Research On Signal-to-noise Separation Method Of Laser Monitoring

Laser monitoring was first proposed and reported in 1988.It uses laser modulation and demodulation to achieve near-real-time modern positioning and monitoring technology for targets.Compared with other electronic monitors,laser monitoring technology has the advantages of strong flexibility,good concealment,and high security.It has received extensive attention in recent years and has become an important research direction in the development of monitoring technology.National defense is also of great significance.Due to the interference of external environmental factors during the laser transmission process,the mixed signal obtained by demodulation contains both the voice target signal and the interference noise.Under different monitoring conditions,different types of interference noise are generated.Therefore,the selection of signal-to-noise separation method is particularly important for noise with different characteristics.The traditional speech enhancement algorithm has a good suppression effect on relatively stable noise,and in recent years,supervised speech signal-to-noise separation has also shown certain advantages when dealing with non-stationary noise.For the various noise problems in laser monitoring,we classify and use different methods to deal with it.The main research contents are as follows:1.When the monitoring device is in working state,the device components and internal circuits will generate a certain amount of system noise,affecting the quality of the monitoring voice.Considering the characteristics of the system noise itself,this paper uses Bayesian-based statistical model algorithm to achieve the separation of signal-to-noise of the monitored speech.Experiments show that this method can improve speech quality and improve speech intelligibility more effectively than other traditional algorithms.2.In response to the echo problems in the laser monitoring system,this paper proposes a cascaded DNN structure based on the deep neural network(DNN)and through the secondary processing of residual noise.The effect of suppressing echo interference and improving voice quality is achieved.3.In view of the problem that multiple types of noise will be generated in different monitoring scenarios,in order to improve the universality of the system,this paper proposes a signal-noise separation method based on fusion features.This method combines the traditional speech enhancement algorithm with supervised learning to calculate the acoustic characteristics of the estimated speech,noise and phase separately,which improves the network's ability to fit the amplitude and phase targets,thereby improving the quality and intelligibility of the final separated speech degree.In this study,according to the characteristics of system noise,echo,and multiple types of noise,statistical model algorithm,cascaded DNN algorithm,and algorithm based on fusion features are used to achieve the separation of signal and noise of monitoring speech.These algorithms have effectively suppressed the related noise and improved the monitoring effect under the premise of ensuring that the voice signal is not distorted.