基于动量项先验信噪比估计的语音增强算法  

Speech enhancement algorithms based on momentum term a-priori SNR estimation

作　　者：沈锁金 刘伟[1] 魏静[1] SHEN Suojin LIU Wei WEI Jing(School of Science and Technology for Opto-Electronics Information of Yantai University, Yantai Shandong 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：《电声技术》2017年第1期55-57,62,共4页Audio Engineering

基　　金：烟台大学研究生科技创新基金重点项目(YDZD1605)

摘　　要：先验信噪比(Signal Noise Ratio,SNR)估计的精度影响着语音增强算法的性能。直接判决(Decision Directed,DD)算法是最常用的先验信噪比估计算法,但是,该算法估计的增强语音语音失真严重。首先介绍了语音增强算法的基本理论并分析了DD算法、TSNR算法和融入动量项算法,然后运用MATLAB软件进行实验仿真,对比分析了融入动量项算法的性能优势。The estimation precision of a priori signal-to-noise （SNR） affects the performance of speech enhancement algo- rithm. DD（Decision Directed） algorithm is the most commonly used method to estimate a prior SNR, however, the en- hanced speech has serious speech distortion. This paper first introduces the basic theory of speech enhancement algorithm and analyzes the DD algorithm, TSNR algorithm and integrated into the momentum item algorithm. Then use MATLAB soft- ware make simulation, comparison and analysis the performance advantage of the momentum item DD algorithm.

关 键 词：先验信噪比 语音增强 语音失真 动量项 

分 类 号：TN912.3[电子电信—通信与信息系统]