语音系统中的子带自适应回声消除技术

Subband adaptive echo cancellation technique in speech system

提出了一种基于DFT调制滤波器组的子带自适应滤波回声消除技术,首先使用分析滤波器组将麦克风输入信号和扬声器输出信号分成若干个频(子)带,扬声器输出信号作为自适应滤波器的输入信号,估计回声反馈信号,并从子带麦克风输入信号中减去。在每个子带中,采用了归一化最小均方自适应算法,使得子带均方误差最小。子带自适应滤波器的阶数通常比全带自适应滤波器更小,而且子带收敛速度会有所提高,同时计算复杂度也大大减少。在实际语音处理实验中证明,这种方法是有效的。

In this paper, we propose a subband adaptive filtering method for echo cancellation based on DFT modulated filter bank. The analysis filter bank is used to divide the microphone input signal, as well as the loudspeaker output signal both into several subbands. The loudspeaker output signal is used as the adaptive filter input signal to estimate the echo signal, which is subtracted from the subband microphone input signal. In each frequency band, the normalized least mean square adaptive algorithm is used to minimize the subband mean square error. The order of the subband adaptive filter usually is smaller than that of the full band one. The subband convergence speed is also improved and the computational complexity is reduced. The actual voice processing experiments demonstrate the effectiveness of the proposed method.