Bird Song Enhancement System Based On Wireless Acoustic Sensor

Traditional bird song enhancement methods based on wireless acoustic sensor networks(WASNs)usually use distributed beamforming with post-filter.These methods have some problems such as slow convergence rate,high energy consumption,poor noise reduction and large bird sound distortion,which limit its practical application.In this paper,a bird song enhancement system based on WASNs is proposed for solving these problems.The main work is as follows:1.An improved distributed MVDR beamforming algorithm(IDMVDR)based on WASNs is proposed.This method serves as a beamformer of the system.Firstly,the average Metroplois weight consistency algorithm is introduced during the consistency phase for consistency update.Then the noise PSD estimation algorithm of MVDR beamforming is optimized.Finally,several experiments are designed.Compared with three traditional methods,the convergence speed and enhancement performance of IDMVDR are verified.2.An improved DNN-based method for bird song enhancement is proposed.This method serves as a post-processing part of the system.A bird song and noise database for training DNNs is established.DNN's noise-aware block is improved and Dropout is introduced to optimize the model.Firstly,the performance and generalization ability of the proposed DNN model are verified by multiple sets of contrast experiments.Then,we evaluate the bird song enhancement system IDMVDR-DNN.The results show that SegSNR and PESQ of the proposed IDMVDR-DNN increased by 129.3%and 64.9%compared with DDSB-WF respectively.So this bird song enhancement system get better enhanced bird song quality.3.The software of the bird song enhancement system is designed based on the MATLAB GUI development tool.This software can realize the simulation of different bird song enhancement methods under different topology of WASNs and automatically calculate the measure index.The bird song enhancement system based on WASNs proposed in this paper has the advantages of fast convergence rate,good enhancement performance.This method is not limited by the WASNs'topology,and greatly reduces the energy consumption.Overall,it is suitable for enhancing the bird song which is collected in natural environment by WASNs,and providing a better quality bird song signal for the further bird identification and research.