啸叫快速抑制的助听器回声抵消算法

发布时间：2018-11-13 13:27

【摘要】：针对助听器回声路径快速变化下易产生啸叫的问题,本文提出一种变步长标准最小均方差-陷波器(Variable Step Normalized least mean square-Notch Filter,VSN-NF)算法。在回声路径相对稳定时,提出一种基于状态分类的变步长标准最小均方差算法来估计回声信号。算法根据滤波器系数能量的长时平均值和短时平均值,将滤波器当前状态分为收敛态、过渡态与稳态,并根据不同状态选择不同的步长。在路径突然变化并产生啸叫时,算法通过关闭变步长NLMS算法来稳定啸叫频点,然后基于ZoomFFT算法动态生成陷波器来进行啸叫抑制;当啸叫抑制后,再开启变步长NLMS进行回声估计。针对易产生多频点啸叫的回声路径,VSN-NF算法还引入不同频带的两个陷波器来进行双频点啸叫抑制。同其它助听器回声抵消算法的对比实验显示,VSN-NF算法的回波抵消性能最好,尤其具有快速啸叫抑制能力。此外,算法生成的语音质量较高,实时性能好,适合于像助听器类的低功耗、小体积产品。
[Abstract]:Aiming at the problem that the echo path of hearing aid is easy to produce howling under the rapid change of echo path, a variable step size standard minimum mean-variance notch (Variable Step Normalized least mean square-Notch Filter,VSN-NF algorithm is proposed in this paper. When the echo path is relatively stable, a variable step size minimum mean square error algorithm based on state classification is proposed to estimate the echo signal. The algorithm divides the current state of the filter into convergence state, transition state and steady-state according to the long-time average and short-time average of the filter coefficient energy, and selects different step sizes according to different states. When the path changes suddenly and produces howling, the algorithm stabilizes the whistling frequency by turning off the variable step size NLMS algorithm, and then dynamically generates a notch based on the ZoomFFT algorithm to suppress the whistling. When the path is suppressed, the variable step size NLMS is turned on to estimate the echo. In view of the echo path which is easy to produce multi-frequency point howling, VSN-NF algorithm also introduces two notch devices with different frequency bands to suppress the dual-frequency point howling. Compared with other echo cancellation algorithms of hearing aids, VSN-NF algorithm has the best performance of echo cancellation, especially the ability of fast howling suppression. In addition, the algorithm has high speech quality and good real-time performance, which is suitable for low power and small volume products such as hearing aids.
【作者单位】：南京工程学院通信工程学院;东南大学信息科学工程学院;
【基金】：国家自然科学基金(61301219,61375028,61301295) 江苏省自然科学基金(BK20130241) 南京工程学院高层次引进人才科研启动基金(YKJ201526)资助
【分类号】：TN912.3

【参考文献】