免手持(hands-free)設備應用在日常生活中越來越頻繁。使用網路通訊系統可省下電話費,讓網路電話更為多數人所使用,而使用免手持式網路通訊系統時最大的詬病就在於聲學回音(acoustic echo)造成通話品質上嚴重的降低及影響,因此回音消除就是一個解決方向。在遠端(far-end)與近端(near-end)兩端使用者的系統加入聲學回音消除(acoustic echo canceller, AEC),利用可適性濾波器(adaptive filter)的特性來達到良好的通話品質。 雙邊通話偵測錯誤會造成大量的殘留回音或近端使用者聲音訊號失真,因此要如何消除回音改善聲音品質,因此精準的雙邊通話偵測是一個相當重要的要求。根據模擬實驗的觀察,如果雙邊通話偵測使用交相關係數(cross-correlation coefficient)來做為檢測指標參數,偵測錯誤通常都是發生在最剛開始收斂的時候,或者在非常吵雜的環境之下,進而影響整體效能。本論文的回音消除分為兩部分。第一部分是利用相關方法(correlation method)與正規化錯誤訊號功率(normalized error signal power),進行第一部分的回音消除。第二部分藉由噪音抑制來消除殘餘回音,加強整個系統的通話品質。我們的方法驗證是透過MATLAB軟體來模擬,並從結果來驗證我們所提出的強健式雙邊通話偵測可以有效的消除回音及殘留噪音。
The hands-free system are more and more popular in everyday life; in addition, using of network communications system can save telephone charges and let more people using VOIP. When using network communications system the biggest problem of the hands-free system are that acoustic echo may degrade the voice quality. Therefore, a solution to this kind of problems is to use echo cancellation. By adding an acoustic echo canceller (AEC) using adaptive filter the far-end and near-end users of a communication system can achieve good voice quality. Double-talk detection (DTD) errors can result in either large residual echo or distorting the near-end talker’s input speech. Thus accurate double-talk detection is an important requirement in the echo canceller to improve the speech quality. We propose a double-talk detector with two stage functions. First, use a correlation method and the normalized error signal power to eliminate acoustic echo. Second, we use a residual noise reduction algorithm to enhance voice quality of the system. Using MATLAB Simulation, the experimental results showed that the proposed system can improve the performance and robustness of the acoustic echo canceller, the residual noise reduction.