Speech Random Impulse Noise Elimination Method Based on Robust PCA Inexact ALM Algorithm

Abstract

In speech enhancement algorithms, many algorithms take gaussian white noise as noise model and objective function based on Euclidean \({L^2}\) norm. Although these algorithms are effective for gaussian white noise, they are poor for random impulse noise. First, this paper analyzes the shortcomings of the typical speech enhancement algorithm based on PCA (Principal Component Analysis) in white noise model, for the PCA algortithm has poor effect in speech contaminated by impact random noise. However, a new method called RPCA (Robust Principal Component Analysis) and non \({L^2}\) norm can be used as the objective function to deal with random impact noise. For the reason of increase of data size, the calculation amount of RPCA algorithm also increases exponentially. So rapid inexact ALM (Augmented Lagrange Multiplier) algorithm is introduced to solve RPCA problem. The experiments show that the proposed algorithm can effectively eliminate the random impact noise of speech.KeywordsRandom impact noiseRobust principal component analysisInexact ALM algorithm