Recognition and Extraction of Cough Sound from Audio Signals

Cough contains a wealth of individual information. Different respiratory diseases potentially in the human body often reflect different cough signal characteristics. The collection and recognition of cough sounds is of great significance for early screening and examination of respiratory diseases. However, in reality, most cough sounds are contained in a long audio segment. These audios also contain silent segments and some noise segments. Therefore, before the subsequent use of cough sounds, the cough sounds in the audio file need to be extracted. This paper introduces the use of a method of identifying and extracting cough sounds. The method is divided into two processes: endpoint detection and cough sound identification. Endpoint detection distinguishes voiced segments from silent segments. In the process of cough sound recognition, the three features of LPC coefficient, spectral centroid and spectral flatness of the sound segment are extracted, and the support vector machine is used as the classification model to divide the audio segment into cough sounds and non-cough sounds. The algorithm achieves a recognition accuracy of 91.4%, which provides a basis for further research on subsequent cough sounds.