Applied Research For Cough Recognition

Cough is a common clinical symptom. With continuous development of science and technology and fast process of industrialization, people's living and working conditions have been significantly improved. Yet their state of health has been threatened to varying degrees as a consequence of environmental pollution brought by industrial waste water and waste gas, with increasing symptom of cough. The diagnosis of coughs with no clear pathogenesis and clinicopathological features depends on their descriptions and assessments that are greatly influenced by subjective factors, which causes difficulties in diagnosis and thus delay in treatment. Therefore it is important to develop an automatic collecting and analyzing system which is not restricted to specific patient usage but can distinguish and identify the most specific coughs recorded.In this paper, key technologies in cough sound recognition and characteristics of cough sounds were analyzed. Wavelet analysis method was used for cough sound recognition. The problems in its application in cough sound recognition were analyzed and solutions were put forward.The research work done is as follows:①Related information of cough identification systems both at home and abroad was reviewed, which included technologies, approaches and system performances. Key technologies concerning cough identification systems were discussed and analyzed and basic characteristics of cough were introduced as well.②Traditional endpoint detection algorithm was adjusted according to the characteristics of cough.The new method updates the threshold using the back-end adaptive update method , improves determine conditions of the starting and ending points and excludes part of speech by Calculating the mean of spectral entropy . It was proved by experiments that the present algorithm can promote the endpoint detection rate and improve its accuracy.③First the paper described two common features of LPCC and MFCC and their differential parameters and applied the combination of these characteristics to recognize the cough sound.Then the paper detailed a concrete realization to achieve of cough sound recognition using HMM models, extracting observation sequence as the input for HMM model by vector quantization of feature parameters using FCM. Finally the experiments proved the effectiveness in cough sound recognition using combination feature parameters of MFCC and the HMM model .④In the feature extraction, frst , the paper used the method of spectral analy cough sounds and other sound signals, and according to the results found optimal nodes by criterion of minimum entropy for the uneven distribution of signal energy. This method can not only improve the performance but also provide prerequisite for improving the processing speed and reducing the storage space .Secondly, the paper extracted the feature parametes based on the method of MFCC extraction and wavelet packet analysis. Proved by experiments, recognition rate using new feature extraction is higher than using the MFCC feature.