一种基于自适应噪声估计的宽带语音增强算法

为解决传统算法对噪声适应性较差，残留音乐噪声较强的问题，本文提出了一种基于自适应噪声估计的宽带语音增强算法。该算法可应用于宽带语音编码器，以提升在噪声环境下的编码质量。本文所提算法利用谱熵对噪声类型进行有效的判别，将背景噪声分为白噪声和有色噪声两类，并根据噪声特性选择适当的噪声估计方法。在白噪声背景下，选择一种谱平滑的方法；在有色噪声背景下，则选择经典的最小值控制递归平均算法。在此基础上结合经典的统计模型方法，构建一种具有较强噪声鲁棒性的宽带语音增强算法。在ITU-T G.160标准下对算法进行性能测试，测试结果表明，在不同强度的背景噪声环境下，增强语音的信噪比提高都较为明显。同时，在低信噪比情况下，该算法有效的抑制了严重影响听觉质量的音乐噪声现象。 

A Wideband Speech Enhancement Method Based on Adaptive Noise Estimation

In order to improve the noise tracking ability and reduce the residual musical noise, an adaptive noise estimation method for wideband speech enhancement is proposed in this paper. The algorithm can be applied to wideband speech coder and improve encoding quality in noisy environment. The spectral entropy is adopted to classify the background noise into two categories, one for white noise, and the other for the colored noise. Then, appropriate noise estimation method is selected based on the noise characteristics. In white noise background, we choose a spectral smoothing method. In colored noise background, we utilize the minima-controlled recursive averaging algorithms. Finally, the proposed method is combined with the classical statistical model based speech enhancement, and the robustness to noise is improved evidently. The performance of the proposed method is evaluated under the standard of ITU-T G.160. The experimental results show that the algorithm is effective for improving the SNR in the different noise environments. Meanwhile, in low SNR conditions, musical noise is reduced effectively.