基于改进小波阈值函数和PSO的语音增强算法

Speech Enhancement Algorithm Based on Optimized Wavelet Thresholding Function and PSO

阈值的估计与阈值函数的选取是小波语音增强技术中的重要部分。针对小波阈值不能随着系统实时改变,仅靠人工因素和统计理论来决策,导致阈值估计非最优,很难满足语音实时处理要求的问题。提出了一种改进的阈值函数和粒子群自适应阈值寻优算法。改进的阈值函数不仅有传统阈值函数同等效果,而且高阶可导,污染的语音小波系数经过改进的PSO算法将全体个体最优值的平均值代替速度更新公式中个体最优值,加快全局收敛速度和提高全局优化精度。利用改进的函数和最优阈值后,得到增强的语音。仿真结果表明:增强后语音信号的信噪比、语音可懂度和均方根误差均优于其它传统方法。

Wavelet threshold functions and the estimation of threshold is a key part of the speech enhancement algorithm with wavelet.For wavelet threshold,without the optimal estimated value,cannot be provided real-time updates and only based on experience or statistics to decision-making,which is difficult to meet the requirements of real-time speech processing.An improved threshold function and the particle swarm adaptive threshold optimization algorithm are proposed. The threshold function can not only realize the effect of traditional threshold function,but also has the second order or higher order more continuous derivative.The optimal value of individual particle in the renewal speed formula are replaced by the average of all particle individuals＇ optimal solution in the optimal algorithm of PSO.That can improve the accuracy of the global optimization and accelerate convergence velocity of the global.Use the improved function and the optimal threshold to obtain the enhanced speech.The simulation results show that： the signal-to-noise ratio（ SNR）,root mean square error（ RMSE） and intelligibility of enhanced speech using the improved threshold function are superior to others.