Cough is an important and present symptom in many respiratory diseases affecting the airways and lungs. Therefore it is interesting to monitor cough in a continuous, on-line way. The objective of this study was to test a cough recognition algorithm in real pig houses. Cough sounds were registered on 150 days old, 60 kg heavy Landrace x Large White x Duroc crosses with a microphone placed at 20-50cm from the animal. The analysis was done on a feature vector, containing energy, timederivate energy and mean power spectral density. The feature vector was compared to the reference set using dynamic time warping. This resulted in a correct classification of 90%.
Index Terms. Sound analysis, Cough, Diagnostic system, Health management