基于递归最小追踪的噪声互功率谱估计算法

针对改进最小追踪噪声互功率谱估计方法存在的噪声过估计的问题。提出一种基于递归最小追踪的噪声互功率谱估计算法。该方法中的平滑因子使用了递归平均技术,在估计噪声互功率谱时,会根据每个频点的实际信噪比作相应的调整。仿真结果表明,该噪声估计算法应用于一个语音增强系统时,取得了较小的噪声均方估计误差及较好的感知语音质量评价(PESQ)得分。

A noise cross power spectral density estimation method based on recursive minimum tracking

Aimed at the over estimation problem of the improved minimum tracking method. A noise cross power spectral density estimation algorithm, based on recursive minimum tracking, is put forward. The method of smoothing factor used the recursive average technology. When estimated the noise CPSD, the smoothing factor will make a corresponding adjustment, according to actual signal-to-noise ratio of each frequency point. Simulation results indicate that the proposed noise estimation algorithm yields lower estimation mean square error and better Perceptual Evaluation of Speech Quality(PESQ) scores when integrated in speech enhancement.