Research On Laser Interception And Intelligent Speech Processing Technology

As one of the main methods of information stealing,voice interception plays an important role in the field of national defense and security.Compared with non-laser voice interception,laser voice interception does not require the installation of listening equipment close to the detected object,and has the advantages of high concealment and portability,which has attracted special attention.Among them,coherent laser voice interception has high accuracy.Antiinterference ability has become a research hotspot in this field.This thesis has studied laser voice interception technology based on superheterodyne laser coherent vibration measurement.Deep learning-based voice noise reduction and voice synthesis have improved the listening effect and functions of laser voice interception,and realized intelligence Voice processing.The specific research content is as follows:First,the principle of laser listening is analyzed,and the laser superheterodyne coherent vibration measurement and demodulation model is established.Two demodulation methods,the structure of the optical system,the surface roughness of the object,the speckle noise,the object under the action of the sound field and the influence of forced vibration characteristics and other factors on the laser vibration detection capability and measurement accuracy are simulated and analyzed.The above research has laid a theoretical foundation for follow-up research;Secondly,through voice signal preprocessing such as vibration measurement signal nonlinear correction,voice signal time-frequency domain noise reduction,voice signal high-frequency restoration,etc.,the nonlinear error introduced by the measurement signal is corrected,the time-frequency noise of the voice detection signal is eliminated,and the reconstruct high-frequency information lost by the low-frequency response of the material;further proposes a voice signal noise reduction method based on the generation of the anti-network structure,according to the characteristics of the human voice,the noise of laser interception voice signal is reduced by Deep learning.On this basis,through GE2E(Generalized End-to-End Loss)network,Tacotron2 network and Wave Net network are combined to realize the simulation and generation of voice listening signals,and broaden the laser voice interception function.Finally,a laser voice interception experiment system was built,and vibration measurement and voice interception experiments were carried out.The instantaneous measurement accuracy of piezoelectric ceramic simple harmonic vibration within 2m is better than 200nm;laser voice interception is performed on the surface of the speaker within 2m and voice noise reduction,the MOS(Mean Opinion Score)score is 3.6,and the paper surface with a distance of 2m,3m,5m and 10 m is further carried out with laser voice detection and noise reduction,and the MOS(Mean Opinion Score)score is 3.4.After noise reduction,realized voice synthesis experiment base on the laser interception voice on the paper surface at a distance of about 10m;verify the effectiveness of the above-mentioned deep learning-based laser listening voice noise reduction and synthesis method,and realize the intelligent processing of laser interception voice.