Group Birds Monitoring Application Based On Birds Voice Separation

Birds,as the important biological indicators in the Earth's ecosystem,need be long-term continuous monitored.Traditional methods of observation is using manual methods such as sample methods and marker replenishment methods.These methods have large errors in monitoring results and make it difficult to provide accurate information.In view of the shortcomings of traditional observation methods,this thesis uses speech separation technology to separate the mixed voices of the birds collected by the microphones.According to the separation results,the number of birds can be estimated to achieve the monitoring effect.This method does not require a lot of manpower and material resources.More effective monitoring results are available.The main work of this thesis is as follows:(1)Constructing a mixed model of bird speech signals and reducing the noise of the voice signal.Because of the reverberation and echo of the voice signal in the open forest,the echo-free model is constructed,and the Wiener filtering signal denoising algorithm is studied and the noisy simulation is performed.On the other hand,according to the actual scene,this paper analyzes the relevant factors affecting the voice collection,determines the appropriate microphone model and the type of microphone array which is suitable in the real forest,and designs a reasonable acquisition end as the signal input port.The use of a microphone array can greatly reduce the noise interference caused by environmental factors.(2)The DUET(Degenerate Unmixing Estimation Technique –DUET)blind speech separation optimization algorithm has been investigated.For the error caused by that the DUET algorithm default source signal does not overlap the sparse representation,this thesis improved effects through rotation transformation after theoretical derivation.(3)The fast density clustering blind separation algorithm and its application in bird speech separation has been investigated.This is a method of estimating by using the cluster center.Firstly,the collected bird voice signal is subjected to time-frequency domain conversion,and then detect single source point,and the single source point interval and the multiple source point interval are divided.The signals of the single source point and the multiple source point interval are processed separately,and then combined the recovered voice which belong to the single source point and the multi-source point and merged according to the DUET algorithm,finally,we can get the high quality source voice signal by Fourier transform.(4)The FastICA(Fast Independent Component Analysis,ICA)algorithm for simulating negative entropy maximization proves its better effect in the separation of mixed bird speech signals.Firstly,the algorithm whitens and centers the collected voice,and then sets the number of iterations and the number of components,selects the initial weight,and iterates according to the steps of the algorithm until the data converges.The simulation proves that the algorithm has good separation effect and is suitable for the monitoring application of microphone array to bird voice.This thesis designs an application scheme,and designs a simulation system GUI which combines with computer and recorder to simulate the real scene,and tests the performance of the bird mixed speech separation algorithm.The test results show that the similarity between the separated speech and the source speech exceeds 85%,and the estimated error is less than 4 as the number of birds increases to 500,which proves the feasibility of the separation algorithm studied in this thesis.