三种经典单通道语音增强算法的比较

Comparison of Three Classical Single Channel Speech Enhancement Algorithms

本文介绍了三种经典的单通道语音增强算法:子空间、最小均方误差估计和谱减法,分析和比较了这三种方法的优缺点。仿真实验表明,子空间去噪能力最强,语音失真度最大,运行时间最长;最小均方误差估计去噪能力不如子空间,语音失真度最小,运行时间介于其它两个算法之间;谱减法去噪能力和语音失真度介于其它两个算法之间,运行时间最短。

This paper introduces three classical single channel speech enhancement algorithms that is subspace,mini-mum mean - square error estimator（MMSE）and spectral subtraction,and then discusses advantage and disadvantage of the three methods. Simulation result shows that the signal enhanced by subspace has the least noises and the maximum speech distortion,and the runtime of subspace is the longest. Although the signal enhanced by MMSE has more noises than signal enhanced by subspace,it has the minimum speech distortion and the runtime of MMSE ranks in the middle. The noise reduction and speech distortion of spectral subtraction rank in the middle and the runtime of spectral subtrac-tion is the shortest.