Development Of Shipborne Voice Recognition And Analysis System

The development of the marine economy has brought abundant material wealth to people and improved quality of life.However,marine operations also lead to frequent marine accidents,threatening people’s economic and personal safety.Therefore,obtaining and identifying maritime distress signals in time becomes an important research direction.The conventional method is to search and identify through infrared and visible light imaging technology.However,this method has a strong dependence on illumination conditions,especially under severe weather conditions with low visibility at sea,which will greatly reduce the accuracy of recognition.To solve this problem,Nevertheless,this study from the perspective of acoustics,by analyzing the characteristics of various typical sound signals in the complex marine environment,proposes a method for reducing noise of weak signals,and then develops a shipboard sound recognition analysis system.Due to the influence of the complicated and changeable climate on the sea,the target sound signal is often submerged in noise.In order to recognize the target sound signal in time and effectively,this study proposes an adaptive noise cancellation method based on whole annealing genetic algorithm to improve the signal-to-noise ratio of the weak target sound signal.It provides a new method for the noise reduction of weak signals.For another thing,in order to further reduce the interference of noise on the subsequent target sound signal recognition,this study adopts a dual-threshold endpoint detection method,which is according to the threshold of both the short-term energy and the short-term zero-crossing rate,and then the target sound signal is cropped and the effective data is extracted.Traditional recognition method only relies on a single feature parameter,resulting in low recognition accuracy.In this study,the robust and easy-to-extract static feature MFCC and dynamic feature first-order difference MFCC are used as feature parameters to effectively improve the accuracy of recognition.Due to the difference and uniqueness of the sound,the characteristic vector length of the target sound signal is different.Aiming at the shortcomings of traditional DTW models that require global matching,based on the core idea of segmented matching and successive elimination,this study proposes an improved DTW model to identify target sound signals.The recognition time is further shortened,and the accuracy and efficiency of recognition are improved.On the basis of theoretical research,according to the system performance indicators set in this paper,the software and hardware design of the system platform is completed.First,the sound signal is picked up by an electret condenser microphone and converted into an analog electric signal.Then it is conditioned by program-controlled amplification and low-pass filtering.Finally,the A/D conversion module is converted into a digital quantity that can be processed by the main control unit.Through a series of operations on the data by the main control unit,such as adaptive noise cancellation of whole annealing genetic,endpoint detection,feature parameter extraction,and improved DTW model matching,the target sound signal is identified and the result is fed back to the user through the host computer.Through simulation verification and actual testing,it is proved that the system can accurately identify the target sound signal in the complex marine sound environment that the user pays attention to.The theory studied in this paper has high academic research value,and the designed system has a good market application prospect.