不需双端检测的声回波对消算法研究

Acoustic Echo Cancellation Algorithms without Double-Talk Detection

声回波对消中双端对讲情况下的近端话音对自适应算法有很大影响。为避免双端话音检测,在滤波型LMS算法基础上,用远端信号和误差输出信号的和代替远端信号去激励预测误差滤波器,降低近端话音的影响。另为进一步提高算法抗近端干扰的能力,做了变步长的改进,首先将步长反比于输出信号预测误差的短时功率,其次将步长正比于预测误差的互相关系数。实验表明,文中提出的两算法在近端话音出现时表现出较好的性能,其中第二种有更好的稳态失调。

The presence of near-end signal in the double talk period is of great influence on the adaptive algorithm in acoustic echo cancellation system. In order to avoid double-talk detection, an AEC algorithm based on Filter-LMS algorithm is proposed, which applies the sum of far-end signal and output error instead of the far-end signal toactiviatetheforecast error filter, and thus to contributes to reducing the influence of near-end signal. In addition, to further enhance the near-end anti-interference ability, two other algorithms with variable step are devised： one makes the step inversely proportional to the short term power of output forecast error, while the other proportional to the correlation coefficient of forecast errors. Experiments show that both algorithms have good performance when the near-end signal appears, of which the second one is more stable.