Research On Feature Extraction Of Cough Sound And Its Application In Identification

At present, there are many biometrics in identification research areas, including fingerprints, palm prints, iris, retinal, vein, gait, ear, voice and hands, facial patterns, but some of these features easy to be forged and theft. In order to improve recognition accuracy and security, there are many scholars turn their attention to internal body information, such as DNA, heart sound signals.Cough is a common clinical symptom and has important diagnostic value. Because of the differences in everyone's articulator, sound track and pronunciation habit, the cough sound for each person is different. As the body's physiological characteristics of information, cough sound is universal, unique and stabile. Therefore, the application of cough sound in identification is theoretically feasible.This paper mainly studies the feature extraction of cough sound and its application in identification. At first, the paper introduces the research status of cough sound. Then the generating mechanism and properties of cough sound in time and frequency domain is analyzed. The paper mainly focuses on the method of feature extraction. By studying the MFCC parameter which is widely used in speech recognition, an improved feature parameter MFCC_N is proposed. In addition, a combined feature parameter is put forward, which composes MFCC_N and short-term energy. The improved parameter can effectively describe the characteristics of cough sound. In order to verify the effectiveness of the improved parameter, the experiments are carried out based on DTW and GMM respectively. The results show that the improved feature parameter has high recognition and low computational complexity. Also the experiments prove that the cough sound can be used as a biometric of identification and has great application prospect.