一种基于改进的谱减法的语音增强算法  

Speech enhancement algorithm based on improved spectral subtraction

作　　者：高留洋 朱文 桑振夏 米兰 

机构地区：[1]63898部队

出　　处：《现代电子技术》2012年第17期60-62,共3页Modern Electronics Technique

基　　金：国家863计划重点项目(2008AA011001)

摘　　要：为改善语音质量,提高语音识别系统的性能,提出了一种新的基于改进的谱减法的语音增强算法。新算法在所有噪声都能转化为加性高斯白噪声的基础上,依据高斯白噪声幅度谱服从瑞利分布的特点,对噪声幅度谱进行精确的估计,并采用频带方差进行端点检测以及时检出语音。仿真分析表明:该方法能够较好地抑制噪声,噪声消除效果较传统算法具有明显的提高。To improve the speech quality and enhance the performance of noisy speech signals recognition, a new speech enhancement algorithm based on improved spectral subtraction is proposed and analyzed. In contrast to the standard spectral subtraction algorithm, the new algorithm accurately estimates the noise based on the fact that all noise can be changed into Additive White Gaussian Noise （AWGN） and according to the feature that the amplitude spectral of narrowband white Gaussian noise obeys Rayleigh distribution. This algorithm also adopts a new speech activity detection technology based on frequency band variance to detect speech activity. The simulation analysis results indicate that the algorithm in this paper performs better in noise elimination than standard spectral subtraction.

关 键 词：语音增强 谱减法 端点检测 加性高斯白噪声 瑞利分布 

分 类 号：TN912.353.4[电子电信—通信与信息系统]