基于Kalman的语音特征参数提取方法研究  

Research of Speech Feature Coefficients Based on Kalman Filtering

作　　者：黄禹胜 张丕状[1] 金东泽[1] 

机构地区：[1]中北大学电子测试技术国家重点实验室,山西太原030051

出　　处：《电声技术》2015年第5期62-65,共4页Audio Engineering

摘　　要：传统的助听器语音去噪方法如谱减法、维纳滤波法等都因计算量大、结构复杂而限制了其实际应用,对一种基于Kalman自适应算法的语音去噪方法进行了分析,在此基础上,对提出的方法进行了实验仿真和误差分析。结果表明:通过Kalman自适应去噪,语音特征参数LPC倒谱系数误差更小,合成语音后的辨识率更高,且除去噪声的同时没有引入新的噪声。Traditional calculation methods such as spectral subtraction or Wiener filtering, are limited to practical applica- tion for the large amount of calculation and complex structure. So a computing method is proposed in this article, which is a speech signal adaptive enhancement algorithm based on Kalman filtering after analyzing the fundamental principles of adaptive algorithm based on Kalman and extraction method of speech signal characteristic coefficients. The simulation results in- dicate that filtering the noise through the Kalman filter, the tolerance of LPC frequency cepstrum coefficient is smaller and the identification is more accurate after the frames consist of the speech signal. Meanwhile, there is no other new noise that is added into the speech signal.

关 键 词：自适应滤波 语音特征参数 KALMAN LPC 

分 类 号：TN911[电子电信—通信与信息系统]