Cough is a symptom of many respiratory and nonrespiratorydisorders. Cough sound has been the subject ofseveral investigations. We simulated real-life data in this work,which included not just coughing but also speaking, eatingrelatedsound events such as chewing and swallowing, and thesubject's movement-related sound. Furthermore, cough datawas induced using the nebulizer, despite the fact that most priorinvestigations relied on pretended cough data. In a real-worldcontext, we aimed to show the usefulness of automatic coughdetection and tried to detect coughing in dysphagia patientstriggered by a mealtime mis-swallow. As a result of coughdetection in real environment, we obtained 0.82 in F-measureby using multi-layer LSTM and a throat microphone. In thecase of a dysphagia patient, we obtained 0.62 in the F-measureusing the same method.
展开▼