基于新型先验信噪比估计语音增强算法的对比研究  

Comparison and Research of Speech Enhancement Algorithms Based on New Type of a prior SNR Estimation

作　　者：沈锁金 魏静[1] 高颖[1] SHEN Suo-jin WEI Jing GAO Ying(School of Science and Technology for Opto-Electronics Information of Yantai University, Yantai 264005, Chin)

机构地区：[1]烟台大学光电信息科学技术学院,山东烟台264005

出　　处：《中国集成电路》2016年第11期41-45,共5页China lntegrated Circuit

基　　金：烟台大学研究生科技创新基金资助重点项目(编号YDZD1605)

摘　　要：先验信噪比是语音增强算法的重要研究参数之一。DD(Decision Directed)算法是常见的先验信噪比估计算法。DD算法具计算复杂度低和抑制音乐噪声的优点,然而其估计的增强语音的语音失真严重。本文首先介绍了语音增强算法的基本理论并分析了DD算法和三种新型改进算法的性能;其次,对MDD(Modified Decision Directed)算法、融合动量项的DD算法和增益因子线性预测算法进行实验仿真;最后,分析了先验信噪比估计算法今后的改进方向。The a prior signal-to-noise ratio （ a priori SNR ） is one of the crucial research parameters of speech en- hancement algorithm. The decision directed （ DD ） method is the widely used a priori SNR estimation algorithm. The advantages of the DD method are low computing complexity and it can effectively suppress the musical noise, however, the enhanced speech based on this method suffers from serious speech distortion. In this paper, we firstly introduced the basic theory of the speech enhancement algorithm and analyzed the performance of the DD method and three modified methods. Then, we made simulation of the modified DD （ MDD ） method, the momentum factor DD algorithm and the gain factor linear prediction method. Finally, we made analysis of the improvement direction of the a priori SNR estimation method.

关 键 词：语音增强 先验信噪比 语音失真 动量项 线性预测 

分 类 号：TN912.35[电子电信—通信与信息系统]