室內聲學迴音消除之研究

Abstract

聲學迴音干擾問題，發生於通訊產品與音訊系統，舉凡有線話機、行動電話、免持聽筒、音響擴音器、助聽器、耳機麥克風…等等，其中皆包含兩大重要聲學元件：揚聲器與麥克風。當此二元件共同置於封閉空間或距離相近時，聲波能量足以由揚聲器傳入麥克風，所產生之聲學迴音將影響通訊品質，甚至造成聽覺上的不適。 本文根據適應性主動控制理論，與數位信號處理技術，採用有限脈衝響應（FIR）數位濾波器為系統架構，並以Normalized LMS演算法進行系統最佳化運算，利用類比/數位–數位/類比信號擷取轉換裝置，結合個人電腦為研發平台，建立一聲學迴音消除系統。 聲學迴音消除實驗，以單頻信號與隨機噪音信號輸入，於壁面反射性高之密閉迴響環境進行測試，分析比較不同頻率、參數設定與運算方式，對聲學迴音消除成效之影響，藉以研究其系統性質。 依實驗結果顯示，單頻信號依頻率高低不同，平均ERLE可達20至50 dB，功率頻譜迴音衰減量可達60 dB以上；隨機噪音依頻帶高低，平均ERLE達約2至10 dB，功率頻譜迴音衰減量約為3至12 dB；固定頻寬之隨機噪音，平均ERLE與功率頻譜迴音衰減量，皆可達約10至15 dB。

The interrupt resulting from the acoustic echo exist in some situations dealing with the communication products and audio systems such as telephone, cellphone, hand–free telephone, stereos, hearing aids and headset etc. Among those products, two important acoustic components such as loudspeaker and microphone are included. When the two acoustic components as described are placed in an enclosure or placed too close to avoid the propagation of acoustic energy from speaker to microphone, the acoustic echo to be produced will influence the communication quality, or further causes an uncomfortable hearing to happen. Upon the adaptively active control theory and the digital signal processing technique, this study adopts the finite impulse response digital filter as the system structure. Regarding to the optimum operation for the control system, a normalized least-mean-square algorithm is selected as well. To further set up the acoustic-echo cancellation system for this study, the combination of an A/D-D/A converter data acquisitive equipment with some personal computer is used as the required platform. For the experiment of the acoustic-echo cancellation, an enclosure with highly reflective walls is used as the experimental site, and the selected input signals are both of pure-tones and random. The system performances are identified by comparing the methods to set up the parameter of system, its operation performance and their influence on the effectiveness of echo cancellation for various frequencies. Experimental results show that the averaged ERLE by 20 to 50 decibels for various frequencies from low to high pure-tone sounds are reached. The reductions of echo power spectrum by more than 60 decibels are reached. As regards to the case of randomly acoustic echo, either the averaged ERLE by 2 to 10 dB or the reduction of power spectrum by 3 to 10 dB is achieved. Moreover, for the case of narrow-band of acoustic echo, the reduction by more than 10 to 15 dB of averaged ERLE and power spectrum are achieved.