Aircraft Cockpit Voice Recognition Based On MFCC And Wavelet Packet Transform And Fuzzy SVM

There are many air disasters in the world every year. A necessary evidence in the analysis of air disaster is Black Box which includes Flight Data Recorder(FDR) and Cockpit Voice Recorder(CVR). CVR records some objective voices which reflect the condition of aircrafe and equipment,and some subjective information which reflects the perception and emotions of pilot,such as voices,aviation noise and background sound.CVR is an important evidence in the analysis of air disaster.It provides important evidence for the air disaster reconstruction.The voice signals in CVR are complex and non-stationary,and they have wide frequency range.This paper studies the classify of cockpit voice according to fourier transform,wavelet packet transform and fuzzy SVM. The major works are summarized as follows:First of all,on the basis of the"aircrafe cabin sound sample library"of the center of aviation safety technology CACC,this paper reduces the noise and intercept the cockpit voice with Adobe Auditio.The alarm sounds,switch,knob and other independent samples are successfully separated from the mixed signals.Secondly,fourier transform and wavelet packet transform are used for the independent cockpit voice, Mel Frequency Cepstrum Coefficient(MFCC) and Wavelet Packet Coefficient (WPC) are extracted as the initial characteristics.The finally characteristics are determined by geometric distance classifiability criterion.Then,the support vector machine (SVM)algorithm is sensitive to outliers and noise present in the datasets and when it comes to imbalanced samples,SVM produces suboptimal classification models. Fuzzy SVM(FSVM) is a variant of the SVM algorithm,which has been proposed to handle the problem of outliers and noise.However,like the normal SVM algorithm,FSVM can also suffer from the problem of imbalanced samples.In this paper,we present a method to improve FSVM for imbalanced samples learning,which can be used to handle the imbalanced samples problem in the presence of outliers and noise.Training samples are assigned two different fuzzy-membership values,and these membership values are incorporated into the SVM learning algorithm. Based on the experiment results,it can be concluded that the proposed method is a very effective method.Lastly, a software to classify the cockpit voice with MATLAB and VC++ is designed.The software fully plays the advantage of MATLAB and VC++ which can classify the cockpit voice intuitively,quickly,accurately. The study of this thesis will have great signigicance in judging the contents in CVR background voice and determining the cause of the air disasters.