Heart Sound Identification Based On MFCC And Short-time Energy

With the rapid development of science technology and communication network, people’s demond for leidentity recognition technology becoming more and more higher. Stability, practicality, security becomes more and more important for people, some traditional identification methods cannot meet the needs of human beings, so more and more people to look at the new field of biometrics. Biometrics refers to using the computer, collection the biological characteristics of human body parts, and use it to the process of identification. It both has easy collection, high stability, uniqueness and some other characteristics, so it is very ideal used for identification. But there is no denying that Biometric recognition in practical application is still facing many challenges, one of the most representative problem is the security of identify. So it is very necessary.to seek a kind of high safety performance of biometric identification methods.This paper use the heart sound signal to realize the identification algorithm, we proposed the Mel frequency cepstrum coefficient with the combination of short-time energy GMM recognition algorithm. This method on the basis of the original MFCC parameters and increase the heart sound signals of short-time energy information,with the combination of gaussian mixture model identification algorithm. We collectthe heart sounds data samples and built a heart sound daterbase, and esearch specific estimation algorithm of the model,we also analyzed the basic principle of gaussian mixture model.At last, we Build the Heart sound identification system,achieved the identification of heart sound signal characteristic parameters. We found that the algorithm has higher recognition rate and the small computational complexity by experiment.This paper mainly completed the following work:Introduce the commonly used biological recognition technology, at the same time, this paper expounds the theoretical basis and the time-frequency characteristics of heart sound signals.Analyzed the possible sources of noise, according to difference of heart sound signal in noise, and the comparison of several denoising methods, proposed wavelet threshold denoising algorithm for heart sound signal preprocessing. Introduces several kinds of common wavelet functions and briefly analyzed its characteristics. Through comparing the different wavelet functions, we found a wavelet basis function for the selection of a reasonable and effective to complete the denoising.At the testing stage,we use heart sound library to test the algorithm, comparing the influence of different parameters for the system identification performance, research the different parameters setting of the algorithm,got the optimal parameter combination, made the system received the best recognition performance.