基于Gamma语音模型的语音增强算法

提出了一种新的基于Gamma语音模型的语音增强算法。首先,在假定语音和噪声的短时DCT系数分别服从Gamma和Gaussian分布的基础上,推导了最小均方误差意义下的语音信号短时DCT系数估计;然后,根据语音存在概率估计,提出了语音信号短时DCT系数估计的修正因子。在增强算法中,提出了基于Gamma语音模型的改进最小统计量控制递归平均(IMCRA)噪声估计算法。仿真结果表明,该算法不仅在噪声抑制性能方面优于近两年国际上提出的几种基于Gaussian语音模型的语音增强算法,而且在增强语音质量方面也具有更好的性能。

Speech enhancement with Gamma speech modeling

A new speech enhancement system with Gamma speech modeling was proposed.The clean speech compo-nents were estimated by minimum-mean-square-error(MMSE) estimator under the assumption that the clean speech DCT coefficients were modeled by Gamma distributions and the noise DCT coefficients were modeled by Gaussian distribu-tions.Then,the MMSE estimator under speech presence uncertainty and a Gamma model was derived.Furthermore,the noise power was estimated by the IMCRA algorithm with the Gamma model.The simulation results show that the pro-posed algorithm yields less residual noise and better speech quality than the Gaussian based speech enhancement algorithms proposed in recent years.