Research And Implementation Of Echo Cancellation Algorithm Based On Deep Neural Network

With the rapid development of intelligent digital information technology,on the one hand,deep neural network algorithms have become a h ot research issue,especially in speech processing,which are mostly used in speech recognition and de-reverberation.On the other hand,because VoIP phones have the advantages of high quality,low cost,and convenience,they are widely welcomed once they are launched.With the widespread application of VoIP telephone technology,some problems have also appeared,especially in the realization of the call function,on the one hand,it is affected by environmental factors such as large noise,and on the other hand,the processing of voice signals and network transmission will also cause delays.Echoes are generated,and when the echo path is too large,the call quality will be affected,and the call function will even be unable to continue.Based on the in-depth study of the structure of commonly used echo cancellation algorithms,this paper studies on the one hand to improve the traditional echo cancellation algorithm to achieve the improvement of echo cancellation effect,and on the other hand,it draws on the principle of the popular deep neural network for speech de-reverberation,and tries Deep neural network algorithms are used for echo cancellation.First of all,the current situation of echo cancellation and deep nerves is studied.When studying the traditional NLMS algorithm,the problem of high complexity and poor stability of the algorithm in the echo cancellation process of the NLMS algorithm is studied,and the steps are changed in different periods.The long factor method improves the NLMS algorithm.The theoretical analysis found that it can significantly reduce the computational complexity of the adaptive filter and improve the stability of echo cancellation.The simulation experiment analysis verifies that the improved algorithm has complexity and stability.However,the echo return loss gain value is not much different from the value of the NLMS algorithm,so the effect of echo cancellation is not improved.Then for the adaptive filter architecture to simulate the echo path of the echo cancellation process,especially the problem of weak echo processing in the noisy place,this paper studies a structure based on the deep neural network algorithm,using the deep neural network algorithm to replace the traditional The adaptive filter in the algorithm characterizes the path impact,maps the echo signal,and then eliminates it.After simulation analysis,the echo return loss gain value based on the deep neural network algorithm is better than the method of changing the step factor in different periods It is better to improve the NLMS algorithm,and it can also greatly improve the effect of echo cancellation.Finally,the echo cancellation structure based on the deep neural network algorithm and WEBRTC technology are combined and applied to the emergency command and rescue software.By using this algorithm on the Android system,the echo problem in the emergency communication process can be improved.Finally,the software is tested on the mobile phone,and the voice information is recorded and analyzed.It is found that the algorithm structure achieves the expected effect in echo cancellation under double-ended calls or complex environments,and improves the quality of voice communication.