基于改进语音存在概率的自适应噪声跟踪算法

在非平稳环境下，由于时间递归平均噪声功率谱估计算法会出现跟踪延迟和估计误差等问题，本文采用一种新的方式对其核心部分语音存在概率（speech presence probability, spp）进行估计。利用时域特征能量与频域特征谱熵的比值能熵比作为新的特征来构建其与spp的正比关系，从而得到当前语音帧的spp估计值；然后用双平滑系数对该值进行平滑；最后结合时间递归平均算法得到估计的噪声功率谱。该算法充分利用语音帧频点的特征信息控制spp的估计值，以此自适应地跟踪噪声变化。实验结果表明:在地空通信环境下，该方法能够准确且连续地跟踪噪声功率谱、快速响应其变化。集成到语音增强系统后，可以提高语音质量，降低残留噪声。 

An Adaptive Noise Tracking Algorithm Using Improved Speech Presence Probability

The time recursive averaging noise power spectrum estimation algorithm had the problems of tracking delay and estimation error in non-stationary noise environment. This paper proposed a new spp(speech presence probability) estimation algorithm which was a core part of noise power spectrum estimation algorithm. Firstly, the ratio of energy in time domain to spectrum entropy in frequency domain was used as a new feature to obtain the spp of current speech frame; then the estimation of spp was smoothed by the double smoothing coefficient; finally, the estimated noise power spectrum was obtained by combining with time recursive averaging algorithm. This algorithm made full use of the feature information of frequency point to control the spp and track the noise changes adaptively. The simulations show that this method can accurately track the noise power spectrum and quickly respond to the noise power spectrum changes in the air-ground communication environment. Integrating it into speech enhancement system can improve speech quality and reduce residual noise.