Echo Cancellation Method Research Based On Prediction Residuals And Adaptive Order

The rapid development of network technology facilitated the emergence of all kinds of communication mode, the emerging instant messaging software Represented by We Chat、Skype、Yixin, which taked network as its carrier, increased its occupied proportion in the communication year by year. However, the existence of echo signal reduced the quality of voice communication and hindered the promotion of IP voice communication. Therefore, theoretical research and practice of echo cancellation was important.Aiming at double-talk testing for echo cancellation system and combining with the double-talk detection algorithm which includes energy and double filter detection algorithm, this article analyzed the principle of echo generation deeply and proposed double-talk detection algorithm that based on prediction residuals. The essence of which was based on energy, its construction was simple and could be easily implemented. Compared with Geigel algorithm, it had a wide range of threshold selection and could effectively reduce misjudgment rate. According to the unknown characteristics of the echo environment in the actual application, we proposed order-variant NLMS algorithm. By using order adapting module, it could adjust the filter order until a better order. Compared with NLMS algorithm、PNLMS algorithm,it reduced steady-state error and promoted echo cancellation effect. In addition to this,we also added voice signal preprocessing module、nonlinear processing module and simple comfort noise generating module, it consummated echo cancellation system solutions and improved the voice quality of IP speech communication system,providing guarantee for echo cancellation.On the basis of the above scheme, we introduced IP voice communication system hardware equipment of Xiamen Leelen Company and seted up IP voice communication hardware environment. we embeded the echo cancellation system algorithm in tht IP voice communication system of Xiamen Leelen Company.Meanwhile, we solved several problems, such as, proximal and distal speech signals were synchronized、the apply data of single and double tracks and numeric types were not matched. The voice communication experiment echo cancellation effect is improved. By saving the data of real-time communication, we analyzed the performance of this paper algorithm.