The Identification Method Of Cough Signals Based On Mel-Frequency Cepstrum Coefficient

Cough is a common symptoms of many respiratory disease, the changes of character ofcough for respiratory disease pathology mechanism research to provide important clues.Evaluate the strength and occurrence frequency can for the diagnosis and treatment of patientswith provide a lot of help. So far this assessment of cough, rely mainly on the subjectivemeasures, such as: cough reflex sensitivity, cough visual simulation test and quality of lifequestionnaire, cough symptoms description and patients diary, etc. By a human monitoringand deal with the slow and tedious, can be subjective factors and error, and the characteristicsof the patients to their own cough description may not complete, professional, studyautomatic identification system and its algorithm of cough is very necessary.Cough the identifying with voice similar research, mainly focused on the pretreatment,feature extraction and recognition phase, if use different endpoint detection methods(pretreatment stage), extraction of different types of cough frequency characteristics, orresearch the time domain and frequency domain features, then using a hidden markov modeland neural networks model building cough identification system.The existing HMM (a hidden markov model (HMM)) method in cough identificationfield application is the most extensive, but its algorithm complexity is high, make identify theefficiency is low, and the medical field too few samples of the training process of HMMadverse. In order to avoid these conditions occur, you need to characteristic parameters of thedimension compression, and find other process and deal with more simple method.This paper used in signal processing of spectrum diagram language, based on againspeech recognition is widely used Mel cepstrum parameters, through the analysis of the twofrequency domain parameters characteristics, obtained the language spectra energyparameters and Mel cepstrum parameter calculation process of logarithmic scale filter Melenergy parameters, of the two kinds of frequency domain parameters for the correspondingquantitative treatment, cough recognition. But Mel cepstrum parameter we as a HMMalgorithm of input signal, get the results, and this paper before the two kinds of parameters ofrecognition results were compared. In the process, in order to get more endpoint detection ofsamples and artificial marker of samples of of all kinds of methods of influence, we are thetwo cases of the experimental results and analysis.Finally in the ward environment to record the files for the experiment, the use ofartificial marker of samples of the experiment, language spectra energy parameters for thesensitivity of the results69.4%, specificity results for86.5%; Mel scale filter logarithm energy parameters to the sensitivity of the results of93.9%, specificity results for94.79%;HMM method to the sensitivity of the results96.2%, specificity results for96.9%. Use ofsamples endpoint detection test, language spectra energy parameters for the sensitivity of theresults78.8%, specificity results for53.5%; Mel scale filter logarithm energy parameters tothe sensitivity of the results of91.5%, specificity results for25.8%; HMM method to thesensitivity of the results85.7%, specificity results for87.5%.