摘要
噪声估计的准确度决定着语音增强效果的好坏。为了提高噪声估计的准确度,并及时跟踪非平稳噪声,在最小值控制递归平均算法的基础之上,提出了连续频谱最小值跟踪的改进最小值控制递归平均算法。通过对含噪语音功率谱的每一个频点进行连续平滑,然后再在子窗内采用最小值搜索的方法来实现噪声谱的估计。仿真实验结果表明,改进后的算法相对于原算法,在输出信噪比上提高大约0.8~2.1dB,能够更加准确的估计噪声谱,提升语音质量。
The accuracy of noise estimation determines the speech enhancement effect.In order to improve the accuracy of noise estimation,track non-stationary noise timely,a speech enhancement approach based on improved minima controlled recursive averaging using continuous spectral minimum tracking is proposed.The proposed algorithm improves the noise estimation by continuously smoothing each frequency of the noisy speech power spectrum and then searching the minimum value in the sub-window.The experimental results show that proposed algorithm can increases the output SNR by approximately 0.8~2.1 dB and more accurately estimate the noise spectrum and improve the speech quality.
作者
邵虹
王杰
Shao Hong;Wang Jie(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《电子测量技术》
2018年第14期16-20,共5页
Electronic Measurement Technology
基金
辽宁省高等学校优秀科技人才支持计划(LJQ2013013)项目资助