采用复高斯分布模型的两步噪声幅度谱估计算法

发布时间：2018-05-08 08:20

本文选题：语音增强 + 非平稳噪声　；参考：《信号处理》2017年07期

【摘要】：噪声幅度谱估计是有效抑制外界噪声干扰、提高语音增强算法整体输出性能的重要环节。但目前针对该问题的研究相对较少,常用的语音激活检测算法只能在语音不存在阶段对噪声信号的幅度谱进行更新或估计,无法适用于更为复杂的非平稳噪声环境。为克服这一问题,本文基于噪声频谱的复高斯分布模型假设,提出了新型的两步噪声幅度谱估计算法。算法首先采用软判决技术计算噪声信号的功率谱,然后再结合复高斯分布条件下信号幅度谱和功率谱之间的数学关系间接地获取噪声幅度谱的估计。文中基于这一结论给出了两种估计算法,并在多种噪声环境下对它们的性能进行了仿真评估,其测试结果有效表明了提出算法优良的估计性能。
[Abstract]:Noise amplitude spectrum estimation is an important step to effectively suppress external noise interference and improve the overall output performance of speech enhancement algorithm. However, there are few researches on this problem at present. The commonly used speech activation detection algorithms can only update or estimate the amplitude spectrum of noise signals at the stage of speech non-existence, and can not be applied to more complex non-stationary noise environments. In order to overcome this problem, a novel two-step noise amplitude spectrum estimation algorithm is proposed based on the assumption of complex Gao Si distribution model of noise spectrum. At first, the soft decision technique is used to calculate the power spectrum of noise signal, then the estimation of noise amplitude spectrum is obtained indirectly by combining the mathematical relation between signal amplitude spectrum and power spectrum under the condition of complex Gao Si distribution. Based on this conclusion, two estimation algorithms are presented, and their performance is evaluated by simulation in various noise environments. The test results show that the proposed algorithm has good estimation performance.
【作者单位】：烟台大学光电信息科学技术学院;烟台大学计算机与控制工程学院;吉林大学通信工程学院;
【基金】：国家自然科学基金(61201457,61005021) 山东省自然科学基金(ZR2014FQ016) 烟台大学研究生科技创新基金重点项目(YDZD1711)
【分类号】：TN912.3

【参考文献】