一种MMSE语音增强算法的研究与实现

介绍了单话筒采集条件下基于语音短时对数谱的最小均方误差(MMSE-LSA)估计的语音增强算法,以及语音帧和噪声帧判别的有声/无声检测方法.将语音信号的相位提取后存储起来,然后对纯净语音的短时对数谱作最小均方误差估计,处理后的语音由估计得到的幅度谱和存储的相位重建.试验证明MMSE-LSA的增强效果很好,尤其在信噪比低时更为明显.

A Study and Realization of a MMSE Speech Enhancement Algorithm

This paper introduces the speech enhancement algorithm based on the Minimum Mean-Square Error Short-Time Log-Spectral Amplitude Estimation (MMSE-LSA) under the single input condition as well as the method for distinguishing between speech and noise frame. The phase of the speech signal is stored and the amplitude is estimated using the MMSE-LSA. The signal processed is reconstructed with the amplitude estimated and the phase stored. Experiments show that the algorithm has good performance in speech enhancement, especially when the SNR is low.