Cough detection : algorithm study

Abstract

Cough is one of the most common illnesses caused by various reasons such as environmental infections or allergies. Symptoms of cough may also be signs of chronic respiratory diseases that will impair one’s quality of life if not detected in its early stages. While chest X-rays and computed tomography (CT) scans can be conducted to detect respiratory diseases, the equipment used are bulky and expensive. As such, comprehensive studies on audio based cough detection algorithms have been pervasive in the recent years due to its effectiveness in diagnosing and treating cough patients. The aim of this project is to study cough detection algorithms and evaluate its performance via numerical experiments. The evaluation will be determined by the algorithm’s accuracy in discerning cough signals from normal breathing signals. Samples of cough and normal breathing audios were collected and subsequently used to carry out feature extraction, classification and validation. From the spectral analysis, results have shown a more distinct fluctuation and higher magnitude for audio samples collected from sick patients as compared to healthy individuals. Through the validation scheme, an overall accuracy of 90.0% was achieved. The audio samples were successfully classified into their respective classes with both the true positive and true negative accuracy rates to be 80.0% and above.