基于TVAR模型的语音增强技术  

Speech enhancement technique based on time-varying autoregressive model

作　　者：石鸿凌[1] 姜琳峰[1] 孙洪[1] 

机构地区：[1]武汉大学电子信息学院,湖北武汉430072

出　　处：《武汉大学学报（工学版）》2004年第2期49-52,共4页Engineering Journal of Wuhan University

基　　金：国家自然科学基金资助项目(60072041)

摘　　要：在分析语音信号的时变自回归TVAR(Time VaryingAutoregressive)模型及其模型参数的随机演化模型的基础上,基于粒子滤波器(ParticleFilter)对TVAR模型参数的序列估计,提出了一种语音增强算法.算法通过引入反射系数,快速简捷实现了模型稳定性的判断,保证了跟踪的模型的稳定性.实验结果表明,算法可以很好地跟踪非平稳信号,采用该方法处理过的语音,信噪比SNR(Signal to NoiseRatio)明显提高,听觉质量得到了很大的改善.By exploring the time-varying autoregressive models and its stochastically evolving parameters, and introducing reflection coefficient to ensure model strict stability, we present a time-varying autoregressive model based speech enhancement algorithm. The algorithm relies on a particle filter estimation of the TVAR model parameters. The experimental results show that after filtered by our algorithm the signal-to-noise ratio and the quality of speech have been improved obviously.

关 键 词：语音增强 时变自回归模型 粒子滤波器 

分 类 号：TN912.35[电子电信—通信与信息系统]