.). The created database allows flexible settings of Signal to Noise Ratio (SNR) levels between background sounds and events (cough and non-cough). This evaluation was complemented using real patient data from an outpatient clinic. The system is able to detect cough events with high sensitivity (up to 88.51%) and specificity (up to 99.77%) in a variety of noisy environments, overcoming other state-of-the-art audio features. Our proposal paves the way for ubiquitous cough monitoring with minimal disruption in daily activities.

Wearable Sensing and Telehealth Technology with Potential Applications in the Coronavirus Pandemic

“…It demonstrated an ability to distinguish the COVID-19 cough from non-COVID-19-related cough with over 90% accuracy [149]. With smartphone acquired audio signals, Monge-Álvarez et al have used local Hu moments as a robust feature set with a k-nearest-neighbor classifier for automatic cough detection, and demonstrated the sensitivity and specificity of cough detection as high as 88% and 99% in various environments [150].…”

A Machine Hearing System for Robust Cough Detection Based on a High-Level Representation of Band-Specific Audio Features

“…The experimental setup used in Section IV-B was used to compare our proposal with three recently proposed cough detectors: 1) the one proposed in [15], based on ensembling multiple frequency subband features; 2) our proposal in [18], based on moment theory cepstrogram characterisation; 3) and the CNN architecture employed by Amoh and Odame in [17]. The obtained results are presented in Table V.…”

“…The most meaningful features are then selected and combined in a high-level representation to perform robust cough detection in noisy conditions. Results on real patient data show that the proposed approach overcomes the best performing of recently proposed robust cough detectors [15], [17], [18].…”

“…Even though the audio signal was recorded during real-life activity, the quality of the acquired signal was favoured by the proximity of the lapel microphone to the mouth. Our work in [18] achieved robust segmentation of audio cough events using short term processing. The proposal applied moment theory to characterise adjacent frames and frequency bands of the cepstrogram audio signal representation.…”

“…The short-time approach employed in [10], [14]- [18] gives a simplistic and time-affordable analysis, enabling the classification of signal frames as belonging to a cough event or not. However, when a high-level representation of the data is used, the actual segmentation of cough events can improve while keeping the classification scheme simple.…”