IIIT-S CSSD: A Cough Speech Sounds Database

Abstract

Paralinguistic sounds such as laughter, cry and cough etc. communicate different messages as well, apart from the linguistic content in speech. These non-verbal speech sounds may possibly communicate some emotion, gesture or physiological condition of a human being. Cough sounds mostly indicate symptoms of a disease and are used sometimes as a gesture to draw attention. Analysing cough sounds using speech signal processing methods can be useful for assisting the medical experts in ailment diagnosis, and also for making machines more intelligent and human-like. This paper describes a database collected for cough speech sounds, named as IIIT-S CSSD. The database consists of three categories, namely, ailment cough, simulated cough and normal speech. Normal speech is recorded for each speaker, as a reference. Spectrograms are used as ground truth reference for preliminary analysis. This database can be helpful in analysing further the differences in cough sounds of different categories such as dry and wet cough, involuntary and voluntary cough, ailment and simulated cough, or age-wise differences in cough sounds etc. Few signal processing methods and initial analysis results indicating the differences in characteristics of different cough sounds are also discussed.