聲學回聲消除系統的設計與實現

Abstract

回聲消除為免持裝置中之一重要部份，負責消除由於同時包含麥克風以及揚聲器的系統所導致的回授回音。 早期的回聲消除主要是透過設計一些適應性濾波器來達成，但使用這種方法需要很大的運算量。 因此我們利用串接固定係數濾波器與適應性濾波器並採用依據頻率範圍分段處理來降低非必要的運算量。 此方法可應用在許多方面，例如藍芽耳機、手機、視訊會議……等。 在適應性濾波器的演算法中，我們採用LMS及NLMS兩種演算法，並用消相關濾波器以及次頻帶法來提高他們的運算速率和效果。 在實驗的部份，我們遵照ITU-T Recommendation G.168上的測試方法，並將結果以統計學上的方法MANOVA做判定。

ABSTRACT An acoustic echo cancellation system is presented in this paper. The proposed systems mainly consist of acoustic echo cancellation (AEC) processing and system. The AEC module seeks to reduce the echo resulting from the acoustic feedback in loudspeaker-enclosure-microphone (LEM) systems. In many audio applications, acoustic echo cancellation processing plays an important role when users in video conference or using hands-free system like Bluetooth. The conventional method was used adaptive filters to identify the echo paths adaptively in response to the environmental changes. Besides, we will introduce some other methods, like fixed filter method, hybrid method, sub-band (N)LMS method, block (N)LMS method. In this paper, we combined pure adaptive filter with fixed filter methods to develop hybrid method. Hybrid method has advantages of the above-mentioned two methods and doesn’t have their defects. A parameter, ERLE (Echo Return Loss Enhancement) is used to evaluate the performance, and the standard of performance is referred to ITU standard G .168. A double-talk detection technique is applied to estimate what situation does users in, we can tune the step size of adaptive filter according as the result of double-talk detection to avoid its algorithm diverging. A de-correlation filter is also used alongside sub-band processing to improve the performance and to reduce the complexity of adaptive filter. We will compare all of above-mentioned methods, and approximate total operations of each method. All algorithms are implemented on the platform of a fixed-point digital signal processor (BF533) and a floating-point digital signal processor (SHARC 21364). The process results were recorded for tests and evaluate performance, and presented by Cool-editor.