The design and development of a system for automatic sound analysis

Barry, Samantha Jane

Abstract

Cough monitoring has been undertaken for many years but the subsequent analysis of cough frequency and the temporal relation to trigger factors have proven problematic. Since cough is episodic, data collection over many hours is required; real-time aural analysis of the resultant data is therefore highly time-consuming. As cough as a symptom becomes a more important area of study, the requirement for a method for cough monitoring and assessment increases.
An objective, portable device has been developed for the 24 hour acoustic monitoring of cough together with a system for the automatic recognition and counting of coughs in audio recordings.
Digital signal processing was applied to extract the spectral features of sound events, followed by a variety of investigated data pre-treatment techniques.
Due to their success in other areas, artificial neural networks were studied in depth for their potential application as a pattern classification step; they were, however, found to be unsuitable for the discriminations between cough and non-cough events due to the high degree of variability present in both classes.
Digital filtering was then applied to remove a range of low frequencies common to many sounds in order to study the characteristics of the higher frequencies.
The system has been demonstrated to correctly classify 82% of all coughs present in audio recordings; the 18% of coughs that were missed were mostly very low in amplitude and would not have been classified as coughs by an experienced cough listener. This figure of 82% is therefore largely underestimated. Only 11% of the events classified coughs were actually non-coughs. Results for sensitivity are calculated to be 82%, specificity is 97% and accuracy is 94%. These are considered to be highly acceptable for clinical applications.
Ambulatory high-quality cough recordings are now possible over a 24 hour period. Automatic analysis affords rapid and reproducible association of cough with environmental triggers.

Citation

Barry, S. J. The design and development of a system for automatic sound analysis. (Thesis). University of Hull. https://hull-repository.worktribe.com/output/4224621

• Thesis Type: Thesis
• Deposit Date: Jan 10, 2023
• Publicly Available Date: Feb 24, 2023
• Keywords: Chemistry
• Public URL: https://hull-repository.worktribe.com/output/4224621
• Additional Information: Department of Chemistry, The University of Hull
• Award Date: Dec 1, 2006