Research On Acoustic Signals Separation Technology Based On Energy Constraints In Improved Generative Networks

With the rapid development of artificial intelligence technology,acoustic detection and recognition technology based on deep learning,as a key means of human-computer interaction,has gradually become a hot spot in the field of signal processing.It has a wide application prospects and important application value in military,industrial and civil fields.Acoustic signal separation technology,which combines signal processing theory with acoustic signal separation model,becomes the key to improve the accuracy of acoustic monitoring and recognition under the conditions of low signal-to-noise ratio and multiple acoustic signal interference,and provides guarantee for its wide application in industrial manufacturing,urban security,aerospace and other fields.The main work of this paper is to take the separation of acoustic signals from unmanned aerial vehicles as the background,the variational Bayesian method as the starting point,and the joint distribution of KL dispersion as the core.To solve the energy loss problem of traditional generative adversarial networks in the separation of acoustic signals,an improved networks for acoustic signal separation model based on the variational energy constraints is proposed.The design uses the mixed sound signal as the input of the generation network process.In the model design of the generator network process,the detailed information of the mixed sound signal is saved by the codec structure of the split network.The estimated source signal amplitude spectrum separated by the generation network process and the target true source signal amplitude spectrum monitored manually are used as the input of the discriminator network.In the model design of discriminator network process,the deep neural network VGG-16 is used,and the parallel training of multiple discriminant networks is designed to reduce the waste of resources in single generator network single discriminator network training.Finally,through the field experiments of multi-element UAV sound signal separation and comparison with other sound signal separation algorithms,the excellent performance of the improved generative adversarial network based on the variational energy constraints is verified from the subjective time-frequency diagram and objective performance indicators.