利用频谱排序和筛选的突变噪声快速估计

针对最小值控制递归平均(Minima Controlled Recursive Averaging,MCRA)算法不能快速跟踪突变噪声的问题，提出了一种基于频谱排序和筛选的突变噪声快速估计方法。该方法在MCRA算法的基础上对带噪语音的功率谱进行排序，筛选出不含语音信号的频点来估计噪声的平均功率谱；当检测到噪声突变时，对当前的平滑参数和状态变量进行校正。仿真结果表明，该方法可以将突变噪声的跟踪时间缩短90%以上；用于语音降噪处理时，音质可以提升约0.4分。该方法具有一定的工程应用价值。

Fast Power Estimation of Mutational Noise by Using Spectrum Sorting and Screening

The noise estimation algorithm of minima controlled recursive averaging(MCRA) cannot rapidly track the mutational noise.To address the problem,a fast power estimation algorithm of mutational noise based on spectrum sorting and screening is proposed.Based on the MCRA algorithm,this method sorts the power spectrum of the noisy speech and screens the frequency points without speech to estimate the average noise power spectrum of the entire frequency band.The current smoothing parameters and status variables are corrected when mutational noise is detected.The simulation results show that the tracking time of mutational noise is shortened by more than ninety percent,and the mean opinion score of de-noised speech has increased by about 0.4.Meanwhile,this method is valuable in practical projects.