一种改进型IMCRA非平稳噪声估计算法

摘要

The Improved Minima Controlled Recursive Averagiiig(IMCRA) algorithm has large delay, this paper presents an improved IMCRA algorithm for non-stationary noise estimation. The speech present probability is calculated in three iterative to determine the length of search window, and noise estimation is limited or compensated with safety mechanism. Experimental result indicates that the algorithm is able to estimate noise more accurately by reducing the delay of the noise minima tracking. It is suitable for the speech enhancement in acute non-stationary noise environment.%针对IMCRA噪声估计算法时延较大引起的噪声欠估计问题,提出一种改进型IMCRA非平稳噪声估计算法.该算法采用3次谱平滑的方法估计语音存在概率,控制搜索窗长度.在噪声过估计或欠估计的情况下,引入安全机制对噪声估计值进行限制或补偿.实验结果表明,与IMCRA算法相比,改进算法能减小噪声估计的时延,提高噪声估计精度,适用于噪声功率变化范围较大的语音增强.