Research On Key Technologies Of Speech Recognition In Airborne Noise Environment

Speech recognition technology is a hot research topic in the past 30 years,and it is also a very comprehensive subject with great application prospects.Although a lot of voice products have emerged in today's society,there are still various problems in the application process,especially in a specific context,so this thesis focuses on the speech recognition technology in the airborne noise environment.The implementation process and system architecture of speech recognition are introduced in detail,and the basic principles of each component of speech recognition system are described,including preprocessing,endpoint detection,feature extraction and identification methods.A detection method combining spectral subtraction and short-time zero entropy is proposed.The method uses spectral subtraction method to reduce noise in the front part of the detection,and it combines the short-time zero-crossing rate and power spectral entropy to construct a new speech parameter,which is named short-term zero entropy.It's proved that the proposed algorithm can get good detect effect of endpoint in the airborne noise environment by experiments.An improved MFCC feature based on HHT transform is proposed.The method replaces the FFT changes with HHT transform for time-frequency analysis.An improved MFCC parameters is obtained by combining HHT transform and the Teager energy,which is used in the high frequency part of the MFCC feature.The experimental results show that the parameter has good robustness and stability in the airborne noise environment.A speech recognition system based on the theory of Hidden Markov Model is built on the MATLAB software platform to identify aircraft voice commands,and the operation interface and simulation results of each part of the identification process are displayed.The final experimental results show that the improved algorithm has good application value in the airborne noise environment.