共極模型分析在聲學回音消除之應用

Abstract

隨著人與人間溝通日益頻繁，視訊會議及免手持聽筒通訊系統的發展已成必然的趨勢。然而有效控制免持聽筒所造成的聲學回音回溯，卻是影響此系統發展成敗的關鍵技術。有別於一般被動式改善通訊環境以期降低回音回溯的方式，本論文中提出採用自迴歸移動平均模型來模擬回音路徑，並利用無限脈衝適應性濾波器和最小均方調變演算法來建構完成，以達到主動消除回音的目的。同時，我們更深入研究聲波在密閉空間中傳播形成駐波的相關理論，推論固定房間中的脈衝響應均擁有共同的極點。並將此共同極點帶入回音消除濾波器中，以期降低計算量達到即時製作要求。實驗結果顯示共極模型除可降低計算量外，亦可提供適應性濾波器一較佳起始值，進而提升回音消除效能。

In many hand-free applications, it is necessary to reduce the acoustic echo incurred by the coupling between the loudspeaker and the microphone. In this thesis, we used an IIR adaptive filter to simulate the room impulse response, which describes the sound transmission characters between a source and a receiver in a room. The simulation of the time-varying room impulse response is formulated as a system identification problem that is amenable to the application of normalized least-mean-square algorithms. The study of sound wave transmission reveals that all possible echo paths in the same enclosure room are characterized by common poles, each of which corresponds to the resonance frequency (eigenfrequency) of one room transfer function. Proceeding in this way, only the zeros of an IIR echo canceller need to be updated. Furthermore, the common poles can characterize part of the echo path and hence it offers a better choice for initial parameter setting of the IIR filter. Simulation results indicate that the proposed echo canceller with common poles not only reduces the computational complexity but also results in more echo cancellation compared with other IIR cancellers.