本文研究了双通道回声抵消系统的三个核心模块:频域多延时自适应滤波器算法(MDF),双端说话检测算法(DTD),以及残留回声抑制算法(RES)。针对频域算法的特点提出了改进的双端说话检测算法和基于维纳滤波与谱减法的残留回声抑制算法,然后以上述三种算法为核心模块实现双通道回声抵消系统,并对系统定点化以便在定点DSP处理器上实时实现,分析并解决了定点化的精度问题所带来的影响。
In this paper, three kernel modules of the stereo acoustic echo cancellation (AEC) system, Multi-delay Frequency Domain (MDF) adaptive filter algorithm, Double Talk Detection (DTD), and Residual Echo Suppression (RES) are investigated. Considering the properties of the frequency domain algorithm, an improved DTD scheme and a novel RES scheme, based on the combination of Wiener filter and spectrum subtraction method, are proposed. An integrated AEC system is designed and the fixed point realization of the system is investigated for the purpose of implementation on a fixed point DSP board. The precision problem caused by the fixed point realization is also discussed.