The Study And Implementation Of Cough Signals Recognition Based On DTW

Cough is a protective response mechanism of body. Cough can remove respiratory tractsecretions and foreign matters that invade into human respiratory tract. Characters of coughare determined by the underlying disease in human body, and perform in type, onset time andintensity and other factors. They will provide important clues in the research of respirationsyndrome.Many patients can not accurately describe their situation, which affects the patients’assessment. If patient’s cough can be real-time monitored, it will provide great help. Coughcan be recorded in real time by a portable recording device, but the analysis of audio contentsstill has to rely on doctors, and the workload is very large. If cough signals can be recognizedthrough computer-aided analysis, it will provide great help for doctors and patients.In order to achieve effective recognition of cough signals, we need a set of effectivecough signals recognition algorithm suitable for outdoor recordings cough recognition. Coughsignals recognition is similar to words recognition, and every cough signal can be regard asisolated rather than continuous fragment. When the outdoor recording samples are not too big,Dynamic Time Warping (DTW) is a more direct and quick recognition pattern. It can achievethe same effect with fewer calculations compared with the complex recognition algorithms.Therefore, we study cough recognition based on cough feature and DTW.This paper investigates present research situation of cough spotting at home and abroadthrough referring to literature. Analyze the characteristics of outdoor recording coughrecognition system and determine the overall framework of cough recognition system, thananalyze the physiological processes of cough as groundwork. Analyze cough signal in timedomain, and use commonly speech endpoint detection methods for cough detecting. Based onacoustic characteristics of cough and background noise, use the sort time zero crossing ratioand short time energy to identify the cough signals. In the feature extraction part, this paperintroduces the human auditory process and local frequency domain energy of cough, than triesto make corresponding improvement for Mel Frequency Cepstral Coefficient (MFCC), whichincludes of the selecting of differential feature parameters and the local feature extraction.Finally, this paper designs a cough signal recognition system based on the basic principles of Dynamic Time Warping, template search path algorithm and training algorithm, to achieve thedetection, extraction and recognition of outdoor cough recordings, and analyzes theexperiment results.