改进的基于语音存在概率的噪声功率谱估计算法

噪声功率谱估计是语音增强算法中的关键技术之一.针对在非平稳噪声环境下噪声功率谱估计不够准确的问题,采用了基于后验语音存在概率的噪声功率谱估计算法,并对其中的语音存在概率估计算法进行了改进.利用语音信号的短时平稳性,在时域和频域上利用相邻帧和相邻频点的相关性估计当前帧的语音存在概率.仿真结果表明,该算法与原始算法及改进的最小值控制递归平均噪声估计(IMCRA)算法相比,能够有效提升非稳态噪声环境下噪声功率谱估计准确度和语音增强效果.

Improved Noise Power Estimation Based on Speech Presence Probability

The noise power estimation is one of the key technologies of speech enhancement algorithm. Aiming at the problem that the estimation of noise power in non-stationary environment is not accurate enough, we adopted the noise power estimation algorithm which based on posterior speech presence probability, and the posterior speech presence probability is improved. Because of the speech signal have short-time stationarity, so we can estimate the probability of the presence of speech by calculating the correlation value of adjacent frames and adjacent frequencies. Experimental results show that the new algorithm can yield good performance on estimating the noise power in non-stationary environment and speech enhancement comparing to the original algorithm and the IMCRA algorithm.