基于归一化自相关语音筛选的鲁棒性说话人确认研究-《郑州大学学报(工学版)》

文章信息/Info

Title:: Confirmation Researching on Robustness Speaker based-on NormalizedAutocorrelation Voice Selection/div>

Author(s):: Wang Shan; GU Cheng; School of Physics and Electronics，Henan University，Kaifeng 475000 China

Keywords:: normalized autocorrelation; Voice filtering; robustness; The speaker confirms

摘要:: 语音在受到加性噪声污染时,不同部分受噪声污染的程度存在差异,保留受污染严重的语音将会对说话人确认系统产生负面影响,笔者提出了一种基于归一化自相关的鲁棒性语音筛选方法.通过归一化自相关舍弃受污染严重的语音.实验表明,通过归一化自相关舍掉受污染严重的语音,能提高噪声环境下的说话人确认性能.

Abstract:: Aiming at the situation that the voice was vitiated by positive noise and the difference of pollutionextent between different parts of the voice，preserving the polluted parts of voice will bring in negative affectionto the confirmation system of speaker．This paper gives a method that based on normalized autocorrelation ro。bustness selection．Normalized autocorrelation not only shows the correlation of short—time voice frame，butalso shows the polluted extent of voice．As a result，we can wipe off the deeply polluted voice through the nor‘malized autocorrelation．Experiments showing，through the normalized autocorrelation wiping off the pollutedvoice，can improve confirmation ability of speaker in high noise environment．