摘要
针对语音自适应增强的递归最小二乘(RLS)算法必须已知参考信号的约束条件,将小波技术引入RLS算法中,提出了一种语音自适应增强算法-小波递归最小二乘算法(WRLS)。该算法无需参考输入和输入信号的延时量,而是用小波分解、合成技术初估期望信号,以此获得先验误差;再用RLS算法求解滤波器权系数修正量;同时采用"块"和"符号"技术减少权系数修正的运算量,提高算法的收敛速度。仿真实验表明该算法的增强效果明显优于谱减法和小波增强法。
According to recursive least square(RLS) algorithm of speech adaptive enhancement, constraints of the reference signal must be known, the wavelet technique is used in the RLS algorithm, a kind of speech adaptive enhancement algorithm based on wavelet recursive least square(WRLS) is put forward. The algorithm doesn't need reference input and the time delay of the input signal, but desired signal is estimated first by wavelet decomposition, synthesis technology, thus obtaining a priori error; the weight coefficients correction of the filter is obtained adopting RLS algorithm; meanwhile, adopting the technique of "block" and "symbol" reduces the amount of operation of weight coefficients correction, and improves the rate of convergence of the algorithm. Simulation results show that the enhancement effect of the algorithm is obviously superior to the spectral subtraction and wavelet enhancement algorithm.
出处
《电子设计工程》
2016年第1期69-72,共4页
Electronic Design Engineering
基金
国家社科基金资助项目(15XYY026)
关键词
语音
自适应增强
小波
RLS算法
滤波器权系数
speech
adaptive enhancement
wavelet
RLS algorithm
weight coefficients of filter