Research On Pilot Voice Commentary System

Under the new situation,the aviation industry is developing rapidly,and the demand of pilots is constantly expanding.In order to meet the needs,a variety of airline companies and institution have the planning courses for training pilots.As the number of aircraft and the surge in routes,flight safety is particularly important,especially the accuracy of the communication between pilot and control in flight has become the most important issue.In recent years,the research enthusiasm in the field of speech recognition has been on the rise.It is possible to apply speech recognition technology to the realization of speech assessment system.In this thesis,the speech recognition technology is introduced elaborately.Based on the human body sounding mechanism,the preprocessing of speech signal is analyzed with the comparison of the detection results of three different endpoints made in detail.At the same time,it mainly studies the linear prediction cepstrum coefficient(LPCC),the peak envelope parameters and the mel-frequency cepstrum coefficients(MFCC).Furthermore,it makes a differential improvement in MFCC and proposes the parameter extraction method of the first order difference MFCC.At the same time,it introduces the structure and algorithm derivation of BP neural network and analyzes the defects in the network training process.It also proposes the improvement of damped least squares algorithm(LM)and particle swarm optimization(PSO)to solve the gradient descent training methods easy to fall into the local minimum point disadvantage of training failure to greatly increases the network stability and system recognition rate.Through the analysis of the main terminology of“Standard of Air Traffic Radiotelephone” formulated by Civil Aviation Administration,this thesis summarizes the characteristics of grammatical structure and special pronunciation,discusses with professional pilots and chooses eight representative instructions as outline.It makes good use of GUI technology of MATLAB to set up a pilot voice comment system to identify the voice data input by the three simulated pilots and compared with the outline by listening to pronunciation to judge the three simulated pilots whether grasp the outline or not according to the identify results.