偵測雙邊對話之適應性回音消除

Abstract

當雙邊同時對話時,適應性回音消除器回音消除的效果會變得很差.本篇 論文針對誤差訊號標準(ERL)與麥克風接收訊號標準(EPL)做數學分析,藉 以偵測雙邊對話的狀態.此外,也會探討影響ERL與EPL的主要因素.當我們 偵測到雙邊對話時,會採用步階調整與係數回存的方法以確保回音消除的 品質.為了要更精確地分辨房間響應改變與雙邊對話的狀態,對於麥克風接 收訊號標準會有更嚴謹的分析.最後,由電腦模擬的結果印證上述的分析. An adaptive acoustic echo canceller(AEC) suffers a significant degradation when double talk arises. To detect double talk, analytical results on ERL and EPL are derived. We propose step- size adjustment and filter coefficients-restore methods to assure echo cancellation quality as double talk is detected. A rigorous analysis on EPL is performned to distinguish room impulse response variations from double talk. Finally, computer simulations are provided to verify our analysis.