摘要
语音信号是一种非稳态的随机信号,而传统的时频分析法缺乏对这类信号进行最稀疏表示的能力,为此提出了一种完备的局部均值分解(Ensemble Local Mean Decomposition,ELMD)联合粒子群优化小波阈值语音消噪分析法:在对原始信号LMD(局部均值分解,Local Mean Decomposition)分解基础上加入高斯白噪声辅助分析的自适应分析法,以减轻分解后的产生模态混叠现象;对于分解后的分量中残留的噪声使用粒子群优化算法获得最优小波阈值滤除。对实际采集语音信号进行Matlab仿真的处理分析结果显示,该算法在抑制语音中的背景噪声有着良好的效果,且有效降低了对语音有效信息的损伤。
The speech signal is a non-stationary and random weak signal,and the traditional time frequency analysis method is insufficient for the most sparsely expression of such signals,therefore a method for speech signals based on ensemble local mean decomposition(ELMD)and joint particle swarm optimization(JPSO)method is proposed.On the basis of the LMD decomposition of the original signal,the adaptive analysis method of Gauss white noise aided analysis is added to reduce the mode aliasing after decomposition.The particle swarm optimization(PSO)algorithm is used to obtain the optimal wavelet threshold for filtering the residual noise in the decomposed component.The Matlab simulation experiment is carried out on the actual speech signal for indicating the analysis results,the proposed algorithm has a good effect on the background noise suppression in speech,and it effectively reduces the damage to the effective speech information.
出处
《长江大学学报(自然科学版)》
CAS
2018年第1期33-38,共6页
Journal of Yangtze University(Natural Science Edition)
基金
国家自然科学基金项目(61272147)
湖北省教育厅项目(B2015446)
长江大学青年基金项目(2016cqn10)
关键词
语音信号
模态混叠
高斯白噪声
完备的局部均值分解
粒子群优化算法
小波阈值
speech signals
mode aliasing
Gauss white noise
ensemble local mean decomposition(ELMD)
particle swarm optimization
wavelet threshold